azz
Contributor
Contributor

Performance issue ESX3.5 and HDS9585

Hi,

After upgarding the firmware of a SAN library HDS 9585, the overall performance became really poor (The firmware is 165F/E).

The CPU of the HDS 9585 raise up to 80% and every action became very slow (for logical and physical servers).

The number of IO is very low (Bellow 1000 IO) ans the bandwith/transfer usage is also very low. So it is not an IO load issue.

It seems that this happen as soon as VMs are started. When VMa are stopped, the problem disappears.

Our SAN team opened a case at HDS support? Until now, there have no real idea of the source. They stated that this firmware is installed at many customer whitout problem.

They also tols us to check our environment especially VMware.

Anybody else met similar issue?

Any help, any idea will be appreciated?

Thanks

Azz

0 Kudos
2 Replies
mvoss18
Hot Shot
Hot Shot

Are you using Fixed multipathing and manually load balancing your paths? We've seen performance issues when folks manually load balanced paths in a simple manor. For example if you have odd numbered LUNs manually set to use SP1 while even numbered LUNs are set to use SP2. If LUN1 and LUN2 are part of the same disk group, the same SP should be preferred. If SP1 and SP2 are trying to hit the disk group with LUN1 and LUN2 simultaneously, the SAN might experience thrashing as control of the disk group is passed back and forth between SPs. Make sure, if manually setting preferred paths, that the same SP is preferred for all LUNs within a disk group.

Are you using LUSE LUNs? This is not recommended. Make sure the Host Mode option is correct. I've seen the Host Mode option cause issues with an HDS SAN.

http://kb.vmware.com/selfservice/viewContent.do?externalId=3408142&sliceId=1

Are the virtual machine partitions aligned? This isn't likely the issue but good to know.

http://www.vmware.com/pdf/esx3_partition_align.pdf

0 Kudos
azz
Contributor
Contributor

Hi Mvoss,

First thanks for your response...

The problem was solved. It was a kind of "bug' in the HDS Firmware when dealing with a Lun which was manually put in an invalide state.

It seems that The new firmware tried to do a "parity check" on this Lun in priorised Mode without success . Why this was amplified when VM started? No idea, no answer (although HDS technicien was on site and Logs and Dumps was analysed at HDS Support).

So, the case is closed although we have not all answers.

Thanks

0 Kudos