MaxDB performance on hardware and VMWare instance

[former_member254634]

Have just made a homogeneous system copy of our Solution Manager 7 production server (hardware instance) onto a VMWare instance to use as a sandbox. Noticed that the performance of the copy "feels" significantly worse than that of the hardware installation.

Is there any commentary/insight out there regarding special tuning or circumstances or actual performance penalty for MaxDB/SAP on VMWare instances? In other words, should I expect copies (or even fresh installs) to be slow on VMWare?

Here are the config details -

SolMan (hardware instance) - 
 Dell PE2850 2 x 2core 2.8 GHz Xeon, 6GB memory, PERC 5i RAID, O/S+ log files on RAID1, rest on RAID 5 filesystems. SLES 10-64 SP2, MaxDB 7.6.04.11, SolMan 7 SP13.

Solman Copy: VMWare ESXi 3.5 Update2 on Dell PE2950 2 x 4core 3 GHz Xeon, 32GB memory, PERC 6i RAID controller, all RAID6. VM instance allocated 4 CPU, 8GB memory: SLES 10-64 SP2, MaxDB 7.6.04.11, SolMan 7 SP13.

Thanks for your insights!

Rytis

[former_member254634]

Sunil - Simple test of I/O actually gives better results on the VMWare instance. Used the same filesystem where the database is stored.

Test: dd if=/dev/zero of=/sapdb/large bs=1024k count=10000

Hardware: 79MB/sec

VMWare instance: 177MB/sec

[lbreddemann]

>

> Sunil - Simple test of I/O actually gives better results on the VMWare instance. Used the same filesystem where the database is stored. 
> 
> Test: dd if=/dev/zero of=/sapdb/large bs=1024k count=10000 
> Hardware: 79MB/sec
> VMWare instance: 177MB/sec

Nice simple test, but useless.

It does not say anything about the I/O Performance the database will get as the database performance is usally impacted by a) reads and b) much smaller I/O tasks.

What you should do is:

1. enable time measurement on both systems

2. enable db analyzer on both systems.

Check the results there. Are there any differences concerning the avg. I/O times?

In general there is no direct disadvantage in using VMWare - in fact this is done inside SAP as well.

But you might get problems in figuring out, where your performance problems are since you have some more system layers to pass here.

regards,

Lars

[former_member254634]

Point taken, Lars. I was looking more for experiences or expectations when running similar instances in virutal and non-virtual worlds. I do understand that closer watching of systems is appropriate to isolating and tuning instances.

Rytis

[Former Member]

looks like issue with storage, have you tested the PERC 6i RAID controller performance before starting the system copy?

is it possible to perform the test now and compare the same with your existing system. you can do copying around 5 GB file?

cheers,

-Sunil