SAP v4.6 SAN Performance Issue

[SANitary]

Our SAP BASIS team is battling performance issues with our new SAP 46 Production system. Part of the on-going analysis involves looking at disk performance.

Short Synopsis: Using information in the Microsoft SQL Server 2000 Performance Tuning Technical Reference there are some values in performance monitor that appear high indicating a performance bottleneck with disk I/O. Specifically, the Average Disk Sec/read and Average disk sec/write are high which indicates disk latency.

The hardware is as follows:

Server: Compaq DL760 8xP3 700 Xeon, 8GB RAM
OS: Windows 2000 AS with SP2 with the /3GB and /pae switches
SQL: SQL 2000 EE

SAN attached via Connectrix Directors, EMC Symmetrix 8730, using 73Gb physicals sliced into 11.6Gb logicals. This server is using six 7-logical member Meta volumes, down three PowerPath connections.

This is similar to a problem we looked at last year. At the time, the conclusion was that Logical Volume Manager mis-reports counters when lots of disks are involved.

The main problem I have is that I think the numbers are whacked. When I see device busies greater than 100% (in this case, you can find devices that are 20,000% busy), I think the numbers are suspect. Write I/O time is high, too, being well over one second in some cases (as I understand it, this is service time which does not include queue).

Is this a counter problem? A reporting problem?

Any insights would be appreciated.

 

[crowe]

The best way to meter the disk IO is on the 8730. If you have support with EMC, they can capture and generate a report on the performance of the symm.
Do you see within power path that your three channels are load balencing?
Are you binding the LUNS into one logical with the OS, or presenting one LARGE LUN to the OS?
Are your 11gig logicals RAID-S or RAID-1? It has been my experience on the 8730 that RAID-1 out performs RAID-S, by a long shot.
Can you access the switch to see the throughput on the ports?

crowe

 

[SANitary]

Thank you Crowe. The entire frame is made up of Raid-1, and all three PowerPaths are load balancing as designed. I am not seeing any error, or CRC errors at the port level, and I am convinced that the problem resides somewhere between the O/S and the HBA (driver perhaps). Perfmon is not a viable tool for monitoring, and after much discussing with our SAP team, I now know that the previous version of SAP was faster until it recently got upgraded to v4.6.

 

[crowe]

Well, I guess I'm confused. Did the upgrade of the SAP software require upgrading the OS or newer HW? I guess what I'm really after is did the older SAP version run on the same HW? Did you have to upgrade/install the HBA and/or drivers?

crowe