从 Hyper-V 复制到辅助站点的测试结果Test results for Hyper-V replication to a secondary site

本文提供了在 System Center Virtual Machine Manager (VMM) 云中将 Hyper-V VM 复制到辅助数据中心时的性能测试的结果。This article provides the results of performance testing when replicating Hyper-V VMs in System Center Virtual Machine Manager (VMM) clouds, to a secondary datacenter.

测试目标Test goals

测试目标是观察在稳定状态复制期间 Site Recovery 的性能如何。The goal of testing was to examine how Site Recovery performs during steady state replication.

  • 当 VM 已完成初始复制并且在同步增量更改时会发生稳定状态复制。Steady state replication occurs when VMs have completed initial replication, and are synchronizing delta changes.
  • 使用稳定状态对性能进行测量非常重要,因为除非发生意外中断,这是大多数 VM 保持的状态。It's important to measure performance using steady state, because it's the state in which most VMs remain, unless unexpected outages occur.
  • 测试部署包括两个本地站点,每个站点中有一台 VMM 服务器。The test deployment consisted of two on-premises sites, with a VMM server in each site. 此测试部署是典型的总公司/分公司部署,总公司作为主站点,分公司作为辅助站点或恢复站点。This test deployment is typical of a head office/branch office deployment, with head office acting as the primary site, and the branch office as the secondary or recovery site.

我们的操作What we did

下面是我们在测试过程中执行的操作:Here's what we did in the test pass:

  1. 使用 VMM 模板创建的 VM。Created VMs using VMM templates.
  2. 启动 VM 并捕获 12 个小时内的基准性能指标。Started VMs, and captured baseline performance metrics over 12 hours.
  3. 在主 VMM 服务器和恢复 VMM 服务器上创建云。Created clouds on the primary and recovery VMM servers.
  4. 在 Site Recovery 中配置复制,包括源云和恢复云之间的映射。Configured replication in Site Recovery, including mapping between source and recovery clouds.
  5. 为 VM 启用保护并让它们完成初始复制。Enabled protection for VMs, and allowed them to complete initial replication.
  6. 等待几个小时以便系统达到稳定状态。Waited a couple of hours for system stabilization.
  7. 捕获 12 个小时内的性能指标,所有 VM 在那 12 个小时内都保持在预期的复制状态。Captured performance metrics over 12 hours, where all VMs remained in an expected replication state for those 12 hours.
  8. 测量基准性能指标与复制性能指标之间的增量。Measured the delta between the baseline performance metrics, and the replication performance metrics.

主服务器性能Primary server performance

  • Hyper-V 副本(由 Site Recovery 使用)以异步方式跟踪对日志文件的更改,从而最大限度地降低了主服务器上的存储开销。Hyper-V Replica (used by Site Recovery) asynchronously tracks changes to a log file, with minimum storage overhead on the primary server.

  • Hyper-V 副本利用自我维护的内存缓存最大程度地降低用于跟踪的 IOPS 开销。Hyper-V Replica utilizes self-maintained memory cache to minimize IOPS overhead for tracking. 它将向 VHDX 进行的写入存储在内存中,并在将日志发送到恢复站点之前将它们刷新到日志文件中。It stores writes to the VHDX in memory, and flushes them into the log file before the time that the log is sent to the recovery site. 如果写入数达到了预先确定的限制,也会发生磁盘刷新。A disk flush also happens if the writes hit a predetermined limit.

  • 下图显示了复制的稳定状态 IOPS 开销。The graph below shows the steady state IOPS overhead for replication. 我们可以看到由复制导致的 IOPS 开销大约为 5%,非常低。We can see that the IOPS overhead due to replication is around 5%, which is quite low.

    主服务器结果

Hyper-V 副本使用主服务器上的内存来优化磁盘性能。Hyper-V Replica uses memory on the primary server, to optimize disk performance. 如下图中所示,主群集中所有服务器上的内存开销都很低。As shown in the following graph, memory overhead on all servers in the primary cluster is marginal. 所显示的内存开销是复制使用的内存占 Hyper-V 服务器上安装的总内存的百分比。The memory overhead shown is the percentage of memory used by replication, compared to the total installed memory on the Hyper-V server.

主服务器结果

Hyper-V 副本具有最小的 CPU 开销。Hyper-V Replica has minimum CPU overhead. 如图中所示,复制的开销范围是 2-3%。As shown in the graph, replication overhead is in the range of 2-3%.

主服务器结果

辅助服务器性能Secondary server performance

Hyper-V 副本使用恢复服务器上的少量内存来使存储操作数量最优化。Hyper-V Replica uses a small amount of memory on the recovery server, to optimize the number of storage operations. 图中汇总了恢复服务器上的内存使用率。The graph summarizes the memory usage on the recovery server. 所显示的内存开销是复制使用的内存占 Hyper-V 服务器上安装的总内存的百分比。The memory overhead shown is the percentage of memory used by replication, compared to the total installed memory on the Hyper-V server.

辅助服务器结果

恢复站点上的 I/O 操作数量是主站点上的写入操作数量的一个函数。The amount of I/O operations on the recovery site is a function of the number of write operations on the primary site. 让我们在与主站点上的总 I/O 操作量和写入操作量进行比较的情况下看一下恢复站点上的总 I/O 操作量。Let's look at the total I/O operations on the recovery site in comparison with the total I/O operations and write operations on the primary site. 图中显示,恢复站点上的总 IOPS 大约是The graphs show that the total IOPS on the recovery site is

  • 主站点上的写入 IOPS 的 1.5 倍。Around 1.5 times the write IOPS on the primary.
  • 主站点上的总 IOPS 的 37%。Around 37% of the total IOPS on the primary site.

辅助服务器结果

辅助服务器结果

对网络利用率的影响Effect on network utilization

在现有的每秒 5 GB 带宽中,在主节点与恢复节点(启用了压缩功能)之间平均每秒使用了 275 MB 网络带宽。An average of 275 Mb per second of network bandwidth was used between the primary and recovery nodes (with compression enabled), against an existing bandwidth of 5 Gb per second.

网络利用率结果

对 VM 性能的影响Effect on VM performance

一个重要的注意事项是复制对在虚拟机上运行的生产工作负荷的影响。An important consideration is the impact of replication on production workloads running on the virtual machines. 如果主站点针对复制进行了充分的设置,则不应当对工作负荷产生任何影响。If the primary site is adequately provisioned for replication, there shouldn't be any impact on the workloads. Hyper-V 副本的轻量跟踪机制可以确保在虚拟机中运行的工作负荷在稳定状态复制期间不受影响。Hyper-V Replica's lightweight tracking mechanism ensures that workloads running in the virtual machines are not impacted during steady-state replication. 下图中对此进行了解释。This is illustrated in the following graphs.

此图显示了在启用复制之前和之后运行各种工作负荷的虚拟机执行的 IOPS。This graph shows IOPS performed by virtual machines running different workloads, before and after replication was enabled. 可以看到两者之间没有差别。You can observe that there is no difference between the two.

副本影响结果

下图显示了在启用复制之前和之后运行各种工作负荷的虚拟机的吞吐量。The following graph shows the throughput of virtual machines running different workloads, before and after replication was enabled. 可以看到复制没有明显影响。You can observe that replication has no significant impact.

副本影响结果

结束语Conclusion

我们的结果清晰地表明,与 Hyper-V 副本配合使用的 Site Recovery 可以针对大型群集以最小的开销很好地进行扩展。The results clearly show that Site Recovery, coupled with Hyper-V Replica, scales well with minimum overhead for a large cluster. Site Recovery 提供了简单的部署、复制、管理和监视。Site Recovery provides simple deployment, replication, management and monitoring. Hyper-V 副本为成功进行复制扩展提供了必要的基础结构。Hyper-V Replica provides the necessary infrastructure for successful replication scaling.

测试环境详细信息Test environment details

主站点Primary site

  • 主站点具有一个群集,其中包含运行着 470 个虚拟机的五台 Hyper-V 服务器。The primary site has a cluster containing five Hyper-V servers, running 470 virtual machines.
  • 各个 VM 运行不同的工作负荷,并且都启用了 Site Recovery 保护。The VMs run different workloads, and all have Site Recovery protection enabled.
  • 群集节点的存储是由一个 iSCSI SAN 提供的。Storage for the cluster node is provided by an iSCSI SAN. 型号 - Hitachi HUS130。Model - Hitachi HUS130.
  • 每台群集服务器具有四个网卡 (NIC),每个为 1 Gbps。Each cluster server has four network cards (NICs) of one Gbps each.
  • 两个网卡连接到一个 iSCSI 专用网络,另外两个网卡连接到一个外部企业网络。Two of the network cards are connected to an iSCSI private network, and two are connected to an external enterprise network. 其中一个外部网络保留给群集通信使用。One of the external networks is reserved for cluster communications only.

主要硬件要求

服务器Server RAMRAM “模型”Model 处理器Processor 处理器数目Number of processors NICNIC 软件Software
在群集中的 Hyper-V 服务器:Hyper-V servers in cluster:
ESTLAB-HOST11ESTLAB-HOST11
ESTLAB-HOST12ESTLAB-HOST12
ESTLAB-HOST13ESTLAB-HOST13
ESTLAB-HOST14ESTLAB-HOST14
ESTLAB-HOST25ESTLAB-HOST25
128128
ESTLAB-HOST25 有 256 个ESTLAB-HOST25 has 256
Dell ™ PowerEdge ™ R820Dell ™ PowerEdge ™ R820 Intel(R) Xeon(R) CPU E5-4620 0 @ 2.20 GHzIntel(R) Xeon(R) CPU E5-4620 0 @ 2.20GHz 44 I Gbps x 4I Gbps x 4 Windows Server Datacenter 2012 R2 (x64) + Hyper-V 角色Windows Server Datacenter 2012 R2 (x64) + Hyper-V role
VMM 服务器VMM Server 22 22 1 Gbps1 Gbps Windows Server Database 2012 R2 (x64) + VMM 2012 R2Windows Server Database 2012 R2 (x64) + VMM 2012 R2

辅助站点Secondary site

  • 辅助站点具有一个六节点的故障转移群集。The secondary site has a six-node failover cluster.
  • 群集节点的存储是由一个 iSCSI SAN 提供的。Storage for the cluster node is provided by an iSCSI SAN. 型号 - Hitachi HUS130。Model - Hitachi HUS130.

主要硬件规格

服务器Server RAMRAM “模型”Model 处理器Processor 处理器数目Number of processors NICNIC 软件Software
在群集中的 Hyper-V 服务器:Hyper-V servers in cluster:
ESTLAB-HOST07ESTLAB-HOST07
ESTLAB-HOST08ESTLAB-HOST08
ESTLAB-HOST09ESTLAB-HOST09
ESTLAB-HOST10ESTLAB-HOST10
9696 Dell ™ PowerEdge ™ R720Dell ™ PowerEdge ™ R720 Intel(R) Xeon(R) CPU E5-2630 0 @ 2.30 GHzIntel(R) Xeon(R) CPU E5-2630 0 @ 2.30GHz 22 I Gbps x 4I Gbps x 4 Windows Server Datacenter 2012 R2 (x64) + Hyper-V 角色Windows Server Datacenter 2012 R2 (x64) + Hyper-V role
ESTLAB-HOST17ESTLAB-HOST17 128128 Dell ™ PowerEdge ™ R820Dell ™ PowerEdge ™ R820 Intel(R) Xeon(R) CPU E5-4620 0 @ 2.20 GHzIntel(R) Xeon(R) CPU E5-4620 0 @ 2.20GHz 44 Windows Server Datacenter 2012 R2 (x64) + Hyper-V 角色Windows Server Datacenter 2012 R2 (x64) + Hyper-V role
ESTLAB-HOST24ESTLAB-HOST24 256256 Dell ™ PowerEdge ™ R820Dell ™ PowerEdge ™ R820 Intel(R) Xeon(R) CPU E5-4620 0 @ 2.20 GHzIntel(R) Xeon(R) CPU E5-4620 0 @ 2.20GHz 22 Windows Server Datacenter 2012 R2 (x64) + Hyper-V 角色Windows Server Datacenter 2012 R2 (x64) + Hyper-V role
VMM 服务器VMM Server 22 22 1 Gbps1 Gbps Windows Server Database 2012 R2 (x64) + VMM 2012 R2Windows Server Database 2012 R2 (x64) + VMM 2012 R2

服务器工作负载Server workloads

  • 针对测试用途,我们选取了企业客户方案中常用的工作负载。For test purposes we picked workloads commonly used in enterprise customer scenarios.
  • 我们使用 IOMeter 与表中汇总的工作负载特性进行模拟。We use IOMeter with the workload characteristic summarized in the table for simulation.
  • 所有 IOMeter 配置文件都设置为写入随机字节以模拟工作负载的最坏情况写入模式。All IOMeter profiles are set to write random bytes to simulate worst-case write patterns for workloads.
工作负荷Workload I/O 大小 (KB)I/O size (KB) 访问百分比% Access 读取百分比%Read 未完成的 I/OOutstanding I/Os I/O 模式I/O pattern
文件服务器File Server 44
88
1616
3232
6464
60%60%
20%20%
5%5%
5%5%
10%10%
80%80%
80%80%
80%80%
80%80%
80%80%
88
88
88
88
88
全部 100% 随机All 100% random
SQL Server(卷 1)SQL Server (volume 1)
SQL Server(卷 2)SQL Server (volume 2)
88
6464
100%100%
100%100%
70%70%
0%0%
88
88
100% 随机100% random
100% 顺序100% sequential
ExchangeExchange 3232 100%100% 67%67% 88 100% 随机100% random
工作站/VDIWorkstation/VDI 44
6464
66%66%
34%34%
70%70%
95%95%
11
11
两者都 100% 随机Both 100% random
Web 文件服务器Web File Server 44
88
6464
33%33%
34%34%
33%33%
95%95%
95%95%
95%95%
88
88
88
全部 75% 随机All 75% random

VM 配置VM configuration

  • 主群集上有 470 个 VM。470 VMs on the primary cluster.
  • 所有 VM 都带有 VHDX 磁盘。All VMs with VHDX disk.
  • VM 运行表中汇总的工作负荷。VMs running workloads summarized in the table. 所有虚拟机都是使用 VMM 模板创建的。All were created with VMM templates.
工作负荷Workload # VM# VMs 最小 RAM (GB)Minimum RAM (GB) 最大 RAM (GB)Maximum RAM (GB) 每个 VM 的逻辑磁盘大小 (GB)Logical disk size (GB) per VM 最大 IOPSMaximum IOPS
SQL ServerSQL Server 5151 11 44 167167 1010
Exchange ServerExchange Server 7171 11 44 552552 1010
文件服务器File Server 5050 11 22 552552 2222
VDIVDI 149149 .5.5 11 8080 66
Web 服务器Web server 149149 .5.5 11 8080 66
总计TOTAL 470470 96.83 TB96.83 TB 41084108

Site Recovery 设置Site Recovery settings

  • 针对本地到本地保护配置了 Site RecoverySite Recovery was configured for on-premises to on-premises protection
  • VMM 服务器配置有四个云,包含 Hyper-V 群集服务器及其 VM。The VMM server has four clouds configured, containing the Hyper-V cluster servers and their VMs.
主 VMM 云Primary VMM cloud 受保护的 VMProtected VMs 复制频率Replication frequency 额外的恢复点Additional recovery points
PrimaryCloudRpo15mPrimaryCloudRpo15m 142142 15 分钟15 mins None
PrimaryCloudRpo30sPrimaryCloudRpo30s 4747 30 秒30 secs None
PrimaryCloudRpo30sArp1PrimaryCloudRpo30sArp1 4747 30 秒30 secs 11
PrimaryCloudRpo5mPrimaryCloudRpo5m 235235 5 分钟5 mins None

性能指标Performance metrics

此表汇总了在部署中测量到的性能指标和计数器。The table summarizes the performance metrics and counters that were measured in the deployment.

指标Metric 计数器Counter
CPUCPU \Processor(_Total)% 处理器时间\Processor(_Total)% Processor Time
可用内存Available memory \Memory\Available MBytes\Memory\Available MBytes
IOPSIOPS \PhysicalDisk(_Total)\Disk Transfers/sec\PhysicalDisk(_Total)\Disk Transfers/sec
VM 读取 (IOPS) 操作数/秒VM read (IOPS) operations/sec \Hyper-V Virtual Storage Device(<VHD>)\Read Operations/Sec\Hyper-V Virtual Storage Device(<VHD>)\Read Operations/Sec
VM 写入 (IOPS) 操作数/秒VM write (IOPS) operations/sec \Hyper-V Virtual Storage Device(<VHD>)\Write Operations/S\Hyper-V Virtual Storage Device(<VHD>)\Write Operations/S
VM 读取吞吐量VM read throughput \Hyper-V Virtual Storage Device(<VHD>)\Read Bytes/sec\Hyper-V Virtual Storage Device(<VHD>)\Read Bytes/sec
VM 写入吞吐量VM write throughput \Hyper-V Virtual Storage Device(<VHD>)\Write Bytes/sec\Hyper-V Virtual Storage Device(<VHD>)\Write Bytes/sec

后续步骤Next steps

设置复制Set up replication