VMware vFlash Read Cache configuration and performance test

VMware vFlash Read Cache configuration and performance test

In previous post I showed you how to configure MLC IO Accelerator from HP (it is basically Fusion IO product branded by HP). Today I will show you how can we use it with VMware vFlash Read Cache.

Benefits of using vFlash Read Cache

Virtual Flash allows you to accelerate virtual machine performance through the use of local SSD disks, which serve flash memory cache to chosen virtual machines running on the ESXi host. Virtual Flash Read Cache can supply low latency for extreme latency sensitive applications, thus enabling the virtualization of some computer systems/applications previously considered too I/O intensive and, therefore, impossible/implausible.

In this picture you can see architecture overview.

VMware vFlash Read Cache configuration and performance test 4

vFlash Read Cache configuration

Whole configuration is done using vSphere Web Client.

  • Navigate to host and click Manage then Settings and Virtual Flash Resource Management and click Add Capacity. VMware vFlash Read Cache configuration and performance test 1
  • Select your local SSD device - in my case Fusion IO / HP MLC IO Accelerator and click ok. VMware vFlash Read Cache configuration and performance test 2
  • Once you click Ok you will see it available for virtual Flash Read Cache. VMware vFlash Read Cache configuration and performance test 3

Enable vFlash Read Cache on Virtual Machine

Virtual Flash Read Cache can be configured only using vSphere Web Client. Before you can do it you have to move virtual machine to host which is configured for virtual Flash Read Cache.

  • Select Virtual Machine you want to accelerate and click Edit.
    Enable vFlash Read Cache on Virtual Machine - 1
  • Select Hard disk you want to accelerate with VMware vFlash Read Cache and expand it. Enable vFlash Read Cache on Virtual Machine - 2
  • Click **Advanced **and select:
    • Enable virtual Flash Read Cache
    • Select desired amount of flash in MB or GB
    • Choose block size
      Enable vFlash Read Cache on Virtual Machine - 3
  • Click OK and finish with editing of Virtual Machine.

Virtual Machine preparation

Before we will start any tests I installed Virtual Machine with Microsoft Windows Server 2012 R2. Here are details about Virtual Machine:

  • Operating System: Windows Server 2012 R2
  • vCPU: 1
  • vRAM: 4GB
  • SCSI Controller: LSI Logic SAS
  • Virtual Disk: 40GB Thick Provisioned Eager Zeroed
  • Virtual Machine Hardware Version: vmx-10
  • Virtual Flash Read Cache size: 40GB

Performance tests

If you want to heave some meaningful data to analyse we need to do some baseline performance tests.

Performance baseline - test with SQLIO

You can download SQLIO performance tool using this link http://www.microsoft.com/en-us/download/details.aspx?id=20163.

I installed SQLIO in default directory on my Windows Server system. Open command prompt as Administrator in SQLIO folder and execute command **sqlio -help **to see various options you can specify during test.
Performance baseline - test with SQLIO - 1

I started with changing file param.txt by setting testfile.dat size from 8MB to 39,8GB. Performance baseline - test with SQLIO - 2

1
E:\testfile.dat 2 0x0 40800

Next step is to simply create batch file with various parameters to find sweet spot between IOPS and MB/s.

Sequential read SQLIO test

Here is content of batch file:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
sqlio -Fparam.txt -kR -t8 -s180 -o8 -fsequential -b8 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o8 -fsequential -b32 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o8 -fsequential -b64 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o8 -fsequential -b128 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o32 -fsequential -b8 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o32 -fsequential -b32 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o32 -fsequential -b64 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o32 -fsequential -b128 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o64 -fsequential -b8 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o64 -fsequential -b32 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o64 -fsequential -b64 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o64 -fsequential -b128 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o128 -fsequential -b8 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o128 -fsequential -b32 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o128 -fsequential -b64 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o128 -fsequential -b128 -BH -LS E:\Testfile.dat

I started batch file by executing command: sequential_read.bat > sequential_read_results_180.txt. After few minutes we got results to analyze.

Sequential read SQLIO test results

To analyze results I used PowerShell script from Jonathan Kehayias which does pretty nice parsing to Excel file with charts.

Results from Excel: SQLIO performance sequential read 180 seconds Excel

Chart from data. SQLIO performance sequential read 180 seconds Excel chart

ESXi performance chart. SQLIO performance sequential read 180 seconds ESXi

As we clearly see best performance is achieved by using IO size 128 KB, with 32 pending IO. We achieved 526.34 MB per second with average latency 14 ms and 210.76 IOPS.

Random read SQLIO test

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
sqlio -Fparam.txt -kR -t8 -s180 -o8 -frandom -b8 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o8 -frandom -b32 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o8 -frandom -b64 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o8 -frandom -b128 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o32 -frandom -b8 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o32 -frandom -b32 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o32 -frandom -b64 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o32 -frandom -b128 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o64 -frandom -b8 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o64 -frandom -b32 -BH -LS E:\Testfile.dat

sqlio -Fparam.txt -kR -t8 -s180 -o64 -frandom -b64 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o64 -frandom -b128 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o128 -frandom -b8 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o128 -frandom -b32 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o128 -frandom -b64 -BH -LS E:\Testfile.dat
  
sqlio -Fparam.txt -kR -t8 -s180 -o128 -frandom -b128 -BH -LS E:\Testfile.dat

Random read SQLIO test results

Results from Excel.

SQLIO performance random read 180 seconds Excel

Chart from data.

SQLIO performance random read 180 seconds Excel chart

ESXi performance chart.

SQLIO performance random read 180 seconds ESXi

As we clearly see best performance is achieved by using IO size 128 KB, with 128 pending IO. We achieved 695.99 MB per second with average latency 45 ms and 5567.95 IOPS.

Performance test with SQLIO accelerated with vFlash Read Cache

I ran the same tests to check what kind of performance boost I will get.

Sequential read SQLIO test results accelerated

Results from Excel. SQLIO performance sequential read 180 seconds cached Excel

Chart from data.

SQLIO performance sequential read 180 seconds cached Excel chart

ESXi performance chart shows quite strange results.

SQLIO performance sequential read 180 seconds cached ESXi

I checked as well vFlash Cache statistics.

vFlash Read Cache cache stats sequential 180 seconds

As we clearly see best performance is achieved by using IO size 128 KB, with 128 pending IO. We achieved 828.02 MB per second with average latency 38 ms and 6624.17 IOPS.

Random read SQLIO test results accelerated

Results from Excel.

SQLIO performance random read 180 seconds cached Excel

Chart from data. SQLIO performance random read 180 seconds cached Excel chart

ESXi performance chart shows quite strange results. SQLIO performance random read 180 seconds cached ESXi

I checked as well vFlash Cache statistics.

vFlash Read Cache cache stats random 180 seconds

As we clearly see best performance is achieved by using IO size 128 KB, with 8 pending IO. We achieved 821.62 MB per second with average latency 1 ms and 6572.98 IOPS.

Performance results - cached vs no cached

For better view on charts I decided to split results.

Average latency results - sequential read test

vFlash Read cache sequential latency results

Average megabytes per second results - random read test

vFlash Read cache sequential megabytes per second results

Average IOPS results - sequential read test

vFlash Read cache sequential IOPS results

Average latency results - random read test

vFlash Read cache random latency results

Average megabytes per second results - random read test

vFlash Read cache random megabytes per second results

Average IOPS results - random read test

vFlash Read cache random IOPS results

Performance results conclusions

After really long test I can clearly say that I am happy with achieved results. As you see I have quite fast underlying Fibre Channel storage. In my case you see improvement with cached solution pretty much from the beginning of tests in all areas. If your storage is quite slow you will see much better performance increase than me. I hope this post was informative for you and you will test vFlash Read Cache as well with same good results as me.

All rights reserved © Wojciech Marusiak
Last updated on 24 Mar, 2015 15:30 ETC
Built with Hugo
Theme Stack designed by Jimmy