3 ways to improve performance from a storage perspective

22
# Vibhor Gupta - NetApp Tim Brazil - Perforce 3 Ways to Improve Performance from a Storage Perspective

Upload: perforce

Post on 17-Nov-2014

1.494 views

Category:

Technology


2 download

DESCRIPTION

In this session, get three takeaways about Perforce performance benchmarks and their results across varying storage protocols, using NetApp storage as an example. Learn how to use Perforce benchmarks and tools to validate the performance of your Perforce deployment; understand Perforce performance across different storage protocols; and get tips and tricks for deploying Perforce on varying storage technologies.

TRANSCRIPT

  • 1. #3 Ways to Improve Performancefrom a Storage PerspectiveVibhor Gupta - NetAppTim Brazil - Perforce

2. #Vibhor GuptaApplications Eng.NetApp Inc.Tim BrazilPerformance Lab Eng.Perforce SoftwareTim Brazil entered the software industry as a SysAdmin and soon moved into Quality Assuranceroles at leading database companies. He joinedPerforce in March 2007, where his focus isperformance-related issues as they apply toclient/server environments.Vibhor Gupta is a Technical MarketingEngineer at in Tech-Apps at NetApp. Hepartnered with Perforce, working on thecharacterization of the Perforce Application onNetApp Storage. He also is involved in thecharacterization of the EDA workload, wherePerforce is heavily used for SCM 3. #Agenda Introduction Benchmarks Test Results and Analysis Key Takeaways 4. #Introduction Perforce Server Deployment: Considerations Performance Cost Backup and Recovery 5. #Performance Lab Benchmarks 6. #Performance Lab Benchmarks Branchsubmit Branchsubmit-edit Browse Sync Deltas 7. Cold vs. Warm FilesystemCache# Cold cache Warm cache 8. Reference01 Dataset#Scenario Test Environment #1 Test Environment #2Description Sizesreference01 total size ~35 GBp4 files //depot/... | wc -l 25200000p4 files -a //depot/... | wc -l 72625000 revisionsp4 files //depot/main/0... | wc -l 70000 filesp4 files //depot/main/00/00/most/sites/have/longer/paths/jam/src/* | wc -l 70 filesp4 dirs //depot/r* | wc -l 35 release directories 9. Branchsubmit#Compute PhaseElapsed TimeExiting TimeCommit Rateintegrate -vcreate changesubmit 10. Branchsubmit-edit#integrateeditsubmitCompute PhaseElapsed TimeExiting TimeCommit Ratecreate change 11. Browse Benchmark#P4D Server32 Browse Children1024 Browses32 Browse Children1024 Browses65536 Total BrowsesTotal Completion Time 12. Sync Benchmark#P4D Server32 Sync Children32 Syncs32 Sync Children32 Syncs2048 Total SyncsTotal Completion TimeGB Synced 13. Deltas Benchmark#ClearCacheStartServerRunCmdRunCmd 7xRunCmdStopServerResult #1 Result #2Completion Time Per Command 14. #Result Analysis 15. #Write Performance (Cold Cache)120001000080006000400020000BranchsubmitFiles/SecondBranchsubmitNFSv3iSCSI9000800070006000500040003000200010000Branchsubmit-EditFiles/SecondBranchsubmit-EditNFSv3iSCSI 16. #Write Performance (Warm Cache)2500020000150001000050000branchsubmitFiles/SecondBranchsubmitFCPiSCSINFS250002000015000100005000NFS/iSCSI split(db only oniSCSI) 0branchsubmit-editFiles/SecondBranchsubmit-EditFCPiSCSINFSNFS/iSCSI split (dbonly on iSCSI)NFS/iSCSi split (db& journal on iSCSI) 17. #Read Performance160140120100806040200syncSecondsSyncFCPiSCSINFSNFS/iSCSI split(db only oniSCSI)300250200150100500browse(1)SecondsBrowseFCPiSCSINFSNFS/iSCSI (logsonly on iSCSI) 18. #Deltas876543210SecondsDeltasNFSiSCSIFCP 19. #Best Practices For best write performance, Perforce metadata and journals should beplaced on faster storage like FC or iSCSI. For NFS, recommended mount options local_lock=all and nocto For heavy server logging, use FCP or iSCSI. 20. #Key TakeawaysFC iSCSI iSCSI and NFS(SplitConfiguration)NFSPerformance Excellent Excellent Good FairCost Fair Excellent Excellent ExcellentEase of backup andrecoveryFair Fair Good ExcellentProtocolCriterion 21. #RESOURCESTechnical Report: http://www.netapp.com/us/media/tr-4164.pdfPublic Benchmarks: ftp://ftp.perforce.com/perforce/tools/benchmarks/ 22. #Questions ?