This white paper compares the performance of Fibre Channel, Hardware iSCSI, Software iSCSI, and NFS storage protocols in VMware vSphere 4. Experiments show that all four protocols can achieve maximum throughput limited only by network bandwidth. However, Fibre Channel and Hardware iSCSI have substantially lower CPU costs than Software iSCSI and NFS. Tests with multiple VMs also demonstrate that vSphere 4 maintains high performance levels with greater efficiency than previous versions.
Improving the Performance of the qcow2 Format (KVM Forum 2017)Igalia
By Alberto García.
qcow2 is QEMU's native file format for storing disk images. One of its features is that it grows dynamically, so disk space is only allocated when the virtual machine needs to store data. This makes the format efficient in terms of space requirements, but has an impact on its I/O performance. This presentation will describe some of those performance problems and will discuss possible ways to address them. Some of them can be solved by simply adjusting configuration parameters, others require improving the qcow2 driver in QEMU, and others need extending the file format itself.
(c) KVM Forum 2017
October 25 - 27, 2017
Hilton Prague, Prague, Czech Republic
http://events.linuxfoundation.org/events/archive/2017/kvm-forum
This document compares the performance of KVM and Docker instances in OpenStack. It finds that for boot and delete operations, KVM is around 20% faster than Docker at high concurrency levels. For CPU and memory tests at low concurrency, KVM is 2-10% slower than bare metal, while Docker is 2-5% slower. For sequential and random read/write file I/O tests, KVM is similar or slightly slower than bare metal, while Docker is slightly slower than KVM. Overall performance decreases for both KVM and Docker as concurrency increases.
The document discusses best practices for deploying MongoDB including sizing hardware with sufficient memory, CPU and I/O; using an appropriate operating system and filesystem; installing and upgrading MongoDB; ensuring durability with replication and backups; implementing security, monitoring performance with tools, and considerations for deploying on Amazon EC2.
This talk explores what has gone in so far in the Linux kernel (version 3.0 and 3.1) and which Linux distributions are deliverinbg Xen again. The otalk explores outstanding challenges and the pieces that are missing and what we can do, and what we cannot do working with Linux.
Kvm performance optimization for ubuntuSim Janghoon
This document discusses various techniques for optimizing KVM performance on Linux systems. It covers CPU and memory optimization through techniques like vCPU pinning, NUMA affinity, transparent huge pages, KSM, and virtio_balloon. For networking, it discusses vhost-net, interrupt handling using MSI/MSI-X, and NAPI. It also covers block device optimization through I/O scheduling, cache mode, and asynchronous I/O. The goal is to provide guidance on configuring these techniques for workloads running in KVM virtual machines.
The document discusses ioremap and mmap functions in Linux for mapping physical addresses into the virtual address space. Ioremap is used when physical addresses are larger than the virtual address space size. It maps physical addresses to virtual addresses that can be accessed by the CPU. Mmap allows a process to map pages of a file into virtual memory. It is useful for reducing memory copies and improving performance of file read/write operations. The document outlines the functions, flags, and flows of ioremap, mmap, and implementing a custom mmap file operation for direct physical memory mapping.
OSDC 2017 - Werner Fischer - Open power for the data centerNETWAYS
IBM's POWER (Performance Optimization With Enhanced RISC) architecture is known to run mission-critical applications and to provide bank-style "RAS" (Reliability, Availability, Serviceability) features since 1990. Opening the architecture in 2013 enabled other vendors like Tyan or Rackspace to build servers based on the current POWER8 edition of this architecture. The current POWER8 CPUs provide up to 12 cores with 8x Simultaneous Multithreading - leading to 96 threads per CPU. Up to eight memory channels enable up to 230 GB/s memory bandwidth per CPU. Increased L1, L2, L3 and new L4 caches help to boost the performance of memory-bound applications like databeses, by providing more than 1 TB/s of bandwidth. In this talk Werner will give an overview of the architecture and show the performance possibilities of POWER8, using the PostgreSQL database as an example. By comparing PostgreSQL 9.4, 9.5 and 9.6 benchmarking results he will visualize the increased efficiency thanks to PowergreSQL's optimizations for POWER over the last years. Finally, he will outline one other benefit of OpenPOWER systems: from the very beginning (the first instruction to initialize the first CPU core, long before DRAM, firmware management or PCIe works) up to running your Linux OS and application like a database, only open source code gets executed.
Improving the Performance of the qcow2 Format (KVM Forum 2017)Igalia
By Alberto García.
qcow2 is QEMU's native file format for storing disk images. One of its features is that it grows dynamically, so disk space is only allocated when the virtual machine needs to store data. This makes the format efficient in terms of space requirements, but has an impact on its I/O performance. This presentation will describe some of those performance problems and will discuss possible ways to address them. Some of them can be solved by simply adjusting configuration parameters, others require improving the qcow2 driver in QEMU, and others need extending the file format itself.
(c) KVM Forum 2017
October 25 - 27, 2017
Hilton Prague, Prague, Czech Republic
http://events.linuxfoundation.org/events/archive/2017/kvm-forum
This document compares the performance of KVM and Docker instances in OpenStack. It finds that for boot and delete operations, KVM is around 20% faster than Docker at high concurrency levels. For CPU and memory tests at low concurrency, KVM is 2-10% slower than bare metal, while Docker is 2-5% slower. For sequential and random read/write file I/O tests, KVM is similar or slightly slower than bare metal, while Docker is slightly slower than KVM. Overall performance decreases for both KVM and Docker as concurrency increases.
The document discusses best practices for deploying MongoDB including sizing hardware with sufficient memory, CPU and I/O; using an appropriate operating system and filesystem; installing and upgrading MongoDB; ensuring durability with replication and backups; implementing security, monitoring performance with tools, and considerations for deploying on Amazon EC2.
This talk explores what has gone in so far in the Linux kernel (version 3.0 and 3.1) and which Linux distributions are deliverinbg Xen again. The otalk explores outstanding challenges and the pieces that are missing and what we can do, and what we cannot do working with Linux.
Kvm performance optimization for ubuntuSim Janghoon
This document discusses various techniques for optimizing KVM performance on Linux systems. It covers CPU and memory optimization through techniques like vCPU pinning, NUMA affinity, transparent huge pages, KSM, and virtio_balloon. For networking, it discusses vhost-net, interrupt handling using MSI/MSI-X, and NAPI. It also covers block device optimization through I/O scheduling, cache mode, and asynchronous I/O. The goal is to provide guidance on configuring these techniques for workloads running in KVM virtual machines.
The document discusses ioremap and mmap functions in Linux for mapping physical addresses into the virtual address space. Ioremap is used when physical addresses are larger than the virtual address space size. It maps physical addresses to virtual addresses that can be accessed by the CPU. Mmap allows a process to map pages of a file into virtual memory. It is useful for reducing memory copies and improving performance of file read/write operations. The document outlines the functions, flags, and flows of ioremap, mmap, and implementing a custom mmap file operation for direct physical memory mapping.
OSDC 2017 - Werner Fischer - Open power for the data centerNETWAYS
IBM's POWER (Performance Optimization With Enhanced RISC) architecture is known to run mission-critical applications and to provide bank-style "RAS" (Reliability, Availability, Serviceability) features since 1990. Opening the architecture in 2013 enabled other vendors like Tyan or Rackspace to build servers based on the current POWER8 edition of this architecture. The current POWER8 CPUs provide up to 12 cores with 8x Simultaneous Multithreading - leading to 96 threads per CPU. Up to eight memory channels enable up to 230 GB/s memory bandwidth per CPU. Increased L1, L2, L3 and new L4 caches help to boost the performance of memory-bound applications like databeses, by providing more than 1 TB/s of bandwidth. In this talk Werner will give an overview of the architecture and show the performance possibilities of POWER8, using the PostgreSQL database as an example. By comparing PostgreSQL 9.4, 9.5 and 9.6 benchmarking results he will visualize the increased efficiency thanks to PowergreSQL's optimizations for POWER over the last years. Finally, he will outline one other benefit of OpenPOWER systems: from the very beginning (the first instruction to initialize the first CPU core, long before DRAM, firmware management or PCIe works) up to running your Linux OS and application like a database, only open source code gets executed.
QEMU is an emulator that uses dynamic translation to emulate one instruction set architecture (ISA) on another host ISA. It translates guest instructions to an intermediate representation (TCG IR) code, and then compiles the IR code to native host instructions. QEMU employs techniques like translation block caching and chaining to improve the performance of dynamic translation. It also uses helper functions to offload complex operations during translation to improve efficiency.
This document summarizes the Texas Advanced Computing Center's (TACC) experience using DDN's Infinite Memory Engine (IME) as a burst buffer for three HPC applications on the Stampede supercomputer. Initial testing showed I/O bottlenecks that were addressed by improving the InfiniBand topology. Performance testing found the IME provided significant acceleration over the Lustre parallel file system, with speedups ranging from 3.7x to 28x for the HACC cosmology code, 6.8x to 22.3x for the S3D combustion code, and 6.2x to 10.1x for the MADBench mini-app. The IME demonstrated its ability to scale and improve
This document discusses performance improvements to the Lustre parallel file system in versions 2.5 through large I/O patches, metadata improvements, and metadata scaling with distributed namespace (DNE). It summarizes evaluations showing improved throughput from 4MB RPC, reduced degradation with large numbers of threads using SSDs over NL-SAS, high random read performance from SSD pools, and significant metadata performance gains in Lustre 2.4 from DNE allowing nearly linear scaling. Key requirements for next-generation storage include extreme IOPS, tiered architectures using local flash with parallel file systems, and reducing infrastructure needs while maintaining throughput.
The document discusses tips for malloc and free in C, including making your own malloc library for troubleshooting. It covers system calls like brk/sbrk and mmap/munmap that are used to allocate memory in user space. It also provides tips for the glibc malloc implementation, such as functions like mallopt, malloc_stats, and malloc_usable_size. Finally, it discusses two methods for hooking and replacing malloc - using LD_PRELOAD and dlsym, or the __malloc_hook mechanism.
The document discusses storage concepts related to VMware including IP storage protocols like iSCSI and NFS, Fibre Channel SAN storage, VMFS datastores, and storage options for ESX/ESXi servers. It covers local storage, SAN storage using Fibre Channel and iSCSI, and NAS storage using NFS and CIFS. Key topics include VMFS specifications, space requirements for installing ESX, raw device mappings (RDM), virtual machine files, datastores, LUN mapping, multipathing policies, and the differences between Fibre Channel and iSCSI storage.
Presentation from 2016 Austin OpenStack Summit.
The Ceph upstream community is declaring CephFS stable for the first time in the recent Jewel release, but that declaration comes with caveats: while we have filesystem repair tools and a horizontally scalable POSIX filesystem, we have default-disabled exciting features like horizontally-scalable metadata servers and snapshots. This talk will present exactly what features you can expect to see, what's blocking the inclusion of other features, and what you as a user can expect and can contribute by deploying or testing CephFS.
- The document provides guidance on deploying MongoDB including sizing hardware, installing and upgrading MongoDB, configuration considerations for EC2, security, backups, durability, scaling out, and monitoring. Key aspects discussed are profiling and indexing queries for performance, allocating sufficient memory, CPU and disk I/O, using 64-bit OSes, ext4/XFS filesystems, upgrading to even version numbers, and replicating for high availability and backups.
The document describes a virtual device project that provides fast backup and replication of data using HDD RAID arrays. It logs I/O information synchronously to provide reliability even if the primary storage crashes. This asynchronous replication ensures the replica is always up to date without impacting the primary storage's performance. The system uses tag markers and a control file to backup data incrementally and support disaster recovery if needed.
As time goes on more OSes are getting Dom0 support, so there's a growing need to provide a platform independent set of tools from which to operate Xen. This talk will expose the different mechanisms used on NetBSD that diverge from the Linux approach, and how Xen is improving its userspace tools to provide a more platform independent support.
The talk also touches upon various features that BSD provides or plans to provide with Xen, thus presenting a coherent roadmap view of where we've come from, and what lies ahead.
What's in this talk:
Xen and BSD
Status updates from the world of BSD
Ecosystem/userbase
This document contains configuration maximums for virtual machines, storage, compute resources, memory, networking, and Virtual Center components in a VMware Infrastructure environment. It lists maximums such as 4 SCSI controllers and 60 devices per virtual machine, 2TB volume sizes, 128 virtual CPUs per server, 64GB RAM per server, 512 port groups, and 1500 virtual machines that can be managed by a single Virtual Center server. It also provides a high-level overview of the key components of VMware Infrastructure, including ESX Server, Virtual Center, and features such as VMotion, HA, and DRS.
The 7 Deadly Sins of Packet Processing - Venky Venkatesan and Bruce Richardsonharryvanhaaren
The document summarizes seven deadly sins of packet processing that can negatively impact performance:
1) Unpredictable branches that confuse the branch predictor. Code should guide the compiler on likely/unlikely cases.
2) Incorrect prefetching that pulls in unnecessary data or data needed by other cores, adding overhead. Hardware prefetchers often help but can also share cache lines inadvertently.
3) Per-packet operations like memory I/O and atomics that have overhead magnified at the per-packet level.
This document discusses Intel's high performance storage solution using Lustre file systems. It provides an overview of Lustre, how it can interface with various Intel technologies like SSDs, networking fabrics and processors. It also summarizes the key features of Lustre including its scalability, POSIX compatibility, shared namespace and how all clients can access data. Specific examples are given around using OpenZFS as the backend storage for Lustre and how technologies like L2ARC can improve performance. Monitoring and management tools for Lustre file systems are also highlighted.
KVM and docker LXC Benchmarking with OpenStackBoden Russell
Passive benchmarking with docker LXC and KVM using OpenStack hosted in SoftLayer. These results provide initial incite as to why LXC as a technology choice offers benefits over traditional VMs and seek to provide answers as to the typical initial LXC question -- "why would I consider Linux Containers over VMs" from a performance perspective.
Results here provide insight as to:
- Cloudy ops times (start, stop, reboot) using OpenStack.
- Guest micro benchmark performance (I/O, network, memory, CPU).
- Guest micro benchmark performance of MySQL; OLTP read, read / write complex and indexed insertion.
- Compute node resource consumption; VM / Container density factors.
- Lessons learned during benchmarking.
The tests here were performed using OpenStack Rally to drive the OpenStack cloudy tests and various other linux tools to test the guest performance on a "micro level". The nova docker virt driver was used in the Cloud scenario to realize VMs as docker LXC containers and compared to the nova virt driver for libvirt KVM.
Please read the disclaimers in the presentation as this is only intended to be the "chip of the ice burg".
Hadoop can effectively utilize many-core systems with large amounts of processing power and storage. The author tested a Hadoop cluster on a personal supercomputer with two nodes, each containing 48 cores, 256GB RAM, and 64TB of storage connected by 40Gb Infiniband. Testing showed the clustered configuration completed a 100GB Terasort in 241 seconds, significantly faster than comparable Amazon clusters. While Hadoop works well on a single fat node, distributing data and tasks across clustered nodes provides even better performance for large workloads.
Ceph Day Melbourne - Ceph on All-Flash Storage - Breaking Performance BarriersCeph Community
The document discusses a presentation about Ceph on all-flash storage using InfiniFlash systems to break performance barriers. It describes how Ceph has been optimized for flash storage and how InfiniFlash systems provide industry-leading performance of over 1 million IOPS and 6-9GB/s of throughput using SanDisk flash technology. The presentation also covers how InfiniFlash can provide scalable performance and capacity for large-scale enterprise workloads.
The document discusses accessing the virtual router in a CloudStack KVM environment. It explains that the virtual router can be accessed by SSHing into its link local IP address from the compute node it is running on. When logging in, it is shown that the link local IP address may change if the virtual router is restarted. The internal interfaces, routing tables, and network configuration of the running virtual router are then displayed and examined.
Design and implementation of a reliable and cost-effective cloud computing in...Francesco Taurino
This document summarizes the INFN Napoli experience in designing and implementing a reliable and cost-effective cloud computing infrastructure. Key aspects included using existing hardware, virtualization and clustering technologies to consolidate services and reduce costs. A network with redundant switches and storage servers using GlusterFS provided high availability. Custom tools were developed to simplify administration tasks like provisioning, migration, and load balancing of virtual machines. The solution provided an efficient and reliable private cloud with over one year of uninterrupted uptime.
LXC – NextGen Virtualization for Cloud benefit realization (cloudexpo)Boden Russell
This document summarizes a presentation on Linux containers (LXC) as an alternative to virtual machines for cloud computing and containerization. It finds that LXC provides significantly better performance than virtual machines in terms of provisioning time, CPU and memory usage, and load on the compute node. Specifically, when packing 15 active VMs/containers on a node, LXC uses 0.54% CPU on average compared to 7.64% for KVM, and 734 MB of memory total compared to 4,387 MB for KVM. When booting VMs/containers serially, the average boot time is 3.5 seconds for LXC versus 5.8 seconds for KVM, and CPU usage is lower overall for
For over 7 years, TYPO3 adoption in the province of Québec has been exponential. How did that happen and why?
The talk consisted of stories from the trenches, tips and tricks, and lessons learned.
Manuale Guida Breve alla Esdebitazione Fallimentare: procedura che consente a coloro che hanno subito un fallimento di chiedere ed ottenere, avendone i requisiti, la cancellazione di tutti i debiti residui al fallimento medesimo. In calce è riportata l'appendice normativa ed un facsimile di ricorso.
Guida Breve alla Riabilitazione Penale con il patrocinio a spese dello Stato: procedura che consente a coloro che sono stati condannati a seguito di sentenza penale passata in giudicato ovvero di decreto penale di condanna non opposto di chiedere ed ottenere, avendone i requisiti, la cancellazione dei reati dal casellario giudiziario e, conseguentemente, l'estinzione degli stessi.
Guida Breve all'avvio della mediazione civile (facoltativa) dopo la sentenza della Corte Costituzionale 272/2012: ecco come risolvere una controversia in tempi brevi e senza andare in tribunale! Con facsimile di domanda.
www.avvocatogratis.com
QEMU is an emulator that uses dynamic translation to emulate one instruction set architecture (ISA) on another host ISA. It translates guest instructions to an intermediate representation (TCG IR) code, and then compiles the IR code to native host instructions. QEMU employs techniques like translation block caching and chaining to improve the performance of dynamic translation. It also uses helper functions to offload complex operations during translation to improve efficiency.
This document summarizes the Texas Advanced Computing Center's (TACC) experience using DDN's Infinite Memory Engine (IME) as a burst buffer for three HPC applications on the Stampede supercomputer. Initial testing showed I/O bottlenecks that were addressed by improving the InfiniBand topology. Performance testing found the IME provided significant acceleration over the Lustre parallel file system, with speedups ranging from 3.7x to 28x for the HACC cosmology code, 6.8x to 22.3x for the S3D combustion code, and 6.2x to 10.1x for the MADBench mini-app. The IME demonstrated its ability to scale and improve
This document discusses performance improvements to the Lustre parallel file system in versions 2.5 through large I/O patches, metadata improvements, and metadata scaling with distributed namespace (DNE). It summarizes evaluations showing improved throughput from 4MB RPC, reduced degradation with large numbers of threads using SSDs over NL-SAS, high random read performance from SSD pools, and significant metadata performance gains in Lustre 2.4 from DNE allowing nearly linear scaling. Key requirements for next-generation storage include extreme IOPS, tiered architectures using local flash with parallel file systems, and reducing infrastructure needs while maintaining throughput.
The document discusses tips for malloc and free in C, including making your own malloc library for troubleshooting. It covers system calls like brk/sbrk and mmap/munmap that are used to allocate memory in user space. It also provides tips for the glibc malloc implementation, such as functions like mallopt, malloc_stats, and malloc_usable_size. Finally, it discusses two methods for hooking and replacing malloc - using LD_PRELOAD and dlsym, or the __malloc_hook mechanism.
The document discusses storage concepts related to VMware including IP storage protocols like iSCSI and NFS, Fibre Channel SAN storage, VMFS datastores, and storage options for ESX/ESXi servers. It covers local storage, SAN storage using Fibre Channel and iSCSI, and NAS storage using NFS and CIFS. Key topics include VMFS specifications, space requirements for installing ESX, raw device mappings (RDM), virtual machine files, datastores, LUN mapping, multipathing policies, and the differences between Fibre Channel and iSCSI storage.
Presentation from 2016 Austin OpenStack Summit.
The Ceph upstream community is declaring CephFS stable for the first time in the recent Jewel release, but that declaration comes with caveats: while we have filesystem repair tools and a horizontally scalable POSIX filesystem, we have default-disabled exciting features like horizontally-scalable metadata servers and snapshots. This talk will present exactly what features you can expect to see, what's blocking the inclusion of other features, and what you as a user can expect and can contribute by deploying or testing CephFS.
- The document provides guidance on deploying MongoDB including sizing hardware, installing and upgrading MongoDB, configuration considerations for EC2, security, backups, durability, scaling out, and monitoring. Key aspects discussed are profiling and indexing queries for performance, allocating sufficient memory, CPU and disk I/O, using 64-bit OSes, ext4/XFS filesystems, upgrading to even version numbers, and replicating for high availability and backups.
The document describes a virtual device project that provides fast backup and replication of data using HDD RAID arrays. It logs I/O information synchronously to provide reliability even if the primary storage crashes. This asynchronous replication ensures the replica is always up to date without impacting the primary storage's performance. The system uses tag markers and a control file to backup data incrementally and support disaster recovery if needed.
As time goes on more OSes are getting Dom0 support, so there's a growing need to provide a platform independent set of tools from which to operate Xen. This talk will expose the different mechanisms used on NetBSD that diverge from the Linux approach, and how Xen is improving its userspace tools to provide a more platform independent support.
The talk also touches upon various features that BSD provides or plans to provide with Xen, thus presenting a coherent roadmap view of where we've come from, and what lies ahead.
What's in this talk:
Xen and BSD
Status updates from the world of BSD
Ecosystem/userbase
This document contains configuration maximums for virtual machines, storage, compute resources, memory, networking, and Virtual Center components in a VMware Infrastructure environment. It lists maximums such as 4 SCSI controllers and 60 devices per virtual machine, 2TB volume sizes, 128 virtual CPUs per server, 64GB RAM per server, 512 port groups, and 1500 virtual machines that can be managed by a single Virtual Center server. It also provides a high-level overview of the key components of VMware Infrastructure, including ESX Server, Virtual Center, and features such as VMotion, HA, and DRS.
The 7 Deadly Sins of Packet Processing - Venky Venkatesan and Bruce Richardsonharryvanhaaren
The document summarizes seven deadly sins of packet processing that can negatively impact performance:
1) Unpredictable branches that confuse the branch predictor. Code should guide the compiler on likely/unlikely cases.
2) Incorrect prefetching that pulls in unnecessary data or data needed by other cores, adding overhead. Hardware prefetchers often help but can also share cache lines inadvertently.
3) Per-packet operations like memory I/O and atomics that have overhead magnified at the per-packet level.
This document discusses Intel's high performance storage solution using Lustre file systems. It provides an overview of Lustre, how it can interface with various Intel technologies like SSDs, networking fabrics and processors. It also summarizes the key features of Lustre including its scalability, POSIX compatibility, shared namespace and how all clients can access data. Specific examples are given around using OpenZFS as the backend storage for Lustre and how technologies like L2ARC can improve performance. Monitoring and management tools for Lustre file systems are also highlighted.
KVM and docker LXC Benchmarking with OpenStackBoden Russell
Passive benchmarking with docker LXC and KVM using OpenStack hosted in SoftLayer. These results provide initial incite as to why LXC as a technology choice offers benefits over traditional VMs and seek to provide answers as to the typical initial LXC question -- "why would I consider Linux Containers over VMs" from a performance perspective.
Results here provide insight as to:
- Cloudy ops times (start, stop, reboot) using OpenStack.
- Guest micro benchmark performance (I/O, network, memory, CPU).
- Guest micro benchmark performance of MySQL; OLTP read, read / write complex and indexed insertion.
- Compute node resource consumption; VM / Container density factors.
- Lessons learned during benchmarking.
The tests here were performed using OpenStack Rally to drive the OpenStack cloudy tests and various other linux tools to test the guest performance on a "micro level". The nova docker virt driver was used in the Cloud scenario to realize VMs as docker LXC containers and compared to the nova virt driver for libvirt KVM.
Please read the disclaimers in the presentation as this is only intended to be the "chip of the ice burg".
Hadoop can effectively utilize many-core systems with large amounts of processing power and storage. The author tested a Hadoop cluster on a personal supercomputer with two nodes, each containing 48 cores, 256GB RAM, and 64TB of storage connected by 40Gb Infiniband. Testing showed the clustered configuration completed a 100GB Terasort in 241 seconds, significantly faster than comparable Amazon clusters. While Hadoop works well on a single fat node, distributing data and tasks across clustered nodes provides even better performance for large workloads.
Ceph Day Melbourne - Ceph on All-Flash Storage - Breaking Performance BarriersCeph Community
The document discusses a presentation about Ceph on all-flash storage using InfiniFlash systems to break performance barriers. It describes how Ceph has been optimized for flash storage and how InfiniFlash systems provide industry-leading performance of over 1 million IOPS and 6-9GB/s of throughput using SanDisk flash technology. The presentation also covers how InfiniFlash can provide scalable performance and capacity for large-scale enterprise workloads.
The document discusses accessing the virtual router in a CloudStack KVM environment. It explains that the virtual router can be accessed by SSHing into its link local IP address from the compute node it is running on. When logging in, it is shown that the link local IP address may change if the virtual router is restarted. The internal interfaces, routing tables, and network configuration of the running virtual router are then displayed and examined.
Design and implementation of a reliable and cost-effective cloud computing in...Francesco Taurino
This document summarizes the INFN Napoli experience in designing and implementing a reliable and cost-effective cloud computing infrastructure. Key aspects included using existing hardware, virtualization and clustering technologies to consolidate services and reduce costs. A network with redundant switches and storage servers using GlusterFS provided high availability. Custom tools were developed to simplify administration tasks like provisioning, migration, and load balancing of virtual machines. The solution provided an efficient and reliable private cloud with over one year of uninterrupted uptime.
LXC – NextGen Virtualization for Cloud benefit realization (cloudexpo)Boden Russell
This document summarizes a presentation on Linux containers (LXC) as an alternative to virtual machines for cloud computing and containerization. It finds that LXC provides significantly better performance than virtual machines in terms of provisioning time, CPU and memory usage, and load on the compute node. Specifically, when packing 15 active VMs/containers on a node, LXC uses 0.54% CPU on average compared to 7.64% for KVM, and 734 MB of memory total compared to 4,387 MB for KVM. When booting VMs/containers serially, the average boot time is 3.5 seconds for LXC versus 5.8 seconds for KVM, and CPU usage is lower overall for
For over 7 years, TYPO3 adoption in the province of Québec has been exponential. How did that happen and why?
The talk consisted of stories from the trenches, tips and tricks, and lessons learned.
Manuale Guida Breve alla Esdebitazione Fallimentare: procedura che consente a coloro che hanno subito un fallimento di chiedere ed ottenere, avendone i requisiti, la cancellazione di tutti i debiti residui al fallimento medesimo. In calce è riportata l'appendice normativa ed un facsimile di ricorso.
Guida Breve alla Riabilitazione Penale con il patrocinio a spese dello Stato: procedura che consente a coloro che sono stati condannati a seguito di sentenza penale passata in giudicato ovvero di decreto penale di condanna non opposto di chiedere ed ottenere, avendone i requisiti, la cancellazione dei reati dal casellario giudiziario e, conseguentemente, l'estinzione degli stessi.
Guida Breve all'avvio della mediazione civile (facoltativa) dopo la sentenza della Corte Costituzionale 272/2012: ecco come risolvere una controversia in tempi brevi e senza andare in tribunale! Con facsimile di domanda.
www.avvocatogratis.com
Questo è un prontuario utile per attivare la tua difesa tecnica nel corso della separazione dei coniugi e per orientarti velocemente in una materia complessa, ricca di prescrizioni normative, giurisprudenziali e prassi. Una guida breve per capire i principi fondamentali della separazione dei coniugi (consensuale e giudiziale).
The document discusses the 5 themes of geography as defined by the National Geographic Society: place, location, human-environment interaction, region, and movement. It provides details on each theme, including examples and key things to consider for each one. For place, it discusses physical and human characteristics that make a place unique. For region, it outlines different ways regions can be defined. For human-environment interaction, it explains how people adapt to and modify their environments. For location, it distinguishes between specific and general locations. And for movement, it discusses the flow of people, goods, and ideas between places.
Non ti pagano lo stipendio? Qui puoi scaricare gratis la Guida al Recupero Crediti da lavoro con tutte le istruzioni su cosa fare, su come ottenere l'ammissione al gratuito patrocinio e con le 22 risposte alle domande che si fanno sempre all'avvocato.
Leggila con attenzione e contatta subito un avvocato di fiducia per attivare la tua difesa senza perdere tempo ed opportunità.
COME FUNZIONANO LE DIMISSIONI PER GIUSTA CAUSA E QUANDO BENEFICIARNEI: istruzioni per l'autodifesa retributiva: tutto quello che serve sapere per usare le dimissioni per giusta causa contro mobbing, insoluti retributivi, molestie sessuali etc. anche con l'ausilio del gratuito patrocinio.
Le generazioni invecchiano e i problemi sociali aumentano.
Scopriamo uno degli strumenti idonei per affrontare questi cambiamenti con le 13 domande per capire come funziona l'Amministrazione di Sostegno.
Guida Breve all'AMMINISTRAZIONE DI SOSTEGNO con il Gratuito Patrocinio: e-book gratis con istruzioni normative, suggerimenti, prassi e tutto quanto serve scoprire sull'amminsitartore di sostegno per gestirlo con il gratuito patrocinio senza perdere tempo e non commettere errori.
Guida Breve con una sintesi tecnica della "PROCEDURA DI GESTIONE DELLA CRISI PER I SOGGETTI NON FALLIBILI IN DIFFICOLTÀ", piccole imprese e consumatori, la cui situazione debitoria sia divenuta oramai cronica e non più superabile con le risorse ordinarie disponibili. Con appendice normativa.
Rainbird: Realtime Analytics at Twitter (Strata 2011)Kevin Weil
Introducing Rainbird, Twitter's high volume distributed counting service for realtime analytics, built on Cassandra. This presentation looks at the motivation, design, and uses of Rainbird across Twitter.
Club E-Tourisme // Les réseaux sociaux, des outils au service de la promotion...Pays Médoc
Club E-Tourisme réalisé avec et pour les prestataires touristiques de Pauillac le 26 avril 2011 sur le thème des réseaux sociaux, pour sensibiliser aux outils des réseaux sociaux que l'on peut mobiliser pour la promotion de ses prestations touristiques.
Veille à suivre sur twitter @mon_etourisme
Hadoop, Pig, and Twitter (NoSQL East 2009)Kevin Weil
A talk on the use of Hadoop and Pig inside Twitter, focusing on the flexibility and simplicity of Pig, and the benefits of that for solving real-world big data problems.
Colapso cardiovascular, paro cardiaco y muerte súbita (final)brielle
Este documento diferencia entre colapso cardiovascular, paro cardiaco y muerte súbita cardiaca. Explica que el colapso cardiovascular es un término inespecífico que puede mostrar reversión espontánea o necesitar intervención, mientras que el paro cardiaco requiere intervención inmediata para revertirlo antes de que cause la muerte. La muerte súbita cardiaca es una interrupción repentina e irreversible de todas las funciones biológicas. Además, describe los factores de riesgo, mecanismos y tratamiento de estos tres conceptos.
View the performance metrics that turned the heads of VMware, EMC, and NetApp at VMworld 2011.
See the reason why Nexenta is now the single biggest threat to legacy storage.
Технологии работы с дисковыми хранилищами и файловыми системами Windows Serve...Виталий Стародубцев
##Что такое Storage Replica
##Архитектура и сценарии
##Синхронная и асинхронная репликация
##Междисковая, межсерверная, внутрикластерная и межкластерная репликация
##Дизайн и проектирование Storage Replica
##Нововведения в Windows Server 2016 TP5
##Графический интерфейс управления, и другие возможности - демонстрация и планы развития
##Интеграция Storage Replica с Storage Spaces Direct
The document discusses accelerating Ceph storage performance using SPDK. SPDK introduces optimizations like asynchronous APIs, userspace I/O stacks, and polling mode drivers to reduce software overhead and better utilize fast storage devices. This allows Ceph to better support high performance networks and storage like NVMe SSDs. The document provides an example where SPDK helped XSKY's BlueStore object store achieve significant performance gains over the standard Ceph implementation.
This document provides an overview of troubleshooting storage performance issues in vSphere environments. It discusses using vCenter performance charts and ESXTop to analyze latency and I/O statistics at the storage path, disk, and LUN level. The document also covers topics like disk alignment, considerations for using SCSI versus SATA disks, identifying APD issues, multipathing, and how VMware uses SCSI reservations for metadata locking on shared VMFS datastores.
CTX138217 - IntelliCache Reduction in IOPS: XenDesktop 5.6 FP1 on XenServer 6.1 - Citrix Knowledge Center http://ow.ly/o3Ma4
The purpose of this document is to provide testing results based on MCS-delivered streamed virtual desktops leveraging IntelliCache
Benchmark results for running bioinformatics platform Galaxy on the Amazon Web Services cloud. Results include info about disks, instance types, sizes, and variable data size.
Amazon EC2 provides a broad selection of instance types to deliver high performance for a diverse mix of applications. In this session, we overview the drivers of system performance and discuss in depth how Amazon EC2 instances deliver system performance while also providing elasticity and complete control over your infrastructure. We also detail best practices and share performance tips for getting the most out of your Amazon EC2 instances.
The document compares several open source virtualization and container technologies. It describes experiments conducted to measure the performance and scalability of KVM, Xen, Linux-VServer, OpenVZ, VirtualBox and KQEMU. The experiments involved benchmarking the technologies using various workloads to test aspects like overhead, throughput, and ability to run multiple virtual machines concurrently. Linux-VServer showed the best overall performance, while Xen performed well except on I/O-bound workloads. The conclusions recommend using OpenVZ for network applications, Linux-VServer for general use, and KVM or VirtualBox for development environments.
The document discusses PROSE (Partitioned Reliable Operating System Environment), an approach that runs applications in specialized kernel partitions for finer control over system resources and improved reliability. It aims to simplify development of specialized kernels and enable resource sharing across partitions. The approach is evaluated using IBM's research hypervisor rHype, which shows PROSE can reduce noise and provide more deterministic performance than Linux. Future work focuses on running larger commercial workloads and further performance/noise experiments.
Real-Time Load Balancing of an Interactive Mutliplayer Game ServerJames Munro
This document summarizes research on implementing real-time load balancing in the game QuakeWorld. The researchers tested four load balancing algorithms (LPTF, SPTF, RR, SRR) across multiple servers and clients. Their initial results showed that LPTF provided the best workload distribution and server throughput, while reducing intra-frame wait times compared to the other algorithms. The researchers concluded that effective load balancing requires optimizing multiple performance metrics to balance workloads and minimize waiting.
Choosing the Right EC2 Instance and Applicable Use Cases - AWS June 2016 Webi...Amazon Web Services
Amazon Elastic Compute Cloud (Amazon EC2) provides a broad selection of instance types to accommodate a diverse mix of workloads. In this technical session, we provide an overview of the Amazon EC2 instance platform, key platform features, and the concept of instance generations. We dive into the design choices of the different instance families, including the General Purpose, Compute Optimized, Storage Optimized, and Memory Optimized families. We also detail best practices and share performance tips for getting the most out of your Amazon EC2 instances.
Learning Objectives: • Understand the differences between instances • Learn best practices and tips for getting the most out of EC2 instances
Yesterday's thinking may still believe NVMe (NVM Express) is in transition to a production ready solution. In this session, we will discuss how the evolution of NVMe is ready for production, the history and evolution of NVMe and the Linux stack to address where NVMe has progressed today to become the low latency, highly reliable database key value store mechanism that will drive the future of cloud expansion. Examples of protocol efficiencies and types of storage engines that are optimizing for NVMe will be discussed. Please join us for an exciting session where in-memory computing and persistence have evolved.
This document provides a summary of the maximum configuration limits for various VMware ESX and vSphere versions. It includes limits for virtual machines, storage, networking, clusters, vCenter Server and other areas. The limits generally increase from older ESX versions to newer vSphere versions. For exact specifications and the latest information, the document directs readers to VMware's documentation on configuration maximums.
The document describes COSBench, a benchmark tool for evaluating the performance of cloud object storage services. It provides an overview of COSBench's key components, including its configurable workload definition file, controller for managing tests, and drivers for generating load. The document also shares sample results from using COSBench to measure the throughput and response times of OpenStack Swift in different configurations. It found that the proxy node's CPU was the bottleneck for larger workloads on one setup. The goal is to open source COSBench to help storage providers optimize performance.
The Lenovo Storage N4610 is a Windows Storage Server NAS appliance that provides up to 72TB of storage capacity. It utilizes the Intel Grantley platform and ThinkServer technologies. Testing showed the N4610 delivered high I/O performance across various access patterns when connected via CIFS and iSCSI, with sequential performance increasing as block size decreased. The N4610 brings strong storage performance for business server infrastructures.
Hands-on Lab: How to Unleash Your Storage Performance by Using NVM Express™ B...Odinot Stanislas
(FR)
Voici un excellent document qui explique étape après étape comment installer, monitorer et surtout correctement benchmarker ses SSD PCIe/NVMe (pas si simple que ça). Autre élément clé : comment analyser la charge I/O de véritables applications? Combien d'IOPS, en read, en write, quelle bande passante et surtout quel impact sur la durée de vie des SSD? Bref à mettre en toute les mains, et un merci à mon collègue Andrey Kudryavtsev.
(EN)
An excellent content which describe step by step how to install, monitor and benchmark PCIe/NVMe SSD (many trick not so simple). Another key learning: how to measure real I/O activities on a real workload? How many R/W IOPS, block size, throughtput, and finally what's the impact on SSD endurance and (real)life? A must read, and a huge thanks to my colleague Andrey Kudryavtsev.
Auteurs/Authors:
Andrey Kudryavtsev, SSD Solution Architect, Intel Corporation
Zhdan Bybin, Application Engineer, Intel Corporation
The document summarizes key topics and industry talks from the China Linux Summit Forum (CLSF) 2010 conference in Shanghai. It discusses presentations on writeback optimization, the BTRFS file system, SSD challenges, VFS scalability, kernel testing frameworks, and talks from companies like Intel, EMC, Taobao, and Baidu on their storage architectures and solutions. Attendees included representatives from Intel, EMC, Fujitsu, Taobao, Novell, Oracle, Baidu, and Canonical discussing topics around file systems, storage, and kernel optimizations.
This document discusses optimizing MySQL performance when deployed in cloud environments. Some key points:
1) Deploying MySQL in the cloud presents new challenges like increased IO latency from network storage, but also opportunities to scale horizontally across multiple servers.
2) InnoDB performance is particularly important to optimize, through techniques like reducing mutex contention, increasing IO throughput, and using patches that improve handling of higher latency storage.
3) Benchmarks show that direct attached storage outperforms network storage, and optimizations like IO threads and capacity tuning can help mitigate higher latencies of network storage. Virtualization also incurs some performance overhead.
This document provides an overview and best practices for running Microsoft Exchange 2010 in a virtualized environment using VMware vSphere.
Key points include:
- Performance testing shows Exchange 2010 performs within 5% of physical hardware when virtualized. Storage protocol performance is comparable between Fibre Channel, iSCSI, and NFS.
- Enabling features like DRS and VMotion can increase performance by up to 18% by load balancing VMs across hosts.
- Best practices include proper sizing of virtual memory, using shared storage, multipathing, and dedicating sufficient resources to Exchange VMs.
GraphRAG for Life Science to increase LLM accuracyTomaz Bratanic
GraphRAG for life science domain, where you retriever information from biomedical knowledge graphs using LLMs to increase the accuracy and performance of generated answers
HCL Notes und Domino Lizenzkostenreduzierung in der Welt von DLAUpanagenda
Webinar Recording: https://www.panagenda.com/webinars/hcl-notes-und-domino-lizenzkostenreduzierung-in-der-welt-von-dlau/
DLAU und die Lizenzen nach dem CCB- und CCX-Modell sind für viele in der HCL-Community seit letztem Jahr ein heißes Thema. Als Notes- oder Domino-Kunde haben Sie vielleicht mit unerwartet hohen Benutzerzahlen und Lizenzgebühren zu kämpfen. Sie fragen sich vielleicht, wie diese neue Art der Lizenzierung funktioniert und welchen Nutzen sie Ihnen bringt. Vor allem wollen Sie sicherlich Ihr Budget einhalten und Kosten sparen, wo immer möglich. Das verstehen wir und wir möchten Ihnen dabei helfen!
Wir erklären Ihnen, wie Sie häufige Konfigurationsprobleme lösen können, die dazu führen können, dass mehr Benutzer gezählt werden als nötig, und wie Sie überflüssige oder ungenutzte Konten identifizieren und entfernen können, um Geld zu sparen. Es gibt auch einige Ansätze, die zu unnötigen Ausgaben führen können, z. B. wenn ein Personendokument anstelle eines Mail-Ins für geteilte Mailboxen verwendet wird. Wir zeigen Ihnen solche Fälle und deren Lösungen. Und natürlich erklären wir Ihnen das neue Lizenzmodell.
Nehmen Sie an diesem Webinar teil, bei dem HCL-Ambassador Marc Thomas und Gastredner Franz Walder Ihnen diese neue Welt näherbringen. Es vermittelt Ihnen die Tools und das Know-how, um den Überblick zu bewahren. Sie werden in der Lage sein, Ihre Kosten durch eine optimierte Domino-Konfiguration zu reduzieren und auch in Zukunft gering zu halten.
Diese Themen werden behandelt
- Reduzierung der Lizenzkosten durch Auffinden und Beheben von Fehlkonfigurationen und überflüssigen Konten
- Wie funktionieren CCB- und CCX-Lizenzen wirklich?
- Verstehen des DLAU-Tools und wie man es am besten nutzt
- Tipps für häufige Problembereiche, wie z. B. Team-Postfächer, Funktions-/Testbenutzer usw.
- Praxisbeispiele und Best Practices zum sofortigen Umsetzen
TrustArc Webinar - 2024 Global Privacy SurveyTrustArc
How does your privacy program stack up against your peers? What challenges are privacy teams tackling and prioritizing in 2024?
In the fifth annual Global Privacy Benchmarks Survey, we asked over 1,800 global privacy professionals and business executives to share their perspectives on the current state of privacy inside and outside of their organizations. This year’s report focused on emerging areas of importance for privacy and compliance professionals, including considerations and implications of Artificial Intelligence (AI) technologies, building brand trust, and different approaches for achieving higher privacy competence scores.
See how organizational priorities and strategic approaches to data security and privacy are evolving around the globe.
This webinar will review:
- The top 10 privacy insights from the fifth annual Global Privacy Benchmarks Survey
- The top challenges for privacy leaders, practitioners, and organizations in 2024
- Key themes to consider in developing and maintaining your privacy program
Cosa hanno in comune un mattoncino Lego e la backdoor XZ?Speck&Tech
ABSTRACT: A prima vista, un mattoncino Lego e la backdoor XZ potrebbero avere in comune il fatto di essere entrambi blocchi di costruzione, o dipendenze di progetti creativi e software. La realtà è che un mattoncino Lego e il caso della backdoor XZ hanno molto di più di tutto ciò in comune.
Partecipate alla presentazione per immergervi in una storia di interoperabilità, standard e formati aperti, per poi discutere del ruolo importante che i contributori hanno in una comunità open source sostenibile.
BIO: Sostenitrice del software libero e dei formati standard e aperti. È stata un membro attivo dei progetti Fedora e openSUSE e ha co-fondato l'Associazione LibreItalia dove è stata coinvolta in diversi eventi, migrazioni e formazione relativi a LibreOffice. In precedenza ha lavorato a migrazioni e corsi di formazione su LibreOffice per diverse amministrazioni pubbliche e privati. Da gennaio 2020 lavora in SUSE come Software Release Engineer per Uyuni e SUSE Manager e quando non segue la sua passione per i computer e per Geeko coltiva la sua curiosità per l'astronomia (da cui deriva il suo nickname deneb_alpha).
Salesforce Integration for Bonterra Impact Management (fka Social Solutions A...Jeffrey Haguewood
Sidekick Solutions uses Bonterra Impact Management (fka Social Solutions Apricot) and automation solutions to integrate data for business workflows.
We believe integration and automation are essential to user experience and the promise of efficient work through technology. Automation is the critical ingredient to realizing that full vision. We develop integration products and services for Bonterra Case Management software to support the deployment of automations for a variety of use cases.
This video focuses on integration of Salesforce with Bonterra Impact Management.
Interested in deploying an integration with Salesforce for Bonterra Impact Management? Contact us at sales@sidekicksolutionsllc.com to discuss next steps.
AI 101: An Introduction to the Basics and Impact of Artificial IntelligenceIndexBug
Imagine a world where machines not only perform tasks but also learn, adapt, and make decisions. This is the promise of Artificial Intelligence (AI), a technology that's not just enhancing our lives but revolutionizing entire industries.
Skybuffer SAM4U tool for SAP license adoptionTatiana Kojar
Manage and optimize your license adoption and consumption with SAM4U, an SAP free customer software asset management tool.
SAM4U, an SAP complimentary software asset management tool for customers, delivers a detailed and well-structured overview of license inventory and usage with a user-friendly interface. We offer a hosted, cost-effective, and performance-optimized SAM4U setup in the Skybuffer Cloud environment. You retain ownership of the system and data, while we manage the ABAP 7.58 infrastructure, ensuring fixed Total Cost of Ownership (TCO) and exceptional services through the SAP Fiori interface.
Fueling AI with Great Data with Airbyte WebinarZilliz
This talk will focus on how to collect data from a variety of sources, leveraging this data for RAG and other GenAI use cases, and finally charting your course to productionalization.
Have you ever been confused by the myriad of choices offered by AWS for hosting a website or an API?
Lambda, Elastic Beanstalk, Lightsail, Amplify, S3 (and more!) can each host websites + APIs. But which one should we choose?
Which one is cheapest? Which one is fastest? Which one will scale to meet our needs?
Join me in this session as we dive into each AWS hosting service to determine which one is best for your scenario and explain why!
Main news related to the CCS TSI 2023 (2023/1695)Jakub Marek
An English 🇬🇧 translation of a presentation to the speech I gave about the main changes brought by CCS TSI 2023 at the biggest Czech conference on Communications and signalling systems on Railways, which was held in Clarion Hotel Olomouc from 7th to 9th November 2023 (konferenceszt.cz). Attended by around 500 participants and 200 on-line followers.
The original Czech 🇨🇿 version of the presentation can be found here: https://www.slideshare.net/slideshow/hlavni-novinky-souvisejici-s-ccs-tsi-2023-2023-1695/269688092 .
The videorecording (in Czech) from the presentation is available here: https://youtu.be/WzjJWm4IyPk?si=SImb06tuXGb30BEH .
OpenID AuthZEN Interop Read Out - AuthorizationDavid Brossard
During Identiverse 2024 and EIC 2024, members of the OpenID AuthZEN WG got together and demoed their authorization endpoints conforming to the AuthZEN API
3. VMware white paper
Introduction
This paper compares the performance of various storage protocols available on VMware vSphere™ 4. The protocols Fibre Channel,
Hardware iSCSI, Software iSCSI, and NFS are tested using virtual machines on an ESX 4.0 host. Iometer is used to generate the I/O workload.
The Fibre Channel experiments were conducted over a 4Gb Fibre Channel network. The Hardware iSCSI, Software iSCSI, and NFS
experiments were conducted over a Gigabit Ethernet connection. Experiments over 10Gb Ethernet and 8Gb Fibre Channel will be
included in a future update to this paper.
This paper will start by describing key aspects of the test environment: ESX host, storage array, virtual machines, and Iometer workload.
Next, performance results for throughput and CPU cost are presented from experiments involving one or more virtual machines.
Finally, the key findings of the experiments are summarized.
The terms “storage server” and “storage array” will be used interchangeably in this paper.
executive Summary
The experiments in this paper show that each of the four storage protocols (Fibre Channel, Hardware iSCSI, Software iSCSI, and NFS)
can achieve line-rate throughput for both single virtual machine and multiple virtual machines on an ESX host. These experiments
also show that Fibre Channel and Hardware iSCSI have substantially lower CPU cost than Software iSCSI and NFS.
experimental Setup
The table below shows key aspects of the test environment for the ESX host, the storage array, and the virtual machine.
ESX Host
Component Details
hypervisor VMware eSX 4.0
processors Four intel Xeon e7340 Quad-Core 2.4Ghz
processors
Memory 32GB
Fibre Channel hBa QLogic QLa2432 4Gb
Fibre Channel network 4Gb FC switch
NiC for NFS and Sw iSCSi 1Gb (intel 82571eB)
MtU for NFS, Sw iSCSi, hw iSCSi 1500 bytes
iSCSi hBa QLogic QL4062c 1Gb (Firmware: 3.0.1.49)
ip network for NFS and Sw/hw iSCSi 1Gb ethernet with dedicated switch and
VLaN (extreme Summit 400-48t)
File system for NFS Native file system on NFS server
File system for FC and Sw/hw iSCSi None (rDM-physical was used)
Storage Array
Component Details
Storage server One server supporting FC, iSCSi, and NFS
Disk Drives: Number per data LUN 9
Disk Drives: Size 300Gb
Disk Drives: Speed 15K rpM
Disk Drives: type Fibre Channel
3
4. VMware white paper
Virtual Machine
Component Details
Guest OS windows Server 2008 enterprise Sp1
Virtual processors 1
Memory 512MB
Virtual disk for data 100MB Mapped raw LUN (rDM-physical)
File system None (physical drives were used)
SCSi controller LSi Logic parallel
VMware’s VMFS file system is recommended for production deployments of virtual machines on iSCSI and Fibre Channel arrays.
Because NFS storage presents files and not blocks, VMFS is not needed or possible. VMFS was therefore not used in the Fibre
Channel and iSCSI experiments to attempt to produce results that could be compared across all protocols.
I/O workload
Iometer (http://sourceforge.net/projects/iometer) was used to generate the I/O workload for these experiments. Iometer is a free
storage performance testing tool that can be configured to measure throughput and latency under a wide variety of access profiles.
Iometer Workload
Component Details
Number of outstanding i/Os 16
run time 2 min
ramp-up time 2 min
Number of workers 1 (per VM)
Each virtual (data) disk of the virtual machines used in these experiments is 100MB in size. The small size of these virtual disks ensures
that the I/O working set will fit into the cache of the storage array. An experiment with a working set size that fits into the cache of
the storage array is commonly referred to as a cached run.
For read operations in a cached run experiment, the data is served from the storage array’s cache, and read performance is independent
of disk latencies.
For write operations in a cached run experiment, the rate of write requests at the storage array may exceed the storage array’s rate of
writing the dirty blocks from the write cache to disk. If this happens, the write cache will eventually fill up. Once the write cache is full,
write performance is limited by the rate at which dirty blocks in the write cache are written to disk. This rate is limited by the latency
of the disks in the storage array, the RAID configuration, and the number of disk spindles used for the LUN.
For these reasons, read performance for cached runs is a better indication of the true performance of a storage protocol on the ESX
host, irrespective of the storage array used.
4
5. VMware white paper
experiment: Single VM: Throughput (read and write)
Figure 1 shows the sequential read throughput (in MB/sec) of running a single virtual machine in the standard workload configuration
for different I/O block sizes, for each of the storage protocols.
Figure 1: Read throughput for different I/O block sizes
450
400
Throughput (MB/sec)
350
300
250
200
150
100
50
0
1KB 4KB 8KB 16KB 32KB 64KB 128KB 256KB 512KB
I/O Block Sizes NFS SW iSCSI HW iSCSI FC
For Fibre Channel, read throughput is limited by the bandwidth of the 4Gb Fibre Channel link for I/O sizes at or above 64KB. For IP-based
protocols, read throughput is limited by the bandwidth of the 1Gb Ethernet link for I/O sizes at or above 32KB.
Figure 2 shows the sequential write throughput (in MB/sec) of running a single virtual machine in the standard workload configuration
for different I/O block sizes, for each of the storage protocols.
Figure 2: Write throughput for different I/O block sizes
450
400
Throughput (MB/sec)
350
300
250
200
150
100
50
0
1KB 4KB 8KB 16KB 32KB 64KB 128KB 256KB 512KB
I/O Block Sizes NFS SW iSCSI HW iSCSI FC
5
6. VMware white paper
For Fibre Channel, the maximum write throughput for any I/O block size is consistently lower than read throughput of the same I/O
block size. This is the result of disk write bandwidth limitations on the storage array. For the IP-based protocols, write throughput for
block sizes at or above 16KB is limited by the bandwidth of the 1Gb Ethernet link.
To summarize, a single Iometer thread running in a virtual machine can saturate the bandwidth of the respective networks for all four
storage protocols, for both read and write. Fibre Channel throughput performance is higher because of the higher bandwidth of the
Fibre Channel link. For the IP-based protocols, there is no significant throughput difference for most block sizes.
experiment: Single VM: CPU Cost Per I/O (read)
CPU cost is a measure of the amount of CPU resources used by ESX to perform a given amount of I/O. In this paper, the CPU cost of
each storage protocol is measured in units of CPU cycles per I/O operation. The cost for different storage protocols is normalized with
respect to the cost of software iSCSI on ESX 3.5.
Figure 3 shows the relative CPU cost of sequential reads in a single virtual machine in the standard workload configuration for a block
size of 64 KB for each of the storage protocols. Results on ESX 4.0 are shown next to ESX 3.5 to highlight efficiency improvements on
all protocols. The CPU cost of write operations for different storage protocols was not compared as write performance is strongly
dependent on the choice of the storage array.
Figure 3: Relative CPU cost of 64 KB sequential reads in a single virtual machine
1.2
1.0
Relative CPU Cost Per I/O
0.8
0.6
0.4
0.2
0
NFS SW iSCSI HW iSCSI FC
ESX 3.5 ESX 4.0
For Fibre Channel and Hardware iSCSI, a major part of the protocol processing is offloaded to the HBA, and consequently the cost
of each I/O is very low. For Software iSCSI and NFS, host CPUs are used for protocol processing which increases cost. Furthermore,
the cost of NFS and Software iSCSI is higher with larger block sizes, such as 64 KB. This is due to the additional CPU cycles needed
for each block for check summing, blocking, etc. Software iSCSI and NFS are more efficient at smaller blocks and are both capable
of delivering high throughput performance when CPU resource is not a bottleneck, as will be shown in the next section.
The cost per I/O is dependent on a variety of test parameters, such as platform architecture, block size, and other factors.
However, these tests demonstrate improved efficiency in vSphere 4’s storage stack over the previous version.
6
7. VMware white paper
experiment: Multiple VMs: aggregate Throughput (read)
Figure 4 shows the aggregate sequential read throughput (in MB/sec) of running 2, 4, 8, 16, and 32 virtual machines in the standard
workload configuration for a block size of 64KB. Each virtual machine performs I/O to its dedicated 100MB LUN.
Figure 4: Throughput of running multiple VMs in the standard workload configuration
450
400
Throughput (MB/sec)
350
300
250
200
150
100
50
0
2 4 8 16 32
Number of VMs NFS SW iSCSI HW iSCSI FC
For each of the storage protocols, the maximum aggregate throughput is limited by the network bandwidth. There is no degradation
in aggregate throughput for a large number of virtual machines.
Conclusion
In this paper, the performance of four storage protocols was compared for accessing shared storage available on VMware ESX 4.0:
Fibre Channel, Hardware iSCSI, Software iSCSI, and NFS.
All four storage protocols for shared storage on ESX are shown to be capable of achieving throughput levels that are only limited by the
capabilities of the storage array and the connection between it and the ESX server. ESX shows excellent scalability by maintaining these
performance levels in cases of heavy consolidation. For CPU cost, Fibre Channel and Hardware iSCSI are more efficient than Software iSCSI
and NFS. However, when CPU resources are not a bottleneck, Software iSCSI and NFS can also be part of a high-performance solution.
Earlier versions of ESX were also able to achieve throughput levels that are only limited by the array and bandwidth to it. VMware vSphere 4
continues to support this maximized throughput but can do so with greater efficiency. Improved efficiency on vSphere 4 means the
same high levels of performance with more virtual machines.
7