This document provides details about a technical seminar report on Blue Gene/L supercomputer. It includes an abstract describing Blue Gene/L's architecture as a 64,000 node scientific supercomputer developed with DOE funding. It also describes the torus interconnection network that connects the nodes and how a parallel performance simulator was used to help design and test the network. The report is submitted by a student to fulfill the requirements for a bachelor's degree in computer science and includes certificates, declarations, acknowledgements and references.
International Journal of Computational Engineering Research(IJCER) ijceronline
International Journal of Computational Engineering Research (IJCER) is dedicated to protecting personal information and will make every reasonable effort to handle collected information appropriately. All information collected, as well as related requests, will be handled as carefully and efficiently as possible in accordance with IJCER standards for integrity and objectivity.
Abstract: Design verification is an essential step in the development of any product. It ensures that the product as designed is the same as the product as intended. Software simulation is the common approach for validating hardware design unfortunately, it will take hours together to execute. Difficulties in validation arise due to the complexity of the design and also due to the lack of on chip observability. One common solution to this problem is to instrument the prototype using trace-buffers to record a subset of internal signals into on-chip memory for subsequent analysis. In the proposed system, an example circuit is implemented to perform the tracing operation and various trace buffers are designed to record the different stages of internal signal states. The resulting signal states are to be stored, like a error outputs. Low power methodologies are also implemented to achieve low power consumption. Thus the errors are separately stored in the memory for analyzing the signals. This might be used for changes in the logic wherever needed. Thus this tracing is performed to monitor signal states of an FPGA.
International Journal of Computational Engineering Research(IJCER) ijceronline
International Journal of Computational Engineering Research (IJCER) is dedicated to protecting personal information and will make every reasonable effort to handle collected information appropriately. All information collected, as well as related requests, will be handled as carefully and efficiently as possible in accordance with IJCER standards for integrity and objectivity.
Abstract: Design verification is an essential step in the development of any product. It ensures that the product as designed is the same as the product as intended. Software simulation is the common approach for validating hardware design unfortunately, it will take hours together to execute. Difficulties in validation arise due to the complexity of the design and also due to the lack of on chip observability. One common solution to this problem is to instrument the prototype using trace-buffers to record a subset of internal signals into on-chip memory for subsequent analysis. In the proposed system, an example circuit is implemented to perform the tracing operation and various trace buffers are designed to record the different stages of internal signal states. The resulting signal states are to be stored, like a error outputs. Low power methodologies are also implemented to achieve low power consumption. Thus the errors are separately stored in the memory for analyzing the signals. This might be used for changes in the logic wherever needed. Thus this tracing is performed to monitor signal states of an FPGA.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
Deep Convolutional Network evaluation on the Intel Xeon PhiGaurav Raina
With a sharp decline in camera cost and size along with superior computing power available at increasingly low prices, computer vision applications are becoming ever present in our daily lives. Research shows that Convolutional Neural Networks (ConvNet) can outperform all other methods for
computer vision tasks (such as object detection) in terms of accuracy and versatility.
One of the problems with these Neural Networks, which mimic the brain, is that they can be very demanding on the processor, requiring millions of computational nodes to function. Hence, it is challenging for Neural Network
algorithms to achieve real-time performance on general purpose embedded platforms.
Parallelization and vectorization are very eective ways to ease this problem and make it possible to implement such ConvNets on energy efficient embedded platforms. This thesis presents the evaluation of a novel ConvNet for road speed sign detection, on a breakthrough 57-core Intel Xeon Phi
processor with 512-bit vector support. This mapping demonstrates that the parallelism inherent in the ConvNet algorithm can be effectively exploited by the 512-bit vector ISA and by utilizing the many core paradigm.
Detailed evaluation shows that the best mappings require data-reuse strategies that exploit reuse at the cache and register level. These implementations are boosted by the use of low-level vector intrinsics (which are
C style functions that map directly onto many Intel assembly instructions).
Ultimately we demonstrate an approach which can be used to accelerate Neural Networks on highly-parallel many core processors, with execution speedups of more than 12x on single core performance alone.
OpenACC and Open Hackathons Monthly Highlights: July 2022.pptxOpenACC
Stay up-to-date with the OpenACC and Open Hackathons Monthly Highlights. July’s edition covers the 2022 OpenACC and Hackathons Summit, NVIDIA’s Applied Research Accelerator Program, upcoming Open Hackathons and Bootcamps, recent research, new resources, and more!
A 64-by-8 Scrolling Led Matrix Display SystemSamson Kayode
For this project, a prototype of a 64 x 8 LED matrix, capable of displaying a maximum of eight (8) characters at a time was constructed. Then, scrolling the characters to show the remaining part of the message and repeating the same loop over and over again was achieved. This process can be interrupted by failure in the power supply or when the reset button is being pushed and released. The system can easily be expanded to handle more characters. The control of this LED matrix is based on a PIC18F4520 microcontroller. This microcontroller was programmed using C language. The number of the microcontroller’s pins used in controlling the LED matrix was strictly minimized to three by adopting the serial to parallel mode of signal transmission. The design of the project was done and simulated with Proteus software. The LED matrix was constructed on a vero board. The drive circuitry which consists of the microcontroller, two ULN2803s, nine 74HC595s and other peripherals was constructed on a printed circuit board. This was done to ensure the project was compact as possible. Cost efficiency was also given a priority in the implementation of this project.
Video and slides synchronized, mp3 and slide download available at URL https://bit.ly/2JrUYLl.
Alison Lowndes talks about the HW & SW that comprise NVIDIA's GPU computing platform for AI, across PC to data center, cloud to edge, training to inference. She details current state-of-the-art research & recent internal work combining robotics with virtual reality & reinforcement learning in an end-to-end simulator for training and testing robots. Filmed at qconlondon.com.
Alison Lowndes is responsible for NVIDIA's Artificial Intelligence Developer Relations in the EMEA region. She consults on a wide range of AI applications, including planetary defence with NASA & the SETI Institute and continues to manage the community of AI & Machine Learning researchers around the world.
International Journal of Engineering Research and Applications (IJERA) is an open access online peer reviewed international journal that publishes research and review articles in the fields of Computer Science, Neural Networks, Electrical Engineering, Software Engineering, Information Technology, Mechanical Engineering, Chemical Engineering, Plastic Engineering, Food Technology, Textile Engineering, Nano Technology & science, Power Electronics, Electronics & Communication Engineering, Computational mathematics, Image processing, Civil Engineering, Structural Engineering, Environmental Engineering, VLSI Testing & Low Power VLSI Design etc.
Deep Convolutional Network evaluation on the Intel Xeon PhiGaurav Raina
With a sharp decline in camera cost and size along with superior computing power available at increasingly low prices, computer vision applications are becoming ever present in our daily lives. Research shows that Convolutional Neural Networks (ConvNet) can outperform all other methods for
computer vision tasks (such as object detection) in terms of accuracy and versatility.
One of the problems with these Neural Networks, which mimic the brain, is that they can be very demanding on the processor, requiring millions of computational nodes to function. Hence, it is challenging for Neural Network
algorithms to achieve real-time performance on general purpose embedded platforms.
Parallelization and vectorization are very eective ways to ease this problem and make it possible to implement such ConvNets on energy efficient embedded platforms. This thesis presents the evaluation of a novel ConvNet for road speed sign detection, on a breakthrough 57-core Intel Xeon Phi
processor with 512-bit vector support. This mapping demonstrates that the parallelism inherent in the ConvNet algorithm can be effectively exploited by the 512-bit vector ISA and by utilizing the many core paradigm.
Detailed evaluation shows that the best mappings require data-reuse strategies that exploit reuse at the cache and register level. These implementations are boosted by the use of low-level vector intrinsics (which are
C style functions that map directly onto many Intel assembly instructions).
Ultimately we demonstrate an approach which can be used to accelerate Neural Networks on highly-parallel many core processors, with execution speedups of more than 12x on single core performance alone.
OpenACC and Open Hackathons Monthly Highlights: July 2022.pptxOpenACC
Stay up-to-date with the OpenACC and Open Hackathons Monthly Highlights. July’s edition covers the 2022 OpenACC and Hackathons Summit, NVIDIA’s Applied Research Accelerator Program, upcoming Open Hackathons and Bootcamps, recent research, new resources, and more!
A 64-by-8 Scrolling Led Matrix Display SystemSamson Kayode
For this project, a prototype of a 64 x 8 LED matrix, capable of displaying a maximum of eight (8) characters at a time was constructed. Then, scrolling the characters to show the remaining part of the message and repeating the same loop over and over again was achieved. This process can be interrupted by failure in the power supply or when the reset button is being pushed and released. The system can easily be expanded to handle more characters. The control of this LED matrix is based on a PIC18F4520 microcontroller. This microcontroller was programmed using C language. The number of the microcontroller’s pins used in controlling the LED matrix was strictly minimized to three by adopting the serial to parallel mode of signal transmission. The design of the project was done and simulated with Proteus software. The LED matrix was constructed on a vero board. The drive circuitry which consists of the microcontroller, two ULN2803s, nine 74HC595s and other peripherals was constructed on a printed circuit board. This was done to ensure the project was compact as possible. Cost efficiency was also given a priority in the implementation of this project.
Video and slides synchronized, mp3 and slide download available at URL https://bit.ly/2JrUYLl.
Alison Lowndes talks about the HW & SW that comprise NVIDIA's GPU computing platform for AI, across PC to data center, cloud to edge, training to inference. She details current state-of-the-art research & recent internal work combining robotics with virtual reality & reinforcement learning in an end-to-end simulator for training and testing robots. Filmed at qconlondon.com.
Alison Lowndes is responsible for NVIDIA's Artificial Intelligence Developer Relations in the EMEA region. She consults on a wide range of AI applications, including planetary defence with NASA & the SETI Institute and continues to manage the community of AI & Machine Learning researchers around the world.
Deep Convolutional Neural Network acceleration on the Intel Xeon PhiGaurav Raina
With a sharp decline in camera cost and size along with superior computing power available at increasingly low prices, computer vision applications are becoming ever present in our daily lives. Research shows that Convolutional Neural Networks can outperform all other methods for computer vision tasks (such as object detection) in terms of accuracy and versatility.
One of the problems with these Neural Networks, which mimic the brain, is that they can be very demanding on the processor, requiring millions of computational nodes to function. Hence, it is challenging for Neural Network algorithms to achieve real-time performance on general purpose embedded platforms. Parallelization is one of the most effective ways to ease this problem and make it possible to implement such Neural Nets on energy efficient embedded platforms.
We present an evaluation of a novel Convolutional Neural Network for Road Speed Sign detection on the new 57 core Xeon Phi processor with 512-bit vector support. This aims to demonstrate that the parallelism inherent in the algorithm can be effectively exploited by the 512-bit vector ISA and by utilizing the many core paradigm.
Ultimately we demonstrate an approach which can be used to accelerate Neural Network based applications on massively parallel many-core processors, with speedups of more than 12x on single core performance alone.
ASSESSING THE PERFORMANCE AND ENERGY USAGE OF MULTI-CPUS, MULTI-CORE AND MANY...ijdpsjournal
This paper studies the performance and energy consumption of several multi-core, multi-CPUs and manycore
hardware platforms and software stacks for parallel programming. It uses the Multimedia Multiscale
Parser (MMP), a computationally demanding image encoder application, which was ported to several
hardware and software parallel environments as a benchmark. Hardware-wise, the study assesses
NVIDIA's Jetson TK1 development board, the Raspberry Pi 2, and a dual Intel Xeon E5-2620/v2 server, as
well as NVIDIA's discrete GPUs GTX 680, Titan Black Edition and GTX 750 Ti. The assessed parallel
programming paradigms are OpenMP, Pthreads and CUDA, and a single-thread sequential version, all
running in a Linux environment. While the CUDA-based implementation delivered the fastest execution, the
Jetson TK1 proved to be the most energy efficient platform, regardless of the used parallel software stack.
Although it has the lowest power demand, the Raspberry Pi 2 energy efficiency is hindered by its lengthy
execution times, effectively consuming more energy than the Jetson TK1. Surprisingly, OpenMP delivered
twice the performance of the Pthreads-based implementation, proving the maturity of the tools and
libraries supporting OpenMP.
ASSESSING THE PERFORMANCE AND ENERGY USAGE OF MULTI-CPUS, MULTI-CORE AND MANY...ijdpsjournal
This paper studies the performance and energy consumption of several multi-core, multi-CPUs and manycore
hardware platforms and software stacks for parallel programming. It uses the Multimedia Multiscale
Parser (MMP), a computationally demanding image encoder application, which was ported to several
hardware and software parallel environments as a benchmark. Hardware-wise, the study assesses
NVIDIA's Jetson TK1 development board, the Raspberry Pi 2, and a dual Intel Xeon E5-2620/v2 server, as
well as NVIDIA's discrete GPUs GTX 680, Titan Black Edition and GTX 750 Ti. The assessed parallel
programming paradigms are OpenMP, Pthreads and CUDA, and a single-thread sequential version, all
running in a Linux environment. While the CUDA-based implementation delivered the fastest execution, the
Jetson TK1 proved to be the most energy efficient platform, regardless of the used parallel software stack.
Although it has the lowest power demand, the Raspberry Pi 2 energy efficiency is hindered by its lengthy
execution times, effectively consuming more energy than the Jetson TK1. Surprisingly, OpenMP delivered
twice the performance of the Pthreads-based implementation, proving the maturity of the tools and
libraries supporting OpenMP.
22 Re-Engineering Traditional Learning Model with Outcome-Based Learning Curv...ijtsrd
This paper presents the design and implementation of an indigenous outcome based 8051 microcontroller training kit. Most of the existing microcontroller training kits available do not have local content, and also have problems with maintenance and repair when they develop faults. Users usually dump these kits when they develop faults due to a lack of trained technical maintenance personnel. This project used components that are locally available to implement an outcome based 8051 microcontroller training kit that is modular and easy to use. In the methodology, the bottom top approach was used to design the individual modules of the kit before fabricating them on a printed circuit board PCB . The user manual was then developed and experiments in it were tested using the developed training kit. The result of the work is an outcome based 8051 microcontroller training kit that meets the National Universities Commission NUC and Council for the Regulation of Engineering in Nigeria COREN standards. It has various modules for the user to practice several 8051 microcontroller programming experiments. The work can be deployed in university laboratories for microcontrollers and embedded systems training. Specifically, there are courses in Mechatronics engineering, FUTO, that can be taught using this kit. Ezenwa Opara | Mbonu, Ekene S. | Udemezue, Obinna E. | Okereke, Chukwunenye G. | Uzoeto, Anurika C. "Re-Engineering Traditional Learning Model with Outcome-Based Learning Curve using Embedded Training Laboratory" Published in International Journal of Trend in Scientific Research and Development (ijtsrd), ISSN: 2456-6470, Volume-7 | Issue-3 , June 2023, URL: https://www.ijtsrd.com.com/papers/ijtsrd55127.pdf Paper URL: https://www.ijtsrd.com.com/engineering/electrical-engineering/55127/reengineering-traditional-learning-model-with-outcomebased-learning-curve-using-embedded-training-laboratory/ezenwa-opara
State of ICS and IoT Cyber Threat Landscape Report 2024 previewPrayukth K V
The IoT and OT threat landscape report has been prepared by the Threat Research Team at Sectrio using data from Sectrio, cyber threat intelligence farming facilities spread across over 85 cities around the world. In addition, Sectrio also runs AI-based advanced threat and payload engagement facilities that serve as sinks to attract and engage sophisticated threat actors, and newer malware including new variants and latent threats that are at an earlier stage of development.
The latest edition of the OT/ICS and IoT security Threat Landscape Report 2024 also covers:
State of global ICS asset and network exposure
Sectoral targets and attacks as well as the cost of ransom
Global APT activity, AI usage, actor and tactic profiles, and implications
Rise in volumes of AI-powered cyberattacks
Major cyber events in 2024
Malware and malicious payload trends
Cyberattack types and targets
Vulnerability exploit attempts on CVEs
Attacks on counties – USA
Expansion of bot farms – how, where, and why
In-depth analysis of the cyber threat landscape across North America, South America, Europe, APAC, and the Middle East
Why are attacks on smart factories rising?
Cyber risk predictions
Axis of attacks – Europe
Systemic attacks in the Middle East
Download the full report from here:
https://sectrio.com/resources/ot-threat-landscape-reports/sectrio-releases-ot-ics-and-iot-security-threat-landscape-report-2024/
Elevating Tactical DDD Patterns Through Object CalisthenicsDorra BARTAGUIZ
After immersing yourself in the blue book and its red counterpart, attending DDD-focused conferences, and applying tactical patterns, you're left with a crucial question: How do I ensure my design is effective? Tactical patterns within Domain-Driven Design (DDD) serve as guiding principles for creating clear and manageable domain models. However, achieving success with these patterns requires additional guidance. Interestingly, we've observed that a set of constraints initially designed for training purposes remarkably aligns with effective pattern implementation, offering a more ‘mechanical’ approach. Let's explore together how Object Calisthenics can elevate the design of your tactical DDD patterns, offering concrete help for those venturing into DDD for the first time!
UiPath Test Automation using UiPath Test Suite series, part 4DianaGray10
Welcome to UiPath Test Automation using UiPath Test Suite series part 4. In this session, we will cover Test Manager overview along with SAP heatmap.
The UiPath Test Manager overview with SAP heatmap webinar offers a concise yet comprehensive exploration of the role of a Test Manager within SAP environments, coupled with the utilization of heatmaps for effective testing strategies.
Participants will gain insights into the responsibilities, challenges, and best practices associated with test management in SAP projects. Additionally, the webinar delves into the significance of heatmaps as a visual aid for identifying testing priorities, areas of risk, and resource allocation within SAP landscapes. Through this session, attendees can expect to enhance their understanding of test management principles while learning practical approaches to optimize testing processes in SAP environments using heatmap visualization techniques
What will you get from this session?
1. Insights into SAP testing best practices
2. Heatmap utilization for testing
3. Optimization of testing processes
4. Demo
Topics covered:
Execution from the test manager
Orchestrator execution result
Defect reporting
SAP heatmap example with demo
Speaker:
Deepak Rai, Automation Practice Lead, Boundaryless Group and UiPath MVP
Builder.ai Founder Sachin Dev Duggal's Strategic Approach to Create an Innova...Ramesh Iyer
In today's fast-changing business world, Companies that adapt and embrace new ideas often need help to keep up with the competition. However, fostering a culture of innovation takes much work. It takes vision, leadership and willingness to take risks in the right proportion. Sachin Dev Duggal, co-founder of Builder.ai, has perfected the art of this balance, creating a company culture where creativity and growth are nurtured at each stage.
GDG Cloud Southlake #33: Boule & Rebala: Effective AppSec in SDLC using Deplo...James Anderson
Effective Application Security in Software Delivery lifecycle using Deployment Firewall and DBOM
The modern software delivery process (or the CI/CD process) includes many tools, distributed teams, open-source code, and cloud platforms. Constant focus on speed to release software to market, along with the traditional slow and manual security checks has caused gaps in continuous security as an important piece in the software supply chain. Today organizations feel more susceptible to external and internal cyber threats due to the vast attack surface in their applications supply chain and the lack of end-to-end governance and risk management.
The software team must secure its software delivery process to avoid vulnerability and security breaches. This needs to be achieved with existing tool chains and without extensive rework of the delivery processes. This talk will present strategies and techniques for providing visibility into the true risk of the existing vulnerabilities, preventing the introduction of security issues in the software, resolving vulnerabilities in production environments quickly, and capturing the deployment bill of materials (DBOM).
Speakers:
Bob Boule
Robert Boule is a technology enthusiast with PASSION for technology and making things work along with a knack for helping others understand how things work. He comes with around 20 years of solution engineering experience in application security, software continuous delivery, and SaaS platforms. He is known for his dynamic presentations in CI/CD and application security integrated in software delivery lifecycle.
Gopinath Rebala
Gopinath Rebala is the CTO of OpsMx, where he has overall responsibility for the machine learning and data processing architectures for Secure Software Delivery. Gopi also has a strong connection with our customers, leading design and architecture for strategic implementations. Gopi is a frequent speaker and well-known leader in continuous delivery and integrating security into software delivery.
LF Energy Webinar: Electrical Grid Modelling and Simulation Through PowSyBl -...DanBrown980551
Do you want to learn how to model and simulate an electrical network from scratch in under an hour?
Then welcome to this PowSyBl workshop, hosted by Rte, the French Transmission System Operator (TSO)!
During the webinar, you will discover the PowSyBl ecosystem as well as handle and study an electrical network through an interactive Python notebook.
PowSyBl is an open source project hosted by LF Energy, which offers a comprehensive set of features for electrical grid modelling and simulation. Among other advanced features, PowSyBl provides:
- A fully editable and extendable library for grid component modelling;
- Visualization tools to display your network;
- Grid simulation tools, such as power flows, security analyses (with or without remedial actions) and sensitivity analyses;
The framework is mostly written in Java, with a Python binding so that Python developers can access PowSyBl functionalities as well.
What you will learn during the webinar:
- For beginners: discover PowSyBl's functionalities through a quick general presentation and the notebook, without needing any expert coding skills;
- For advanced developers: master the skills to efficiently apply PowSyBl functionalities to your real-world scenarios.
UiPath Test Automation using UiPath Test Suite series, part 3DianaGray10
Welcome to UiPath Test Automation using UiPath Test Suite series part 3. In this session, we will cover desktop automation along with UI automation.
Topics covered:
UI automation Introduction,
UI automation Sample
Desktop automation flow
Pradeep Chinnala, Senior Consultant Automation Developer @WonderBotz and UiPath MVP
Deepak Rai, Automation Practice Lead, Boundaryless Group and UiPath MVP
SAP Sapphire 2024 - ASUG301 building better apps with SAP Fiori.pdfPeter Spielvogel
Building better applications for business users with SAP Fiori.
• What is SAP Fiori and why it matters to you
• How a better user experience drives measurable business benefits
• How to get started with SAP Fiori today
• How SAP Fiori elements accelerates application development
• How SAP Build Code includes SAP Fiori tools and other generative artificial intelligence capabilities
• How SAP Fiori paves the way for using AI in SAP apps
Smart TV Buyer Insights Survey 2024 by 91mobiles.pdf91mobiles
91mobiles recently conducted a Smart TV Buyer Insights Survey in which we asked over 3,000 respondents about the TV they own, aspects they look at on a new TV, and their TV buying preferences.
Welocme to ViralQR, your best QR code generator.ViralQR
Welcome to ViralQR, your best QR code generator available on the market!
At ViralQR, we design static and dynamic QR codes. Our mission is to make business operations easier and customer engagement more powerful through the use of QR technology. Be it a small-scale business or a huge enterprise, our easy-to-use platform provides multiple choices that can be tailored according to your company's branding and marketing strategies.
Our Vision
We are here to make the process of creating QR codes easy and smooth, thus enhancing customer interaction and making business more fluid. We very strongly believe in the ability of QR codes to change the world for businesses in their interaction with customers and are set on making that technology accessible and usable far and wide.
Our Achievements
Ever since its inception, we have successfully served many clients by offering QR codes in their marketing, service delivery, and collection of feedback across various industries. Our platform has been recognized for its ease of use and amazing features, which helped a business to make QR codes.
Our Services
At ViralQR, here is a comprehensive suite of services that caters to your very needs:
Static QR Codes: Create free static QR codes. These QR codes are able to store significant information such as URLs, vCards, plain text, emails and SMS, Wi-Fi credentials, and Bitcoin addresses.
Dynamic QR codes: These also have all the advanced features but are subscription-based. They can directly link to PDF files, images, micro-landing pages, social accounts, review forms, business pages, and applications. In addition, they can be branded with CTAs, frames, patterns, colors, and logos to enhance your branding.
Pricing and Packages
Additionally, there is a 14-day free offer to ViralQR, which is an exceptional opportunity for new users to take a feel of this platform. One can easily subscribe from there and experience the full dynamic of using QR codes. The subscription plans are not only meant for business; they are priced very flexibly so that literally every business could afford to benefit from our service.
Why choose us?
ViralQR will provide services for marketing, advertising, catering, retail, and the like. The QR codes can be posted on fliers, packaging, merchandise, and banners, as well as to substitute for cash and cards in a restaurant or coffee shop. With QR codes integrated into your business, improve customer engagement and streamline operations.
Comprehensive Analytics
Subscribers of ViralQR receive detailed analytics and tracking tools in light of having a view of the core values of QR code performance. Our analytics dashboard shows aggregate views and unique views, as well as detailed information about each impression, including time, device, browser, and estimated location by city and country.
So, thank you for choosing ViralQR; we have an offer of nothing but the best in terms of QR code services to meet business diversity!
Key Trends Shaping the Future of Infrastructure.pdfCheryl Hung
Keynote at DIGIT West Expo, Glasgow on 29 May 2024.
Cheryl Hung, ochery.com
Sr Director, Infrastructure Ecosystem, Arm.
The key trends across hardware, cloud and open-source; exploring how these areas are likely to mature and develop over the short and long-term, and then considering how organisations can position themselves to adapt and thrive.
Key Trends Shaping the Future of Infrastructure.pdf
blue gene report
1. A Technical Seminar Report on
Blue Gene
Submitted to
Jawaharlal Nehru Technological University, Hyderabad
in partial fulfillment of the requirements for the award of the degree
BACHELOR OF TECHNOLOGY
IN
COMPUTER SCIENCE & ENGINEERING
BY
Rabindra Raj Sah (16831A05J4)
Under the Esteemed Guidance of
Mrs. B.Ranjitha
Assistant Professor
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
GURU NANAK INSTITUTE OF TECHNOLOGY
(Affiliated to JNTUH-Hyderabad)
Ranga Reddy District -501506
2. i
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
GURU NANAK INSTITUTE OF TECHNOLOGY
(Affiliated to JNTUH-Hyderabad)
Ranga Reddy District -501506
CERTIFICATE
This is to certify that the Technical Seminar entitled “Blue Gene” is being presented with
report by Rabindra Raj Sah (16831A05J4) in partial fulfillment for the award of Degree of
Bachelor of Technology in Computer Science and Engineering, to Jawaharlal Nehru
Technological University, Hyderabad.
INTERNAL GUIDE SEMINAR COORDINATOR
(Mrs. B.Ranjitha) (Ms. Anuradha)
HEAD OF DEPARTMENT PRINCIPAL
(Dr. S.DeepaJothi) (Dr. Sreenatha Reddy)
3. ii
DECLARATION
I hereby declare that Technical Seminar report entitled “Blue Gene” is the work done
by Rabindra Raj Sah bearing the roll no. 16831A05J4 towards the fulfillment of the
requirement for the award of the Degree of Bachelor of Technology in Computer Science
and Engineering, to Jawaharlal Nehru Technological University, Hyderabad, is the
result of the work carried out under the guidance of Mrs. B.Ranjitha Guru Nanak Institute
of Technology, Hyderabad.
I further declare that this Technical Seminar report has not been previously submitted
before either in part or full for the award of any degree or any diploma by any organization or
any universities.
Mr. Rabindra Raj Sah (16831A05J4)
4. iii
ACKNOWLEDGEMENT
"Task successful" makes everyone happy. But the happiness will be gold without
glitter if i didn't state the persons who have supported me to make it a success.
I would like to express my sincere thanks and gratitude to our Principal,
Dr. S. SREENATHA REDDY and Head of the Department Dr. S. DeepaJothi, Department
of Computer Science and Engineering, Guru Nanak Institute of Technology for having
guided me in developing the requisite capabilities for taking up this Technical Seminar.
I thank Technical Seminar Coordinator Ms. Anuradha CSE, GNIT for providing
seamless support and right suggestions that are given in the development of the Technical
Seminar.
I specially thank our internal guide Mrs. B.Ranjitha for her constant guidance in
every stage of the Technical Seminar. I would also like to thank all our lecturers for helping
me in every possible way whenever the need arose.
On a more personal note I thank my beloved parents and friends for their moral
support during the course of my Technical Seminar. .
5. iv
ABSTRACT
Blue Gene/L (BG/L) is a 64K (65,536) node scientific and engineering
supercomputer that IBM is developing with partial funding from the United States
Department of Energy. This paper describes one of the primary BG/L interconnection
networks, a three dimensional torus. We describe a parallel performance simulator that was
used extensively to help architect and design the torus network and present sample simulator
performance studies that contributed to design decisions. In addition to such studies, the
simulator was also used during the logic verification phase of BG/L for performance
verification, and its use there uncovered a bug in the VHDL implementation of one of the
arbiters. Blue Gene/L (BG/L) is a scientific and engineering, message-passing,
supercomputer that IBM is developing with partial funding from the U.S. Department of
Energy Lawrence Livermore National Laboratory. A 64K node system is scheduled to be
delivered to Livermore, while a 20K node system will be installed at the IBM T.J. Watson
Research Center for use in life sciences computing, primarily protein folding. A more
complete overview of BG/L may be found in 1999, but we briefly describe the primary
features of the machine.
6. v
TABLE OF CONTENTS
Contents Page No
Certificate..................................................................................................................i
Declaration................................................................................................................ii
Acknowledgement.....................................................................................................iii
Abstract .....................................................................................................................iv
List of Figure.............................................................................................................vi
1. INTRODUCTION .............................................................................................1
1.1. GENERAL.....................................................................................................1
1.2. MAJOR FEATURES.....................................................................................2
1.3. TECHNICAL DETAILS...............................................................................3
2. ARCHITECTURE DETAILS...........................................................................4
2.1. PROGRAMMING MODELS.......................................................................5
2.2. CONTROL SYSTEM...................................................................................5
2.3. AUTONOMIC FEATURES..........................................................................6
2.4. SYSTEM OVERVIEW................................................................................6
3. DESIGN AND ANALYSIS OF THE BLUE GENE/L TORUS
INTERCONNECTION NETWORK................................................................9
3.1. TORUS NETWORK.....................................................................................9
3.2. SIMULATOR OVERVIEW..........................................................................12
4. SAMPLE PERFORMANCE STUDIES...........................................................15
5. APPLICATION...................................................................................................19
5.1. GENERAL.....................................................................................................19
5.2. PROTEIN ARCHITECTURE ......................................................................21
5.3. THE PROTEIN FOLDING PROBLEM.......................................................22
5.4. CURRENT VIEW OF FOLDING MECHANISMS ....................................22
5.5. CHALLENGES FOR COMPUTATIONAL MODELING.........................25
6. CONCLUSION.......................................................................... 26
REFERENCES............................................................................... 27
7. vi
LIST OF FIGURES
Contents Page No
Fig 2.4: System Overview 6
Fig 4.1: General Structural of Torus Router 16
Fig 4.2: Sample Response Time in Light Traffic 16
Fig 4.3: Average Link Utilization during All-to-All 17
Fig 5.1: DNA 19
Fig 5.2: Protein architecture 21
Fig 5.4: folding 23
8. 1
CHAPTER 1
INTRODUCTION
1.1 GENERAL
The first computer in the Blue Gene series, Blue Gene/L, developed through a
partnership with Lawrence Livermore National Laboratory, cost US$100 million and is
intended to scale to speeds in the hundreds of TFLOPS, with a theoretical peak performance
of 360 TFLOPS. This is almost ten times as fast as the Earth Simulator, the fastest
supercomputer in the world before Blue Gene. In June 2004, two Blue Gene/L prototypes
scored in the TOP500 Supercomputer List at the #4 and #8 positions.
On September 29, 2004, IBM announced that a Blue Gene/L prototype at IBM Rochester
(Minnesota) had overtaken NEC's Earth Simulator as the fastest computer in the world, with
a speed of 36.01 TFLOPS, beating Earth Simulator's 35.86 TFLOPS. The machine later
reached a speed of 70.72 TFLOPS.
Linux will be the main operating system for IBM's upcoming family of "Blue Gene"
supercomputers--a major endorsement for the operating system and the open-source
computing model it represents. The decision to adopt Linux came, in part, as a result of the
growing size and strength of the open-source community. Thousands of developers around
the world are participating in the evolution of Linux. Creating a new OS inside of IBM
would require a massive engineering effort.
On March 24, 2005, the US Department of Energy announced that Blue Gene/L broke its
current world speed record, reaching 135.5 TFLOPS. This feat was possible because of
doubling the number of racks to 32 with each rack holding 1,024 compute nodes. This is still
only half of the final configuration with 65,536 compute nodes. The final Blue Gene/L
installation will have a total of 65,536 compute nodes (i.e., 216 nodes) and an additional
1024 I/O nodes. Each compute or IO node is a single ASIC with associated DRAM memory
chips. The ASIC integrates two PowerPC 440 embedded processors, a cache sub-system and
communication sub-systems. Each node is attached to three parallel communications
networks: a 3D toroidal network for peer-to-peer communication between compute nodes, a
collective network for collective communication, and a global interrupt network for fast
barriers. The I/O nodes, which run the Linux operating system, provide communication with
the world via an Ethernet network. Finally, a separate and private Ethernet network provides
9. 2
1.2 MAJOR FEATURES
The Blue Gene/L supercomputer was unique in the following aspects:
• Trading the speed of processors for lower power consumption.
• Blue Gene/L used low frequency and low power embedded PowerPC cores with
floating point accelerators. While the performance of each chip was relatively low,
the system could achieve better performance to energy ratio, for applications that
could use larger numbers of nodes.
• Dual processors per node with two working modes: co-processor mode where one
processor handles computation and the other handles communication; and virtual-
node mode, where both processors are available to run user code, but the processors
share both the computation and the communication load.
• System-on-a-chip design. All node components were embedded on one chip, with
the exception of 512 MB external DRAM.
• A large number of nodes (scalable in increments of 1024 up to at least 65,536)
• Three-dimensional torus interconnect with auxiliary networks for global
communications (broadcast and reductions), I/O, and management .
• Lightweight OS per node for minimum system overhead (system noise).
10. 3
1.3 TECHNICAL DETAILS
The performance spectrum:
In computing, FLOPS is an abbreviation of FLoating point Operations Per Second.
This is used as a measure of a computer's performance, especially in fields of scientific
calculations that make heavy use of floating point calculations. (Note: a hertz is a cycle (or
operation) per second. Compare to MIPS -- million instructions per second.) One should
speak in the singular of a FLOPS and not of a FLOP, although the latter is frequently
encountered. The final S stands for second and does not indicate a plural. Computing devices
exhibit an enormous range of performance levels in floating-point applications, so it makes
sense to introduce larger units than the FLOPS. The standard SI prefixes can be used for this
purpose, resulting in such units as the megaFLOPS (MFLOPS, 106 FLOPS), the gigaFLOPS
(GFLOPS, 109 FLOPS), the teraFLOPS (TFLOPS, 1012 FLOPS), and the petaFLOPS
(PFLOPS, 1015 FLOPS).
A cheap but modern desktop computer using, for example, a Pentium 4 or Athlon 64 CPU,
typically runs at a clock frequency in excess of 2 GHz and provides computational
performance in the range of a few GFLOPS. Even some video game consoles of the late
1990s' vintage, such as the Gamecube and Dreamcast had performance in excess of one
GFLOPS .
The original supercomputer, the Cray-1, was set up at Los Alamos National Laboratory in
1976. The Cray-1 was capable of 80 MFLOPS (or, according to another source, 138–250
MFLOPS). In fewer than 30 years since then, the computational speed of supercomputers has
jumped a millionfold.
The fastest computer in the world as of November 5, 2004, the IBM Blue Gene
supercomputer, measures 70.72 TFLOPS. This supercomputer was a prototype of the Blue
Gene/L machine IBM is building for the Lawrence Livermore National Laboratory in
California. During a speed test on 24 March 2005, it was rated at 135.5 TFLOPS. Blue
Gene's new record was achieved by doubling the number of current racks to 32. Each rack
holds 1,024 processors, yet the chips are the same as those found in high-end computers
11. 4
CHAPTER 2
ARCHITECTURE DETAILS
Blue Gene/L (BG/L) is a scientific and engineering, message-passing, supercomputer
that IBM is developing with partial funding from the U.S. Department of Energy Lawrence
Livermore National Laboratory. A 64K node system is scheduled to be delivered to
Livermore, while a 20K node system will be installed at the IBM T.J. Watson Research
Center for use in life sciences computing, primarily protein folding.
BG/L is built using system-on-a-chip technology in which all functions of a node (except for
main memory) are integrated onto a single ASIC. This ASIC includes two 32- bit Power PC
cores (the 440); the 440 was developed for embedded applications. Associated with each core
is a 64- bit “double” floating-point unit (FPU) that can operate in SIMD mode. Each (single)
FPU can execute up to two multiply-adds per cycle, meaning that the peak performance of
the chip is 8 floating-point operations per cycle. Each 440 has its own instruction and data
caches (each 32KB), a small L2 cache that primarily serves as a pre-fetch buffer, a 4MB
shared L3 cache built from embedded DRAM, and a DDR memory controller. In addition,
the logic for five different networks is integrated onto the ASIC. These networks include a
JTAG control and monitoring network, a Gbit Ethernet macro, a global barrier and alert
network, a “tree” network for broadcasts and combining operations such as those used in the
MPI collective communications library, and a three dimensional torus network for point-
point communications between nodes. The ASIC can be used as either an I/O node or as a
Compute node. I/O nodes have their Ethernet macro connected to an external switch enabling
connectivity to hosts, however they do not use the torus network. Compute nodes do not
connect their Ethernet, and talk to the I/O nodes over the tree network. The Livermore
machine will have 64 Compute nodes for each I/O node. I/O nodes will have at least 512MB
and Compute nodes will have at least 256 MB of memory, depending on the cost of memory
at the time of delivery. Because of the high level of integration and relatively low target
clock speed (700 MHz target), the system is designed to deliver unprecedented aggregate
performance at both low cost and low power consumption. At this clock rate, each node has a
peak of 5.6 GFlops, while the 64K node system has a peak of 367 Tera Flops. Each ASIC
will consume only 12 watts of power. Because of the low power, a very high density of
packaging can be achieved.
12. 5
2.1 PROGRAMMINGMODELS
Message passing is expected to be the dominant parallel programming model for
BG/L applications. It is supported through an implementation of the MPI message-passing
library. In developing MPI for BG/L, we are paying particular attention to the issue of
efficient mapping of operations to the torus and tree networks. Also important is the issue of
efficient use of the second (communication) processor in a compute node. We are also
investigating two approaches to the global address space programming model for BG/L: Co-
arrays [NuRe98] and Unified Parallel C (UPC) [CDCYBW99].
2.2 CONTROLSYSTEM
The BG/L system software includes a set of control services that execute on the host
system. Many of these services, including system bring up, machine partitioning, measuring
system performance, and monitoring system health, are nonarchitected from a user
perspective, and are performed through the backdoor JTAG network described in Section 2
(which is also nonarchitected from a user perspective).
The resource management system of BG/L provides services to create electrically isolated
partitions of the machine and to allocate resources to jobs. Each partition is dedicated to the
execution of a single job at a time. Job scheduling and job control is also performed from the
host.
2.3 AUTONOMIC FEATURES
Given the scale of BG/L, there is clearly a need to recover from failures of individual
components. Support for long-running applications will be provided through a
checkpoint/restart mechanism. We are currently developing an application-assisted
checkpoint infrastructure. In this approach, the application programmer is responsible for
identifying points in the application in which there are no outstanding messages. The
programmer can then place calls to a system checkpoint in those points. When executed, the
checkpoint service will synchronize all tasks of the application and take a complete
application checkpoint, writing the state of all compute processes to disk. Application state
that resides on I/O nodes, particularly file pointers and list of open files, is also saved to disk.
In case of unexpected termination, the application can then be restarted from its latest
checkpoint.
13. 6
2.4 SYSTEM OVERVIEW
BlueGene/L is a scalable system in which the maximum number of compute nodes
assigned to a single parallel job is 216 = 65,536. BlueGene/L is configured as a 64 x 32 x 32
three-dimensional torus of compute nodes. Each node consists of a single ASIC and memory.
Each node can support up to 2 GB of local memory; our current plan calls for 9 SDRAM-
DDR memory chips with 256 MB of memory per node. The ASIC that powers the nodes is
based on IBM’s system-on-a-chip technology and incorporates all of the functionality needed
by BG/L. The nodes themselves are physically small, with an expected 11.1-mm square die
size, allowing for a very high density of processing. The ASIC uses IBM CMOS CU-11 0.13
micron technology and is designed to operate at a target speed of 700 MHz, although the
actual clock rate used in BG/L will not be known until chips are available in quantity.
The current design for BG/L system packaging is shown in Figure 1. (Note that this is
different from a preliminary design shown in [ISSCC02] as are certain bandwidth figures that
have been updated to reflect a change in the underlying signaling technology.) The design
calls for 2 nodes per compute card, 16 compute cards per node board, 16 node boards per
512-node midplane of approximate size 17”x 24”x 34,” and two midplanes in a 1024-node
rack. Each processor can perform 4 floating point operations per cycle (in the form of two
64-bit floating point multiply-add’s per cycle); at the target frequency this amounts to
approximately 1.4 teraFLOPS peak performance for a single midplane of BG/L nodes, if we
count only a single processor per node. Each node contains a second processor, identical to
the first although not included in the 1.4 teraFLOPS performance number, intended primarily
for handling message passing operations. In addition, the system provides for a flexible
number of additional dual-processor I/O nodes, up to a maximum of one I/O node for every
eight compute nodes. For the machine with 65,536 compute nodes, we expect to have a ratio
one I/O node for every 64 compute nodes. I/O nodes use the same ASIC as the compute
nodes, have expanded external memory and gigabit Ethernet connections. Each compute
node executes a lightweight kernel. The compute node kernel handles basic communication
tasks and all the functions necessary for high performance scientific code. For compiling,
diagnostics, and analysis, a host computer is required. An I/O node handles communication
between a compute node and other systems, including the host and file servers. The choice of
host will depend on the class of applications and their bandwidth and performance
requirements.
14. 7
The nodes are interconnected through five networks: a 3D torus network for point-topoint
messaging between compute nodes, a global combining/broadcast tree for collective
operations such as MPI_Allreduce over the entire application, a global barrier and interrupt
network, a Gigabit Ethernet to JTAG network for machine control, and another Gigabit
Ethernet network for connection to other systems, such as hosts and file systems. For cost
and overall system efficiency, compute nodes are not hooked directly up to the Gigabit
Ethernet, but rather use the global tree for communicating with their I/O nodes, while the I/O
nodes use the Gigabit Ethernet to communicate to other systems.
In addition to the compute ASIC, there is a “link” ASIC. When crossing a midplane
boundary, BG/L’s torus, global combining tree and global interrupt signals pass through the
BG/L link ASIC. This ASIC serves two functions. First, it redrives signals over the cables
between BG/L midplanes, improving the high-speed signal shape and amplitude in the
middle of a long, lossy trace-cable-trace connection between nodes on different midplanes.
Second, the link ASIC can redirect signals between its different ports. This redirection
function enables BG/L to be partitioned into multiple, logically separate systems in which
there is no traffic interference between systems. This capability also enables additional
midplanes to be cabled as spares to the system and used, as needed, upon failures. Each of
the partitions formed through this manner has its own torus, tree and barrier networks which
are isolated from all traffic from all other partitions on these networks.
System fault tolerance is a critical aspect the BlueGene/L machine. BlueGene/L will have
many layers of fault tolerance that are expected to allow for good availability despite the
large number of system nodes. In addition, the BlueGene/L platform will be used to
investigate many avenues in autonomic computing.
16. 9
CHAPTER 3
DESIGN AND ANALYSIS OF THE BLUE GENE/L TORUS
INTERCONNECTION NETWORK
3.1 TORUS NETWORK
Many of the design decisions were driven by simulation performance studies. The
torus network uses dynamic routing with virtual cut through buffering. A torus was chosen
because it provides high bandwidth nearest neighbor connectivity, which is common in
scientific applications, but also for its scalability, cost and packaging considerations. A torus
requires no long cables and, because the network is integrated onto the same chip that does
computing, no separate switch is required. Previous supercomputers such as the Cray T3E
have also used torus networks. Torus packets are variable in size – from 32 to 256 bytes in
increments of 32 byte chunks. The first eight bytes of each packet contain link level protocol
information (e.g., sequence number) and routing information including destination, virtual
channel and size. A 24-bit CRC is appended to each packet, along with a one byte valid
indicator. The CRC permits link level checking of each packet received, and a timeout
mechanism is used for retransmission of corrupted packets. The error detection and recovery
protocol is similar to that used in IBM SP interconnection networks as well as in the HIPPI
standard. For routing, the header includes six “hint” bits, which indicate in which directions
the packet may be routed. For example, hint bits of 100100 means that the packet can be
routed in the x+ and y- directions. Either the x+ or x- hint bits, but not both, may be set. If no
x hops are required, the x hint bits are set to 0. Each node maintains registers that contain the
coordinates of its neighbors, and hint bits are set to 0 when a packet leaves a node in a
direction such that it will arrive at its destination in that dimension. These hint bits appear
early in the header, so that arbitration may be efficiently pipelined. The hint bits can be
initialized either by software or hardware; if done by hardware, a set of two registers per
dimension is used to determine the appropriate directions. These registers can be configured
to provide minimal hop routing. The routing is accomplished entirely by examining the hint
bits and virtual channels, i.e., there are no routing tables. Packets may be either dynamically
or statically (xyz) routed. Besides point-topoint packets, a bit in the header may be set that
causes a packet to be broadcast down any dimension. The hardware does not have the
capability to route around “dead” nodes or links, however, software can set the hint bits
17. 10
appropriately so that such nodes are avoided; full connectivity can be maintained when there
are up to three faulty nodes, provided they are not co-linear.
The torus logic consists of three major units, a processor interface, a send unit and a receive
unit. The processor interface consists of network injection and reception FIFOs. Access to
these FIFOs is via the double FPU registers, i.e., data is loaded into the FIFOs via 128 bit
memory mapped stores from a pair of FPU registers, and data is read from the FIFOs via 128
bit loads to the FPU registers. There are a total of 8 injection FIFOs organized into two
groups: two high priority (for inter-node OS messages) and six normal priority FIFOs, which
are sufficient for nearest neighbor connectivity. Packets in all FIFOs can go out in any
direction. Each group of reception FIFOs contains 7 FIFOs, one high priority and one
dedicated to each of the incoming directions. More specifically, there is a dedicated bus
between each receiver and its corresponding reception FIFO. Up to six injection and six
reception FIFOs may be simultaneously active. Each of the six receivers, as shown in Figure
1, has four virtual channels (VCs). Multiple VCs help reduce head-ofline blocking [4], but in
addition, mesh networks including tori with dynamic routing, can deadlock unless
appropriate additional “escape” VCs are provided. We use a recent, elegant solution to this
problem, the “bubble” escape VC as proposed in BG/L has two dynamic VCs, one bubble
escape VC that can be used both for deadlock prevention and static routing, and one high
priority bubble VC. Each VC has 1 KB of buffering, enough for four full-sized packets. In
addition to the VCs, the receivers include a “bypass” channel so that packets can flow
through a node without entering the VC buffers, under appropriate circumstances. Dynamic
packets can only enter the bubble escape VC if no valid dynamic VCs are available. A token
flow control algorithm is used to prevent overflowing the VC buffers. Each token represents
a 32B chunk. For simplicity in the arbiters, a VC is marked as unavailable unless 8 tokens (a
full-sized packet) are available. However, token counts for packets on dynamic VCs are
incremented and decremented according to the size of the packet. The bubble rules, as
outlined in require that tokens for one full-sized packet are required for a packet already on
the bubble VC to advance, but that tokens for two full-sized packets are required for a packet
to enter the bubble VC, upon either injection, a turn into a new direction, or when a dynamic
VC packet enters the bubble. This rule ensures that buffer space for one packet is always
available after an insertion and thus some packet can always, eventually move. However, we
discovered that this rule is incomplete for variable-sized packets when our simulator
deadlocked using this rule. With this rule, the remaining free space for one full-sized packet
18. 11
can become fragmented resulting in a potential deadlock. To prevent this, the bubble rules
are simply modified so that each packet on the bubble is accounted for as if it were a
fullsized (8 chunk) packet. Eight byte acknowledgement (ack-only) or combined token-
acknowledgement (token-ack) packets are returned when packets are either successfully
received, or when space has freed up in a VC. Acknowledgements permit the torus send units
to delete packets from their retransmission FIFOs, which are used in the error recovery
protocol. The send units also arbitrate between requests from the receiver and injection units.
Due to the density of packaging and pin constraints, each link is bit serial. The torus is
internally clocked at onefourth the rate of the processor, so at the target 700 MHz clock rate,
each torus link is 175 MB/sec. There are sufficient internal busses so that each of the 6
outgoing and 6 incoming links can be simultaneously busy; thus each node can be sending
and receiving 1.05 GB/sec. In addition, there are two transfer busses (paths) coming out of
each receiver that connect with the senders. Thus, a single receiver can have up to 4
simultaneous transfers, e.g., one to its normal reception FIFO, one to the high priority
reception FIFO, and two to two different senders.
Arbitration is distributed and pipelined, but occurs in three basic phases. It generalizes an
approach used in [3] and represents tradeoffs between complexity, performance, and ability
to meet timing constraints. First, each packet at the head of the injection or VC FIFOs
decides in which direction and on what VC it prefers to move. For statically routed packets,
there is only one valid choice, but dynamically routed packets may have many choices. The
preferred direction and VC are selected using a modified “Join the Shortest Queue” (JSQ)
algorithm as follows. The senders provide the receivers and injection FIFOs with a bit
indicating both link and token availability for each VC in each direction. This bit vector is
and-ed with a bit vector of possible moves constructed from the packet’s hint bits and VC.
This defines the set of possible and available arbitration requests. In addition, the sender
provides 2 bits for each VC indicting one of four ranges of available downstream tokens. Of
all the possible and available dynamic direction/VC pairs, the packet selects the one with the
most available downstream tokens. Ties are randomly broken. If no dynamic direction/VC
combination is available, the packet will request its bubble escape direction/VC pair (if
available), and if that is also unavailable, the packet makes no arbitration request. This is a
somewhat simplified description since bus availability must also be taken into account. In
addition, when a packet reaches its destination, the “direction” requested is simply .
19. 12
3.2 SIMULATOR OVERVIEW
Given the complexity and scale of the BG/L interconnection network, having an
accurate performance simulator was essential during the design phase of the project. Due to
the potential size of such a model, simulation speed was a significant concern and a proven
shared memory parallel simulation approach was selected. In particular, parallel simulation
on shared memory machines has been shown to be very effective in simulating
interconnection networks whereas success with message passing parallel interconnection
network simulators is harder to come by .We also recognized the difficulties in developing an
execution driven simulator for a system with up to 64K processes, and therefore decided
upon a simulator that would primarily be driven by application pseudo-codes, in which
message passing calls could be easily passed to the simulator; such calls include the time
since the last call (the execution burst time), the destination and size of the message, etc. This
pseudo-code included a subset of the MPI point to point messaging calls as a workload driver
for the simulator. We also extended the IBM UTE trace capture utility that runs on IBM SP
machines and were able to use such traces as simulator inputs (for up to several hundreds of
nodes). The basic unit of simulation time is a network cycle, which is defined to be the time
it takes to transfer one byte. As BG/L is organized around 512 node (8x8x8) midplanes, the
simulator partitions its work on a midplane basis, i.e., all nodes on the same midplane are
simulated by the same processor (thread) and midplanes are assigned to threads in as even a
manner as possible.
Because different threads are concurrently executing, the local simulation clocks of the
threads need to be properly synchronized. To deal with this problem, we use a simple but
effective “conservative” parallel simulation protocol known as “YAWNS” .In particular, we
take advantage of the fact that the minimum transit time between midplanes is known and is
at least some constant w≥1 cycles. In this protocol, time “windows” of length w are
simulated in parallel by each of the threads. Consider an event that is executed during the
window (starting at time t) on processor i that is destined to arrive on processor j in the
future; such an event represents the arrival of the first byte of a packet. Since the minimum
transit time is w, the arrival cannot occur during the current window, represented by the
interval [t, t+w-1]. Processor i simply puts a pointer to the event on an i-to-j linked list. When
each processor reaches the end of the window, it enters a barrier synchronization. Upon
leaving the barrier, each processor is sure that every other processor has executed all events
20. 13
up to time t+w-1 and that all inter-processor events are on the appropriate inter-processor
linked lists. Processor j can therefore go through all its i-to-j linked lists, remove events from
them, and put the events on its own future event list. Once this is done, the processors can
simulate the next window [t+w, t+2w-1]. If w=1, then this protocol requires a barrier
synchronization every cycle, however, on BG/L, the minimum inter-midplane delay will be
approximately w=10 network cycles. When a large number of BG/L nodes are being
simulated, each processor will execute many events during a window, i.e., between barriers,
and thus the simulator should obtain good speedups. The simulator runs on a 16-way IBM
“nighthawk” SMP with 64 GB of memory. The model of the torus hardware contains close to
100 resources per node (links, VC token counters, busses, FIFOs, etc), so that a full 64K
node system can be thought of as a large queuing network with approximately 6 million
resources. It consumes a large amount of memory and runs slowly; a 32K node simulation of
fully loaded network advances at about 0.25 microseconds of BG/L time per second of wall
clock time. However, it obtains excellent speedup, typically more than 12 on 16 nodes, and
sometimes achieves superlinear speedup due to the private 8MB L3 caches on the SMP and
the smaller per node memory footprint of the parallel simulator. The model, which was
written before the VHDL, is thought to be a quite accurate representation of the BG/L
hardware, although a number of simplifications were made. For example, in BG/L the
arbitration is pipelined and occurs over several cycles. In the simulator, this is modeled as a
delay of several cycles followed by presentation of the arbitration request. Because the
simulator focuses on what happens once packets are inside the network, a gross
simplification was the assumption that the injection FIFOs were of infinite size, and that
packets are placed in these FIFOs as early as possible rather than as space frees up in the
FIFOs. This has little effect on network response time and throughput measurements during
the middle of a run, but can affect the dynamics particularly near the end of runs. The
simulator also did not model the error recovery protocol, i.e., no link errors were simulated
and the ackonly packets that are occasionally sent if a link is idle for a long time were not
modeled. However, the arbitration algorithms and token flow control are modeled to a high
level of detail.
21. 14
CHAPTER 4
SAMPLE PERFORMANCE STUDIES
In this section, we present some examples of use of the simulator to study design
trade-offs in BG/L. The studies presented are illustrative and sometimes use assumptions and
corresponding parameters about the system that do not reflect the final BG/L design.
Response Time in Light Traffic: Figure 2 plots the response time for various 32K node BG/L
configurations when the workload driver generates packets for random destinations and the
packet generation rate is low enough so that the average link utilization is less than one. This
Figure compares static routing to dynamic routing with one or more dynamic VCs and one or
more busses (paths) connecting receivers to senders. Simpler, random, arbitration rules than
SLQ and JSQ were used and the plot was generated early in our studies when the target link
bandwidth was 350 MB/sec. (The 350 MB/sec. assumption essentially only affects results by
a rescaling of the y-axis.) The figure shows the clear benefit of dynamic over static routing. It
also shows that there is little benefit in increasing the number of dynamic VCs unless the
number of paths is also increased. Finally, it shows only marginal benefit in going from a 2
VC/2 path to 4 VC/4 path configuration.
All-to-All: MPI_AlltoAll is an important MPI collective communications operation in which
every node sends a different message to every other node. plots the average link utilization
during the communications pattern implied by this collective. The Figure again shows the
benefit of dynamic over static routing. For this pattern, there is marginal benefit in going
from 1 to 2 dynamic VCs, but what is important is that the average link utilization is, at
approximately 98%, close to the theoretical peak. This peak includes the overhead for the
token-ack packets, the packet headers and the 4 byte CRC trailers. A reasonable assumption
for the BG/L software is that each packet carries 240 bytes of payload, and with this
assumption the plot shows that the payload occupies 87% of the links. Not shown in these
plots is the fact that a very low percentage of the traffic flows on the escape bubble VC and
that statistics collected during the run showed that few of the VC buffers are full. Three-
dimensional FFT algorithms often require the equivalent of an All-to-All, but on a subset of
the nodes consisting of either a plane or a line in the torus. Simulations of these
communications patterns also resulted in near-peak performance. The above simulation was
for a symmetric BG/L. However, the situation is not so optimistic for an asymmetric
22. 15
BG/L.For example, the 64K node system will be a 64x32x32 node torus. In such a system,
the average number of hops in the x dimension is twice that of the y and z dimensions, so that
even if every x link is 100% busy, the y and z links can be at most 50% busy. Thus, the peak
link utilization is at most 66.7%. Since 12% of that is overhead, the best possible payload
utilization is 59%. However, we expect significantly more blocking and throughput
degradation due to full VC buffers. Indeed a simulation of the All-to-All communications
pattern on a 32x16x16 torus resulted in an average link utilization of 49% and payload
utilization of 44%, corresponding to 74% of the peak. This figure is probably somewhat
pessimistic due to the simulator artifact of infinite-sized injection FIFOs, which distorts the
effects at the end of the simulation. We also believe that appropriate injection flow control
software algorithms can reduce VC buffer blocking and achieve closer to peak performance.
Nevertheless, the above study points out a disadvantage of the torus architecture for
asymmetric machines in which the application cannot be easily mapped so as to result in a
close proximity communications pattern.
Virtual Channel Architecture: Here we consider several different deadlock prevention escape
VC architectures. The first proposed has two escape VCs per direction. Each dimension has a
“dateline.” Before crossing the dateline, the escape VC is the lower numbered of the pair, but
after crossing the dateline the escape VC is the higher numbered of the pair. In addition we
consider dimension ordered or direction ordered escape VCs. In dimension ordered, the
escape VC is x first, then y if no x hops remain, then z if no x or y hops remain. In direction
ordered, the escape VCs are ordered by x+, y+, z+, x-, y-, z- (other orderings are possible).
We also consider dimension and direction ordered escape VCs for the bubble escape. We
again use the hot region workload where the hot region starts at coordinates (0,0,0) and the
datelines are set at the maximum coordinate value in each dimension. plots the throughput as
a function of time. The dimension ordered dateline pair shows particularly poor and wild
behavior, with a steep decline in throughput, followed by a rise and then another steep
decline. plots the throughput on a per VC basis for a longer period of time. The decreasing
and increasing bandwidth waves persist even over this much longer time scale. An
appreciable fraction of the traffic flows on the escape VCs, indicating a high level of VC
buffer occupation. What causes these waves? First, the placement of the dateline causes an
asymmetry in the torus, whereas the bubble escape is perfectly symmetrical in each
23. 16
dimension. Since there are two escape VCs, we thought it likely that packets at the head of
the VC buffers could be waiting for one of the escape VCs but tokens are returned for the
other escape VC. In such a situation, no packets could move even though the link may be
available and downstream buffer space is available. To confirm this, the simulator was
instrumented to collect additional statistics. In particular, we measured the fraction of time a
token-ack is returned that frees at least one previously blocked packet to move. plots this
unblocking probability along with the throughput as a function of time. The unblocking
probability is relatively constant for the bubble (after the initial decline), but varies directly
with the throughput for the dateline pair; when the unblocking probability increases, the
throughput increases and vice-versa.
Performance Verification: To verify the VHDL logic of the torus, we built a multi-node
verification testbench. This testbench, which runs on the Cadence VHDL simulator,
consisted of workload drivers that inject packets into the injection FIFOs, links between
nodes on which bits could be corrupted to test the error recovery protocol, and packet
checkers that pull packets out of the reception FIFOs and check them for a variety of
conditions, such as whether the packet arrived at the correct destination and whether its
contents were received correctly. The workload drivers could be flexibly configured to
simulate a number of different traffic patterns. As we neared the end of the logic verification
process, we wanted to ensure that network performance was as intended. One of the
benchmarks we tested was the All-to- All. The VHDL simulator was limited (by memory) to
a maximum of 64 nodes, so we simulated both a 4x4x4 torus and an 8x8x1 torus and
compared the average link utilizations to those predicted by the performance simulator.
While these agreed to within 2%, the VHDL (corresponding to the actual network hardware)
indicated that VC buffers were fuller than that predicted by the performance simulator. A
close inspection of the arbitration logic revealed that a one cycle gap in the arbitration
pipeline of the receivers could occur when all possible outgoing links/VCs were busy. This
gap was sufficient to permit packets from the injection FIFOs to sneak into the network,
leading to fuller VCs than intended. A simple fix to eliminate this possibility was
implemented, and subsequent VHDL simulations indicated greatly reduced levels of VC
buffer occupation.
26. 19
CHAPTER 5
APPLICATION
5.1 GENERAL
Machines like Blue Gene/L are designed to handle data-intensive applications like
content distribution, simulations, and modeling, webserving, data mining or business
intelligence.
• Veselin Topalov, the challenger to the World Chess Champion title in 2010,
confirmed in an interview that he had used a Blue Gene/P supercomputer during his
preparation for the match.
• The Blue Gene/P computer has been used to simulate approximately one percent of a
human cerebral cortex, containing 1.6 billion neurons with approximately 9 trillion
connections.
• The IBM Kittyhawk project team has ported Linux to the compute nodes and
demonstrated generic Web 2.0 workloads running at scale on a Blue Gene/P. Their
paper, published in the ACM Operating Systems Review, describes a kernel driver
that tunnels Ethernet over the tree network, which results in all-to-all TCP/IP
connectivity. Running standard Linux software like MySQL, their performance
results on SpecJBB rank among the highest on record
• In 2011, a Rutgers University / IBM / University of Texas team linked the KAUST
Shah een installation together with a Blue Gene/P installation at the IBM Watson
Research Center into a "federated high performance computing cloud", winning the
IEEE SCALE 2011 challenge with an oil reservoir optimization application
Another most important application is to predict how chains of biochemical building blocks
described by DNA fold into proteins--massive molecules such as hemoglobin. Most
biological functions involve proteins and while a protein's chemical composition is
determined by a sequence of amino acids joined like links of a chain, a protein folds into a
highly complex, three-dimensional shape such as illustrated in the two figures below.
27. 20
Fig 5.1 :-DNA
It is hypothesized that the shape of a protein is the principal determinant of its function.
Arbitrary strings of amino acids do not, in general, fold into a well-defined three-dimensional
structure, but evolution has selected out the proteins used in biological processes for their
ability to fold reproducibly (sometimes with assistance, sometimes without) into a particular
three-dimensional structure within a relatively short time. Some diseases are actually caused
by slight misfoldings of a particular protein. Understanding the mechanisms that cause a
string of amino acids to fold into a specific three-dimensional structure is an outstanding
scientific challenge. Appropriate use of large scale biomolecular simulation to study protein
folding is expected to shed significant light into this process. Extensive collaborations with
the biological research community will be needed to find the best way of applying the unique
computational resources available to the Blue Gene project to advance our understanding of
protein folding. The level of performance provided by Blue Gene (sufficient to simulate the
folding of a small protein in a year of running time) is expected to enable a tremendous
increase in the scale of simulations that can be carried out as compared with existing
supercomputers.
The scientific community considers protein folding one of the most significant "grand
challenges" -- a fundamental problem in science or engineering that has broad economic and
scientific impact and whose solution can be advanced only by applying high-performance
computing technologies.
Proteins control all cellular processes in the human body. Comprising strings of amino acids
that are joined like links of a chain, a protein folds into a highly complex, three-dimensional
shape that determines its function. Any change in shape dramatically alters the function of a
protein, and even the slightest change in the folding .
28. 21
5.2 PROTEIN ARCHITECTURE
Protein architecture8 is based on three principles:
• The formation of a polymer chain .
• The folding of this chain into a compact function-enabling structure, or native
structure .
• Post-translational modification of the folded structure .
The protein chain (or peptide chain if short in length) is a heteropolymer built up from alpha
amino acid monomers, as shown in Figure . The sequence of amino acid residues in the
peptide chain is termed the primary structure of the protein. The 20 different choices for each
amino acid in the chain give the possibility of enormous diversity, even for small proteins.
For example, a peptide of 30 residues yields the astonishing number of about 2030, or
approximately 1039, possible unique sequences.
Fig 5.2 :- Protein architecture
5.3 THE PROTEIN FOLDING PROBLEM
There are two important facets to the protein folding problem: prediction of three-
dimensional structure from amino acid sequence, and understanding the mechanisms and
pathways whereby the three-dimensional structure forms within biologically relevant
timescales.
The prediction of structure from sequence data is the subject of an enormous amount of
research and a series of conferences that assess the state of the art in structure prediction.9
While this area is extremely important, good progress in the area of structural predictions has
been made using only modest amounts of computational power. The effort described in this
paper is aimed at improving our understanding of the mechanisms behind protein folding,
rather than at structure prediction. Even though biologists have been most interested .
29. 22
5.4 CURRENT VIEW OF FOLDING MECHANISMS
Groups (see Figure, caption) fall into three main classes: (1) charged, (2) hydrophilic
(“water-loving”), and (3) hydrophobic (“water-hating”). In the simplest picture, the folded
state of the peptide chain is stabilized primarily (for a globular protein in water), by the
sequestration of much of the hydrophobic groups into the core of the protein—out of contact
with water, while the hydrophilic and charged groups remain in contact with water. The
stability can be described in terms of the Gibbs free-energy change G
G = H – TS,
where H is the enthalpy change and S is the entropy change. H is negative due to the more
favorable hydrophobic interactions in the folded state, but so is S because the folded state is
much more ordered and has lower entropy than the unfolded state. The balance between the
enthalpy and entropy terms is a delicate one, and the total free-energy change is only of order
15 kilocalories per mole. Evidently the internal hydrophobic/external hydrophilic packing
requirement places strong constraints on the amino acid sequence, as does the requirement
that the native state be kinetically accessible.
It is helpful to think of the physics of the folding process as a “free-energy funnel Since the
folding process is slow relative to motions at atomic scale, we can think of partially folded
configurations as having a quasi-equilibrium value of the free energy. The free energy
surface may be displayed as a function of some reduced dimensionality representation of the
system configuration in a given state of the protein.12 . The most unfolded configurations are
the most numerous, but have the highest free energy, and occur on the rim of the funnel.
Going into the funnel represents a loss of number of configurations (decrease of entropy), but
a gradual decrease in free energy, until the native state with very few configurations and the
lowest free energy is reached at the bottom of the funnel. The walls of the funnel contain
only relatively shallow subsidiary minima, which can trap the folding protein in non-native
states, but only for a short time. Now the evolution of the system as it folds can be described
in terms of the funnel. The system starts off in a physically probable state on the rim of the
funnel, and then makes transitions to a series of physically accessible states within the funnel,
until the bottom of the funnel is gradually approached.
Figure 3 illustrates folding. Here the unfolded peptide chain on the left already contains some
folded secondary structure, alpha helices (red), and a beta hairpin (blue). It is still a long way
30. 23
from the compact native structure at right. The folding process in different proteins spans an
enormous dynamic range from approximately 20 microseconds to approximately 1 second.
Fig 5.4 :- Folding
The scientific knowledge derived from research on protein folding can potentially be applied
to a variety of related life sciences problems of great scientific and commercial interest,
including:
• Protein-drug interactions (docking)
• Enzyme catalysis (through use of hybrid quantum and classical methods)
• Refinement of protein structures created through other methods
We shall also explore the use of Blue Gene in other scientific computing areas. We expect
that lessons learned from this project will apply to future high performance IBM systems in a
broader range of scientific and commercial applications.
Examples of those applications include the modeling of the aging and properties of materials,
and the modeling of turbulence. This technology opens the door to a number of applications
of great interest to civilian industry and business, like biology and other life sciences. The
future of US high-performance computing will benefit tremendously from pursuing both of
these paths in parallel.
"One day, you're going to be able to walk into a doctor's office and have a computer
analyze a tissue sample, identify the pathogen that ails you, and then instantly prescribe a
treatment best suited to your specific illness and individual genetic makeup."
Consider the following three types of protein science studies that might employ large-scale
numerical simulation techniques:
• Structure prediction
31. 24
• Folding pathway characterization
• Folding kinetics
Protein structure prediction can be carried out using a large number of techniques8 and, as
previously discussed, it is unnecessary to spend a “petaflop year” on the prediction of a
single protein structure. That said, there is some reason to believe that atomistic simulation
techniques may be useful in refining structures obtained by other methods.
Folding pathway characterization typically involves the study of thermodynamic properties
of a protein in quasi-equilibrium during the folding process. Mapping out the free-energy
“landscape” that the protein traverses as it samples conformations during the folding process
can give insights into the nature of intermediate states along the folding pathway and into the
“ruggedness” of the free-energy surface that is traversed during this process. Because such
studies involve computations of average values of selected functions of the system's state,
one has the choice of either averaging over time as the system samples a large number of
states (molecular dynamics) or averaging over configurations (Monte Carlo). Aggressive
sampling techniques that may improve the computational efficiency with which such
averages can be computed can be used to good effect in these studies. Simulation techniques
to compute these averages over the appropriate thermodynamic ensembles are available.
Simulation studies of folding kinetics are aimed at understanding the rates at which the
protein makes transitions between various conformations. In this case, the calculation of
thermodynamic averages is not enough; the actual dynamics of the system must be simulated
with sufficient accuracy to allow estimation of rates. Of course, a large number of transition
events must be simulated in order to derive rate estimates with reasonable statistical
uncertainties. Another challenge faced in such simulations is that the simulation techniques
used to reproduce thermodynamic averages in ensembles other than constant particle number,
volume, and energy (NVE) are, strictly speaking, inappropriate for studies of folding
kinetics.
32. 25
5.5. CHALLENGES FOR COMPUTATIONAL MODELING
The current expectation is that it will be sufficient to use classical techniques, such as
molecular dynamics (MD), to model proteins in the Blue Gene project. This is because many
aspects of the protein folding process do not involve the making and breaking of covalent
bonds. While disulfide bonds play a role in many protein structures, their formation will not
be addressed by classical atomistic simulations. In classical atomistic approaches, a model
for the interatomic interactions is used. This is known as a potential, or force field, since the
forces on all the particles can be computed from it, if one has its mathematical expression
and all its parameters. The MD approach is to compute all the forces on all the atoms of the
computer model of the protein and solvent, then use that force to compute the new positions
of all the atoms a very short time later. By doing this repeatedly, a trajectory of the atoms of
the system can be traced out, producing atomic coordinates as a function of time.
system samples a large number of states (molecular dynamics) or averaging over
configurations (Monte Carlo). Aggressive sampling techniques that may improve the
computational efficiency with which such averages can be computed can be used to good
effect in these studies. Simulation techniques to compute these averages over the appropriate
thermodynamic ensembles are available.
Simulation studies of folding kinetics are aimed at understanding the rates at which the
protein makes transitions between various conformations. In this case, the calculation of
thermodynamic averages is not enough; the actual dynamics of the system must be simulated
with sufficient accuracy to allow estimation of rates. Of course, a large number of transition
events must be simulated in order to derive rate estimates with reasonable statistical
uncertainties. Another challenge faced in such simulations is that the simulation techniques
used to reproduce thermodynamic averages in ensembles other than constant particle number,
volume, and energy (NVE) are, strictly speaking, inappropriate for studies of folding
kinetics.
33. 26
CHAPTER 6
CONCLUSION
"Blue Gene" is an ambitious project to expand the horizons of supercomputing, with
the ultimate goal of creating a system that can perform one quadrillion calculations per
second, or one petaflop. IBM is hoping that expanded performance, more efficient data
access for processors, and lower operational costs will give Blue Gene a big leg up in the
world of high-performance computing.
The Blue Gene project represents a unique opportunity to explore novel research into a
number of areas, including machine architecture, programming models, algorithmic
techniques, and biomolecular simulation science. Every aspect of this highly adventurous
project involves significant challenges. Carrying out our planned program will require a
collaborative effort across many disciplines and the involvement of the worldwide scientific
and technical community. In particular, the scientific program will engage with the life
sciences community in order to make best use of this unique computational resource.
34. 27
REFERENCES
[1] www.research.ibm.com/bluegene
[2] www.research.ibm/journals/sj/402/allen.html
[3] www.bio-itworld.com/news/071503-report2898.html
[4] www.linuxdevices.com
[5] Adiga et al., (2002). An Overview of the BG/L Supercomputer. Proceedings of the
2002 Supercomputing Conference www.scconference. org/sc2002/
[6] Benveniste, C. and Heidelberger, P. (1995). Parallel Simulation of the IBM
Interconnection Network. In Proceedings of the 1995 Winter Simulation Conference.
IEEE Computer Society Press, 584 – 589.
[7] Dally, W.J. (1992). Virtual-Channel Flow Control. IEEE Transactions on Parallel and
Distributed Systems 3, No. 2, 194-205.
[8] Puente, V., Beivide, R., Gregorio, J.A., Prellezo, J.M., Duato, J., and Izu, C. (1999).
Adaptive Bubble Router: A Design to Improve Performance in Torus Networks. In
Proceedings of the 1999 International Conference on Parallel Processing, 58-67.
[9] Dally, W.J. and Seitz, C.L. (1987). Deadlock-Free Message Routing in
Multiprocessor Interconnection Networks. IEEE Transactions on Computers C-36,
No. 5, 547-553.
[10] Dickens, P.M., Heidelberger, P., and Nicol, D.M. (1996). Parallelized Direct
Execution Simulation of Message-Passing Parallel Programs. IEEE Transactions on
Parallel and Distributed Systems 7, No. 10, 1090-1105
[11] IBM Research Report on torus interconnection network by M. Blumrich, D. Chen, P.
Coteus, A. Gara, M. Giampapa and P. Heidelberger.