what’s new in vmware vsphere 5.1 – performance€¦ · · 2016-11-24the web client also...

What’s New in VMware vSphere® 5.1 – PerformanceVMware vSphere 5.1

T E C H N I C A L M A R K E T I N G D O C U M E N TAT I O NV 1 . 0 /A U G U S T 2 0 1 2

What’s New in VMware vSphere 5.1 – Performance

T E C H N I C A L W H I T E P A P E R / 2

Table of Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

vCenter Server Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

vSphere Web Client . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4

Single Sign-On Server . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5

vCenter Server Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5

Core Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

64-Way vCPU Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5

New Physical Hardware Support and Scalability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6

Memory Overhead Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6

Virtualized CPU Performance Counters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6

Latency Sensitivity Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

Storage I/O Control – Automatically Tuning Congestion Threshold . . . . . . . . . . . . . . . .8

VmObservedLatency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8

Storage DRS Enhancements – Correlation Improvements . . . . . . . . . . . . . . . . . . . . . . . . .9

16Gb FC Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

iSCSI Storage Driver Upgrade for Jumbo Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11

Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

VDS Scalability Improvements and Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Single-Root I/O Virtualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

vMotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

vMotion Enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Storage vMotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14

Platform Features for vCloud Director and View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

Hardware-Accelerated 3D Graphics Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14

VAAI Snapshots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15

VXLAN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16



IntroductionVMware vSphere® 5.1 continues to enhance the performance features and capabilities of the vSphere platform, making it the most robust and highest-performing cloud platform. vSphere 5.1 supports even larger virtual machines and physical hosts to accommodate even the most demanding of workloads. It also introduces several new features that reduce latency and increase throughput for network, storage and compute.

The following are some of the performance highlights in vSphere 5.1:

VMware® vCenter Server™•VMware vSphere Web Client™

•VMware vCenter™ Single Sign-On server scalability

•vCenter Server database and statistics collection enhancements

Core platform•64-way vCPU scalability

•256 pCPU support

•New physical processor enablement

•Memory overhead reduction

•Virtualized CPU performance counters

•Latency setting

Storage

•VMware vSphere Storage I/O Control enhancements

•VMware vSphere Storage DRS™ enhancements

•16GB Fibre Channel support

•Hardware iSCSI jumbo frames support

Network

•Single-root I/O virtualization (SR-IOV)

•VMware vSphere Distributed Switch™ (VDS) scalability improvements

VMware vSphere vMotion®•vMotion enhancements

•VMware vSphere Storage vMotion® enhancements

Platform features for VMware vCloud® Director™ and VMware View™•3D graphics

•VMware vSphere APIs – Array Integration (VAAI) snapshots

•VXLAN



vCenter Server ScalabilityManageability has become just as important as the day-to-day performance of applications. Simplifying IT operations reduces costs and makes businesses more flexible and agile, ready to quickly respond to changes in the environment. The improvements made in the VMware vSphere 5.1 Web Client, Single Sign-On server, and vCenter Server backend provide the flexibility to manage several different services from anywhere with a single authentication. VMware vCenter 5.1 provides the performance and data required to run a growing virtualization environment.

vSphere Web ClientIn vSphere 5.1, the VMware vSphere Web Client is the new primary point for vSphere management. It has several advantages over the previous Microsoft Windows–based VMware vSphere Client™. It is platform agnostic, built to handle thousands of objects and be highly extensible. The Web Client also improves vCenter Server tasks and workflows by preventing “workflow lock,” where one task in the GUI takes precedence and prevents the user from switching to another task without first ending the current task. In the Web Client, vSphere introduces a new feature called “Work in Progress,” with which the user can pause a current task to work on another task in the UI and then later resume the original task at the point it was interrupted.

The Web Client also expands the scalability of vSphere management and supports up to 300 concurrent Web Client users, a 200 percent increase over the previous standard vSphere Client.

Figure 1. The VMware vSphere Web Client



Single Sign-On ServervSphere 5.1 and vCenter Server 5.1 now utilize vCenter Single Sign-On server to provide centralized user access control and management. In a vSphere environment, users might have a need to access several vSphere infrastructure components, such as VMware vCenter Orchestrator™, vCenter Server and the vCenter Inventory Service. Utilizing vCenter Single Sign-On server, the user must provide security credentials to the client only once; this validates them against vCenter Single Sign-On server. If the validation is successful, vCenter Single Sign-On server issues a security token that can be used by the client to access any of the vSphere infrastructure components according to the user’s permissions and rights. vCenter Server 5.1 and the Web Client utilize vCenter Single Sign-On server for all user authentication.

vCenter Single Sign-On server can support more than 30,000 transactions per hour, enough to meet the needs of large-scale cloud environments with hundreds of concurrent clients and authentication activities. vCenter Single Sign-On server also supports a clustered mode where two or more servers work together for high availability and load balancing. In clustered mode using two vCenter Single Sign-On servers, performance for authentication tasks improved by up to 42 percent over that of a single server while providing redundancy for high availability.

vCenter Server EnhancementsvSphere 5.1 enables greater levels of consolidation and larger virtualized environments. As a result, vCenter Server 5.1 has also been enhanced to provide better scalability and support for these larger environments, reduce latency on vCenter Server database activities, and enable the collection of a greater number of vSphere statistics.

vCenter Server 5.1 is now more efficient in how it does its statistics collection and processing. As a result, it is now able to handle higher statistics collection levels for larger environments—up to 80 million statistics/hour—which corresponds to collecting vCenter Server statistics at statistics level 4 for a massively large cloud environment consisting of 1,000 vSphere hosts, 2,000 connected datastores and 10,000 powered-on virtual machines.

STATISTICS COLLECTION LEVEL

NUMBER OF STATISTICS/HOUR

NUMBER OF STATISTICS/ 5 MINUTES

Level 1 10M 0.83M

Level 2 15M 1.25M

Level 3 68M 5.67M

Level 4 80M 6.67M

Table 1. Statistics/Hour for Various vCenter Server Statistics Levels

Core Platform64-Way vCPU ScalabilityVMware vSphere 5.1 can now support virtual machines with up to 64 vCPUs. Being able to utilize larger virtual machines enables large-scale tier 1 applications to be virtualized while still achieving the mission-critical high performance that they demand. VMware performance engineering has demonstrated that large CPU and memory-intensive workloads (HPC workloads: SPEC MPI and SPEC OMP benchmarks) running on 64-vCPU virtual machines in vSphere 5.1 can achieve near native performance. For full details on the SPEC MPI and SPEC OMP HPC benchmark results, visit http://labs.vmware.com/publications/performance-evaluation-of-hpc-benchmarks-on-vmwares-esxi-server.

http://labs.vmware.com/publications/performance-evaluation-of-hpc-benchmarks-on-vmwares-esxi-server

http://labs.vmware.com/publications/performance-evaluation-of-hpc-benchmarks-on-vmwares-esxi-server



Other business-critical applications have also been shown to scale well in a vSphere virtualized environment as compared to a physical environment. Testing using Oracle Database and an OLTP-intensive workload has shown that even the most strenuous workloads scale well from 1 to 64 vCPUs in vSphere 5.1, comparable to the application’s native physical scaling capabilities. The ability to create larger 64-way virtual machines, combined with the scaling performance of vSphere, enables users to virtualize large business-critical workloads with confidence.

New Physical Hardware Support and Scalability vSphere 5.1 supports physical systems with up to 256 physical CPUs, up from 160. These larger servers enable even greater levels of consolidation and the capability to run hundreds of virtual machines per physical host. But consolidation alone is not enough. vSphere 5.1 includes a number of advanced features that enable a significant increase in consolidation levels on larger physical servers while maintaining performance fairness according to administrator-assigned virtual machine priorities. These features include the following:

•VMware vSphere Network I/O Control and Storage I/O Control ensure that lower-priority virtual machines cannot adversely impact the I/O performance of higher-priority virtual machines.

•VMware vSphere Distributed Resource Scheduler™ (vSphere DRS) and Storage DRS provide proper initial placement of virtual machines on hosts and storage and perform ongoing monitoring and load balancing to ensure optimal utilization.

vSphere 5.1 introduces support for the latest Intel and AMD CPUs including the Intel “Ivy Bridge” and “Sandy Bridge” series and the AMD “Piledriver” series.

For a comprehensive list of the supported guest operating systems (OS) and server hardware, refer to the VMware Compatibility Guide: http://www.vmware.com/resources/compatibility.

Memory Overhead ReductionLarge and complex virtual machines are typical in vSphere virtualized environments. The management processes that support them require resources. These management processes for large virtual machines can reserve large amounts of memory that prior to vSphere 5.1 were locked in the vSphere host’s physical memory—even if the process was actively using only a small amount of that memory. To balance the vSphere management processes’ memory usage against virtual machines demanding memory, vSphere 5.1 has introduced the ability to swap out and reduce the memory reservation for several of the vSphere backend processes. This feature can reduce memory overhead for vSphere backend processes by up to 1GB when the vSphere host is under memory pressure.

To use this feature, a system swap file must first be manually created. This can be accomplished by issuing the following command from the VMware ESXi™ console:

esxcli sched swap system set -d true -n <datastore name>

The swap file is typically between 500MB and 1GB, depending on the host memory size, and is created at the root of the specified datastore.

NOTE: This system swap file is different from the per–virtual machine swap files, which store memory pages from the VMX component of the per–virtual machine memory space overhead.

Virtualized CPU Performance Counters Virtual machines are becoming the standard in IT organizations as a stable and extensible platform to run many workloads. More and more software developers and quality assurance engineers are utilizing vSphere virtual machines as their development and testing platform. This trend has given rise to the need for virtual machines to provide more accurate counters from the underlying physical hardware.

http://www.vmware.com/resources/compatibility



vSphere 5.1 is able to expose more information about the CPU architecture to the virtual machine. This feature is called Virtualized Hardware Virtualization (VHV). VHV provides the guest OS with near-native access to the physical CPU. It enables, for example, running Windows XP Mode from inside a Windows 7 or Windows 8 virtual machine.

NOTE: Windows XP Mode enables running legacy Windows XP applications that are not compatible with newer Windows versions inside a simulated Windows XP environment on Windows 7 or Windows 8.

In addition to the improved CPU virtualization with VHV, vSphere 5.1 provides the ability to expose more low-level CPU counters to the guest OS, which enables improved debugging, tuning and troubleshooting of operating systems and applications—and enhances the ability to develop, test and optimize applications—running inside a VMware virtual machine.

Latency Sensitivity SettingvSphere 5.1 simplifies the process of configuring a virtual machine to support low-latency applications. It now offers a latency sensitivity advanced setting that automatically makes low-level changes in the vSphere kernel to reduce latency for the virtual machine. The latency sensitivity setting might improve the performance of latency-sensitive virtual machines when they are contending for CPU resources with non–latency-sensitive virtual machines. However, if the workload is not latency sensitive, the setting might have a negative performance impact. Therefore this feature should be enabled only after testing has shown performance benefits for the user’s particular application.

Figure 2. Latency Sensitivity Feature



StorageStorage I/O Control – Automatically Tuning Congestion ThresholdThe default latency threshold for Storage I/O Control is 30ms. Not all storage devices are equal, so this default value is set to the middle of the range. There are devices that hit their natural contention point earlier than others. One example of this is SSDs, for which the user should decrease the threshold. However, manually determining the correct latency can be difficult, so the latency threshold should be automatically determined for each device. Storage I/O Control now can do this instead of employing a default value or requiring user selection.

The latency threshold is set automatically to the value determined by the I/O injector (a new part of Storage I/O Control). When the I/O injector calculates the peak throughput, it then detects the 90 percent throughput value and measures the latency at that point to determine the threshold.

Figure 3. Automatic Latency Threshold Advantages

vSphere administrators can change this from 90 percent to another value or they can input a millisecond value if they choose.

Another enhancement to Storage I/O Control is that it now supports a passive statistics-only mode by default. This new feature enables Storage I/O Control to collect statistics even if it is not actively managing and controlling the user’s storage workloads. The new Storage I/O Control statistics-only mode also enables Storage DRS to get statistical information on datastores immediately when they are added to a cluster.

VmObservedLatencyThe improved Storage I/O Control now can monitor and control storage congestion occurring not only on the storage device but also inside the rest of the storage stack. In vSphere 5.1, VmObservedLatency is a new Storage I/O Control metric, replacing the datastore latency metric used in previous versions. The new metric measures the time between receipt of the I/O by the VMkernel from the virtual machine and receipt of the response from the datastore. Previously, Storage I/O Control measured the latency only after the I/O had left the vSphere host (device latency only). Now Storage I/O Control measures and controls storage workload latency throughout the whole virtualized storage stack. If all hosts using the datastore cluster leverage vSphere 5.1, Storage DRS uses the VmObservedLatency metric, which closely represents the load on the datastore as detected by virtual machines and is visible in the Web Client performance charts.



VMware

VM VM

Kernel LatencyTime an I/O request spent waiting inside the vSpherestorage stack.

VmObservedLatencyStorage I/O Control now monitorsand controls storage performance and congestion occurring at boththe device and kernel layers ofthe storage stack.

Storage Stack in a vSphere Virtual Environment

Device LatencyLatency coming from the physical hardware, HBA, and storage device.

ApplicationGuest OS

VMM

vSCSI

Driver

HBA

Fabric

Array SP

ESX StorageStack

Figure 4. VmObservedLatency

Storage DRS Enhancements – Correlation ImprovementsStorage DRS was a significant new feature introduced in VMware vSphere 5.0 that provided smart virtual machine placement and load-balancing capabilities based on I/O and space capacity. It helped decrease the operational effort associated with the provisioning of virtual machines and monitoring of the storage environment. In vSphere 5.1, Storage DRS has been further improved to detect correlated datastores better, preventing unnecessary Storage vMotion operations.

Datastore correlation refers to the process of discovering whether two distinct datastores might be using the same set of disk spindles on a storage array. The purpose of this feature is to help Storage DRS determine where to move a virtual machine. For instance, there is little advantage to moving a virtual machine from one datastore to another if the two datastores are backed by the same set of physical spindles on the array. The datastore-correlation detector also can assist in following antiaffinity rules, making sure that virtual machine disks are kept not only on separate datastores but also on different spindles on the backend array.

Although the initial release of Storage DRS included datastore-correlation detection capabilities, it relied on VMware vSphere APIs for Storage Awareness (VASA), which might not be supported by all storage arrays. The vSphere 5.1 datastore-correlation detection mechanism uses the I/O injector to determine whether a source and destination datastore are correlated. The I/O injector places a load on one datastore and monitors the latency on another. If the datastore-correlation mechanism measures a latency increase on both datastores when a load is placed on one datastore, it can detect that the datastores are correlated. This enhancement enables datastore-correlation detection on storage arrays without VASA support.

T E C H N I C A L W H I T E P A P E R / 1 0


16Gb FC SupportFibre Channel (FC) speeds continue to increase, with 16Gb FC cards now generally available. To take full advantage of these new cards, vSphere 5.1 has added support for 16Gb FC HBAs to enable better storage performance.

With the availability of 16Gb FC, throughput for larger block I/Os has doubled compared to 8Gb FC. In addition to increased throughput, 16Gb FC provides better CPU efficiency per I/O. For random I/Os in small blocks, 16Gb FC enables much higher IOPS than with 8Gb FC. These improvements in random read and write I/Os for small block sizes boost performance for most I/O-intensive applications.

In single–virtual machine testing with new 16Gb FC HBAs using a single storage I/O thread, vCPU throughput improved by as much as 25 percent for a 4KB block size, 75 percent for an 8KB block size and 100 percent when using a 16KB or larger block size. CPU efficiency for all block sizes was improved by approximately 3–15 percent compared to tests using the 8Gb FC HBA.

Block size

1KB

0

200

400

600

800

1000

1200

1400

1600

1800

4KB 8KB 16KB 32KB 64KB 256KB

8Gb16Gb

Sequ

entia

l rea

d th

roug

hput

(M

Bps

)

Figure 5. Sequential Read Throughput in Megabytes per Second



Block size

100%

95%

90%

85%

80%

75%

1KB 4KB 8KB 16KB 32KB 64KB 256KB

CPU

cos

t per

I/O

(low

er is

bet

ter)

8Gb16Gb

Figure 6. Sequential Reads, Relative CPU Cost per I/O (Lower Is Better and Means a Higher CPU Efficiency)

When testing with multiple virtual machines using multiple storage I/O threads, the 16Gb FC HBA achieved 196K read IOPS for an 8KB block size and more than 500K read IOPS for a 1KB block size. This represents a throughput of 1,535MBps, which is near the 16Gbps wire speed for 16Gb FC and a 100 percent improvement in throughput over 8Gb FC.

For full details, see the Storage I/O Performance on VMware vSphere 5.1 over 16 Gigabit Fibre Channel technical white paper available on the vmware.com Web site.

iSCSI Storage Driver Upgrade for Jumbo Frames The emergence of 10Gb Ethernet in the datacenter and the popularity of network-based storage have created the need to provide features that enable users to fully utilize larger networking pipes. Jumbo frame support was available in previous releases of vSphere for virtual machine, NFS, and software-based iSCSI network traffic. vSphere 5.1 adds jumbo frame support to all hardware iSCSI adapters, including both dependent and independent hardware iSCSI adapters. Utilizing 10Gb networking and the new vSphere 5.1 jumbo frame support, hardware-based iSCSI throughput can be significantly increased.

PROTOCOL READ THROUGHPUT WRITE THROUGHPUT

HWiSCSI +88% +20%

SWiSCSI +11% +40%

NFS +9% +32%

64Kb Block Size, 100% Sequential

Table 2. Throughput Improvements When Using Jumbo Frames with 10Gb Ethernet



NetworkvSphere 5.1 adds a number of powerful new features and enhancements to the networking capabilities of the vSphere platform. These new features enable users to run network latency-sensitive applications and scale their virtual environment to new heights.

VDS Scalability Improvements and EnhancementsThe vSphere Distributed Switch is a centrally managed, datacenter-wide virtual switch. It offers the same raw performance as the standard virtual switch but includes advanced networking features and scalability. Having one centrally managed virtual switch across the entire vSphere environment greatly simplifies networking and network management in the datacenter. vSphere 5.1 increases the manageability and recovery of the VDS by providing automatic rollback from misconfigurations as well as recovery and reconfiguration capabilities directly from the local console of the vSphere host.

In addition to the manageability and recovery enhancements made to the VDS, vSphere 5.1 greatly increases the number of switches supported per server and doubles the number of port groups supported per vCenter Server.

VDS PROPERTIES 5.0 LIMIT 5 .1 LIMIT

Number of VDS per vCenter Server 32 128

Number of Static Port Groups per vCenter Server 5,000 10,000

Number of Distributed Ports per vCenter Server 30,000 60,000

Number of Hosts per VDS 350 500

Table 3. New VDS Scalability Numbers for vCenter Server

Single-Root I/O Virtualization Single-root I/O virtualization (SR-IOV) is a standard that enables one PCI Express (PCIe) adapter to be presented as multiple separate logical devices to virtual machines. The hypervisor manages the physical function (PF) while the virtual functions (VFs) are exposed to the virtual machines. In the hypervisor, SR-IOV–capable network devices offer the benefits of direct I/O, including reduced latency and host-CPU utilization. The ESXi platform’s virtual machine direct-path (pass-through) functionality provides similar benefits to the user but requires a physical adapter per virtual machine. In SR-IOV, the pass-through functionality can be provided from a single adapter to multiple virtual machines through virtual functions. While a paravirtualized device such as vmxnet3 is generally capable of matching SR-IOV for throughput, SR-IOV can typically achieve the same throughput using a lower amount of CPU resources on the host. Testing with netperf at very high packet rates (>=100,000 packets per second) has shown that SR-IOV consumes up to 45 percent less in CPU resources when compared to vmxnet3. SR-IOV was also shown to have lower latencies, approximately 5µs lower compared to vmxnet3. SR-IOV can be beneficial for applications that have extremely low latency requirements (end-to-end latency <10µs) or extremely high packet rates (>100K packets per second).

vMotionvMotion EnhancementsvSphere 5.1 offers new enhancements to vMotion that provide a new level of ease and flexibility for live virtual machine migrations. It adds the ability to perform a vMotion migration between two hosts without a SAN or shared storage. This is done by combining vMotion and Storage vMotion into one operation. The combined migration copies both the virtual machine memory and its disk over the network to the destination host. In smaller environments, the ability to simultaneously migrate both memory and storage enables virtual machines to be moved between hosts that do not have shared storage. In larger environments, this capability enables virtual machines to be moved between clusters that do not have a common set of datastores.



APPOS

TCP/IP

No shared storage required

OS

APP

OS

APP

OS

APPVMware ESX

OS

APP

OS

APP

OS

APPVMware ESX

Figure 7. vMotion Migration with vSphere 5.1 Using the TCP/IP Network Connection

•vSphere 5.1 vMotion can leverage the multi-NIC feature, added in vSphere 5.0, to load-balance the vMotion network traffic over multiple network adapters. The multi-NIC feature enables users to provision multiple vMotion network interfaces on the source and destination hosts. When a migration is initiated, vSphere matches source and destination vMotion NICs based on link speed, pairing multiple NICs to a single NIC as needed to fully utilize the link. vSphere opens a TCP connection per network adapter pair and transparently load-balances the migration traffic over all the connections.

0.6

0.8

1

1.2

Memory Migration Time(Normalized to Traditional vMotion)

Traditional vMotion vSphere 5.1 vMotion

0.00

0.20

0.40

0.60

0.80

1.00

1.20

1.40

Traditional Storage vMotion(Locally Connected)

Disk Migration Time(Normalized to Traditional vMotion)

vSphere 5.1 vMotion (Network Based)

Figures 8–9. Disk Migration Time and Memory Migration Time .

•During storage migration, vMotion 5.1 maintains the same performance as Storage vMotion, even when using the network to migrate, due to the optimizations added to the vMotion 5.1 network data path.

•During memory migration, vMotion 5.1 maintains near-identical performance to the traditional vMotion, due to the optimizations added to the vMotion 5.1 memory copy path.

•vMotion 5.1 retains the proven reliability, performance and atomicity of the traditional vMotion and Storage vMotion technologies, even at metropolitan area network (MAN) distances.



Storage vMotionIn vSphere 5.1, up to four parallel disk copies per Storage vMotion operation can be performed. In previous versions, vSphere serially copied disks belonging to a virtual machine. Now, if a request to copy six disks in a single Storage vMotion operation is received, the first four copies are initiated simultaneously. Then, as soon as one of the first four finishes, the next disk copy is started.

Migration

30%

0%

0%

0%

0%

Migration

30%

20%

26%

18%

0%

VMDK-1

VM-1

VMDK-2

VMDK-3

VMDK-4

VMDK-5

VMDK-1

VM-1

VMDK-2

VMDK-3

VMDK-4

VMDK-5

VMDK-1

VM-1

VMDK-2

VMDK-3

VMDK-4

VMDK-5

VMDK-1

VM-1

VMDK-2

VMDK-3

VMDK-4

VMDK-5

Storage vMotion process virtual machine 1Storage vMotion process virtual machine 1

vSphere 5.1 Storage vMotion processPre vSphere 5.1 Storage vMotion process

Figure 10. Storage vMotion in vSphere 5.1

To reduce performance impact on other virtual machines sharing the datastores, parallel disk copies apply only to disk copies between distinct datastores. This means that if a virtual machine has multiple VMDK files on datastores A and B, parallel disk copies will occur only if the destination datastores are C and D.

A fan-out disk copy—in other words, copying two VMDK files on datastore A to datastores B and C—will not have parallel disk copies. The common use case of parallel disk copies is the migration of a virtual machine configured with an antiaffinity rule inside a datastore cluster.

The limit of eight concurrent Storage vMotion operations doesn’t directly relate to the parallel disk copy change. For example, even if only one Storage vMotion operation is issued (leaving room for another seven operations on the target datastores), that single operation might be moving multiple disks related to a virtual machine.

Platform Features for vCloud Director and ViewHardware-Accelerated 3D Graphics SupportWith vSphere 5.1, VMware has partnered with NVIDIA to provide hardware-based virtual GPU (vGPU) support inside the virtual machine. vGPUs improve the graphics capabilities of a virtual machine by offloading graphic-intensive workloads to a physical GPU installed on the vSphere host. In vSphere 5.1, the new vGPU support targets View environments that run graphic-intensive workloads such as graphics design and medical imaging.

Hardware-based vGPU support in vSphere 5.1 is limited to View environments running on vSphere hosts with supported NVIDIA GPU cards (refer to the VMware Compatibility Guide on the vmware.com Web site for details on supported GPU adapters). In addition, the initial release of vGPU is supported only with desktop virtual machines running Windows 7 or Windows 8. Refer to the View documentation for more information on the vGPU capabilities of vSphere 5.1.



vGPU support enables virtual machines to have four times the amount of video memory configured, from 128MB to 512MB per virtual machine. The increased video processing and memory support now enables graphic-intensive applications to be run in a View virtual desktop environment. Utilizing hardware-based vGPU support can reduce up to 15 percent of the CPU utilization in the physical server while delivering improved video performance and a better user experience.

NOTE: vGPU support is enabled in vSphere 5.1, but the ability to leverage this feature is dependent on a future release of View. Refer to the View documentation for information on when these features will be available.

VAAI SnapshotsVMware vSphere APIs – Array Integration (VAAI) enables certain storage operations to be offloaded from the vSphere host to the storage array. vSphere 5.0 introduced the offloading of View desktop snapshots to the array via VAAI NAS primitives. vSphere 5.1 introduces additional VAAI NAS enhancements to enable array-based snapshots to be used for vCloud Director fast-provisioned VMware vSphere vApps™. This feature in vSphere 5.1 enables vCloud Director to offload the creation of linked clones to the NAS storage array in a similar manner to how View does it in vSphere 5.0. When vCloud Director does a fast provision of a vApp/virtual machine, it transparently uses VAAI NAS to offload the creation of the subsequent linked clones to the VAAI-supported storage array.

Similar to VAAI NAS support for View in vSphere 5.0, this feature requires a special VAAI NAS plug-in from the storage array vendor.

VXLANPrior to vSphere 5.1, vSphere and vCloud Director provided network isolation using VLANs or cross-host fencing, based on a MAC-in-MAC encapsulation scheme. While providing network isolation is valuable within a datacenter, users needed to instantly provision isolated networks that can span across datacenters seamlessly and scale to handle the number of isolated networks required to support even the largest cloud environment. vSphere 5.1 introduces VXLAN (Virtual eXtensible LAN) support, which enables multitenant networks at scale and is the first step toward logical, software-based networks that can be created on demand, enabling enterprises to leverage capacity wherever it’s available.

Using “MAC-in-UDP” encapsulation, VXLAN provides a layer 2 abstraction to virtual machines, independent of where they are located. It completely untethers the virtual machines from physical networks by enabling them to communicate with each other using a transparent overlay scheme over physical networks that can span layer 3 boundaries. Because virtual machines cannot detect physical network constraints and distinguish only the virtual layer 2 adjacency, the fundamental properties of virtualization such as mobility and portability are extended across traditional network boundaries. Furthermore, logical networks easily can be separated from one another, simplifying the implementation of true multitenancy.

vSphere hosts using VXLAN have been shown to deliver NIC line rate performance, fully utilizing the latest 10Gbps NIC hardware, with CPU processing overhead comparable to other previous network isolation solutions. VXLAN gives users the flexibility and ease provided by software-defined networking along with the performance and scalability required to support large cloud environments.



ReferencesFor more information on the performance features and studies mentioned in this paper, see the following documents on the vmware.com Web site:

What’s New in VMware vSphere 5.1 – Networking

What’s New in VMware vSphere 5.1 – Platform

What’s New in VMware vSphere 5.1 – Storage

VMware vSphere 5.1 vMotion Architecture, Performance and Best Practices

Storage I/O Performance on VMware vSphere 5.1 over 16 Gigabit Fibre Channel

Performance Best Practices for VMware vSphere 5.1

VMware, Inc. 3401 Hillview Avenue Palo Alto CA 94304 USA Tel 877-486-9273 Fax 650-427-5001 www.vmware.comCopyright © 2012 VMware, Inc . All rights reserved . This product is protected by U .S . and international copyright and intellectual property laws . VMware products are covered by one or more patents listed at http://www .vmware .com/go/patents . VMware is a registered trademark or trademark of VMware, Inc . in the United States and/or other jurisdictions . All other marks and names mentioned herein may be trademarks of their respective companies . Item No: VMW-WP-WHATS-NEW-vSPHR-5 .1-PERF-USLET-103 Docsouce: OIC-12VM007 .09

what’s new in vmware vsphere 5.1 – performance€¦ · · 2016-11-24the web client also...

Documents