MSA1000/ DL380G4/ SAP4.7 --->> Performance Problem

[VASANTH KUMAR D]

Dear Experts,

One of our customer has been running this setup for SAP Application for the last 3 years. Since last 4 months, he has observed that SAP response has degraded. Plz refer the attached screen shot.

Following is the existing system information.

HP ProLiant DL380G4 2 * Intel Xeon 3.2GHz DC Processor, 8GB RAM, 2 * 72.8GB Ultra SCSI HDD with RAID1

HBA Type on Server QLA2340 DP FC HBA
HBA Firmware 9.1.3.18

HP StorageWorks MSA1000 7 * 72.8 Ultra SCSI HDD

MSA1000 Firmware v5.20

Logical partitions on MSA1000 Q: 100MB on RAID 5 and P: 416GB on RAID5

SAP Version 4.7

SAP Deployment 2-Tier Architecture (DB and Apps on a single server)

SAP Users 40 Concurrent Users

SAP Modules FI-CO, SD, MM

Database MSSQL2000 DB

OS Windows2003 Enterprise Server

Current configuration :
1. Logical Volume Raid Level: DISTRIBUTED PARITY FAULT TOLERANCE (RAID 5) stripe size=16kB
2. Disk Read Queue Length goes to 80 with a Read of 6,000,000 bytes/sec.
3. We have two logical drives presented from MSA1000 to Windows server.
Unit 0 = 99MB ->> Q Drive
Unit 1 = 416,686MB ->>P Drive

Would appreciate any solution/ workaround/ pointers to solve this problem.

Best Regards,
VASANTH
+91 98804 83100

[Prokopets]

Generally there's a classical answers:
1) By more disks
2) Don't use raid5 (especially, if there's many write operations on disk subsystem). Use raid1.

IMHO, it's a good idea to upgrade your very old MSA1000 to something new, like MSA2324 with approximately 25-30 disks 32Gb@15k. It should solve your disk subsystem issue.

[Patrick Terlisten]

Hello,

what has changed in the last months? Is there any mentionable data growth?

You can check with perfmon some performance counters. Take a look at the logical drive disk queue length and the response time. There should be not more then 2 waiting IOs for each physical disks. If your logical array contains 7 disks, there shouldn't be more then 14 waiting IOs in the queue. If the response time is more then 30ms you should more disks to the array to distribute the load over more disks.

Best regards,
Patrick