PernixData FVP testing - part 3

Configured the Flash Cluster, it’s time to verify the enhancements PernixData FVP brings to the storage running some stress tests. To test FVP VMware-IO-Analyzer 1.6, FIO 2.1.9 and Jetstress 2010 benchmark tools have been used.

Tests have been performed in my LAB environment where hardware and workload are of course different compared to a real production environment therefore results should be taken just to have an'idea of the FVP potential.

Blog series

• PernixData FVP installation – part 1
• PernixData FVP configuration – part 2
• PernixData FVP testing – part 3
• PernixData FVP uninstall – part 4

Test FVP with VMware-IO-Analyzer

VMware-IO-Analyzer parameters

Exchange 2007: 55% read, 80% random, block size 8k
SQL_Server_64k: 66% read, 100% random, block size 64k

FVP: OFF

FVP: ON > Write Policy: write-through

FVP: ON > Write Policy: write-back

Test FVP with FIO

FIO parameters

100% read, 100% sequential, block size 8k, IO depth 32, 16 jobs

FVP: OFF

FVP: ON  > Write Policy: write-through

Test FVP with Jetstress 2010

Jetstress parameters

Operation mix: Sessions 8, Inserts 40%, Deletes 20%, Replaces 5%, Reads 35%, Lazy Commits 70%

Test Scenario and Exchange Profile Summary
Test Scenario: Disk Subsystem Throughput Test
Test type: Performance
Run Database Maintenance: True
Test duration: 02:00:00
Capacity percentage: 100
Throughput percentage: 100
Suppress tuning: True
ThreadCount: 4
Output path: D:\JetStressResult
Database source: CreateBrandNewDatabases
Number of copies per database: 1
Database paths:
D:\DB\db1
Log paths:
D:\DB\db1

FVP: OFF

• IOPS: 51,365

FVP: ON > Write Policy: write-through

• IOPS: 67,112

Notes

PernixData FVP is definitely an interesting solution that brings real benefits to the storage.

Having 30 days fully working trial the administrators can test the product with different scenarios getting an overview of the FVP impact to the storage I/O performance.

Results strongly depend on used hardware and workload in the network and real benefits can be analyzed comparing parameters before and after FVP activation.