VenturiCoreExploiting the power of multi-core and dense memory x86 systems to implement a Grid-In-A-Box. Delivering Value at Risk In-A-Box
VenturiCoreThe Forward Scaling x86 PlatformVenturiCore is a patented technology and: • A library of C++ classes which implement software components delivering a dataflow architecture based on x86 technologies . • An x86 code base lifetime extender which makes optimum use of the increasing number of processors, associated cache and system memory. • An effective alternative to large scale and costly software rewrites.
What Venturicore is… In memory Multi-core 64bit address spaces Parallel calculation Forward Scaling x86 architecture
What Venturicore does…Pre-process 200Mrecords Calculate and select from 200M records Gather results200 Million records < 10 secondssynchronous processing over 24 threadsForward scaling technology more cores = more processing = less time
How VenturiCore works… • Allows minimum recalculation DAG Based • Generationally scheduled Automated • Less expensive to implement multi- • Goes faster with more processors threading Processor • Accelerates performance Cache • Allows optimisation of Management heterogeneous systems
VenturiCoreForward Scaling x86 Compute PerformanceComparing compute on x86, GPGPU & FPGA systems:In recent tests at Intel in August 2011 a VenturiCore-enabled application demonstrated superior throughput on anx86 platform compared to the documented performance of either FPGA or GPGPU systems. Monte Carlo Comparisons 5000 4500 4000 3500 3000 M per Second 2500 2000 1500 1000 500 0 VenturiCore & Altera - FPGA GPGPU – M2050 Intel Quad E7 2011 1800 1900 4700CommentsVenturiCore & Quad 2.4G E7 processor at Winnersh, M2050 nVidia GPGPUSources: Altera, HP, nVidia
VenturiCoreForward Scaling x86 Cache Performance Throughput effect of Vector Cache Management 70.00 60.00 50.00 40.00 Millions of Records processed per second 30.00 20.00 10.00 0.00 0.5 2 5 20 50 200 500 Vector Cache Management Off 3.82 5.11 6.70 19.05 23.81 23.47 23.96 Vector Cache Management On 8.77 22.47 33.12 43.48 52.03 58.93 57.20 Millions of records to be processed ( Data size)
VenturiCoreForward Scaling x86 Cache PerformanceResults comparing UCS Xeon x5670 and E7 servers:• Selecting from 200 Million in memory records ( 8 fields) and calculating analytics on a 16.2 Million subset.• Within the Intel Nehalem processor family VenturiCore demonstrated the significant gains in throughput in moving from a single, non hyper-threaded Intel x5670 to dual hyper-threaded E7 processors – using the same installed software in each instance – processing different tasks in parallel to deliver superior overall application performance.
VenturiCoreForward Scaling x86 Cache PerformanceMaintaining throughput on x86 systems:VenturiCore actively manages the data flow through each processor package. Memory access times Main Memory •>45ns L3 Cache •>15ns L2 Cache •>5ns L1 Cache •>1ns Intel & AC confirmed that VenturiCore uses L1 & L2 Cache more than 98% of the time during data processing using E7 systems. This factor alone would account for a 10 fold speed increase in data processing using VenturiCore; this technology makes extensive use of the capabilities of the x86 processor family.
VenturiCoreForward Scaling x86 Throughput 6 12 Xeon x5670 Threads in total 24 40 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 Millions of Records per second
VenturiCoreForward Scaling x86 ThroughputResults comparing UCS Xeon x5670 and E7 servers:• Selecting from 200 Million in memory records ( 8 fields) and calculating analytics on a 16.2 Million subset.• Within the Intel Nehalem processor family VenturiCore demonstrated the significant gains in throughput in moving from a single, non hyper-threaded Intel x5670 to dual hyper-threaded E7 processors – using the same installed software in each instance – processing different tasks in parallel to deliver superior overall application performance.
Where VenturiCore fits… Client Data Data Processing Client ResultsSources Export In From Import User Interface Memory Memory SQL OLEDB OLEDB Load Data Select XML ODBC Browser subsets File Custom ODBC Pre-Process Import DataWarehouse Data XLS Output Custom Custom Process data Application Txt Data VenturiCore
Use Cases Client Data Data Processing Client Results From Sources Export In Memory Memory SQL OLEDB Load Data Select XML subsets File ODBC Pre-Process Data Warehouse Data XLS Output data Txt Custom Process Data Big Data = Big MemoryOption calculations; VaR models; Co-location; JVM extender
About Attova• Partnership with Applied Cytometry to address move towards forward scaling software and parallel computing.• Attova Technology Services is primarily concerned with the introduction of market changing technologies to achieve transformational business success.• Offers high performance business consulting & technology services into financial markets• Knowledge Network- a powerhouse of efficient, experienced and expert level independent consultants and performance technologists• Compelling operating model – Multiple solution options for client- flexibility and choice – Ability to precisely align the resource & technology to the solution – Lower cost operating model which is passed through to client – Define, Design, Deliver- single port of call – Experience- know how and know when based on relevant client engagements
VenturiCoreExploiting the power of multi-core and dense memory x86 systems to implement a Grid-In-A-Box. Delivering Value at Risk-In-A-Box Contact : email@example.com Attova (+44 (0)20 7618 6100)