Performance issue with PERC4e/DC and PowerVault 220S

Question

We have a strange issue with the captioned hardware running Windows Server 2003 cluster.

We recently setup a new active/standby cluster with two PE2950 servers with PREC4e/DC adapters and both are connected to PV220S with the LVD cable provided by Dell. SCSI ID 6 and 7 are given to the PERCs respectively. PV220S has 12 disks installed and its running in clustering mode. PERCs are also configured with clustering FW.

The configurations can be summarized as follows:
PowerEdge 2950 x 2
PERC5/i (system, 2 x SAS, RAID-1)
Firmware: 5.1.1-0040
Driver: 2.08.00.32
PERC4e/DC (shared disk, 12 x SCSI, RAID-5)
Firmware: 5A2D
Driver: 6.46.2.32
PowerVault 220S
Firmware: E.19

Windows cluster itself is running without any problems, file storing and printing are functioning perfectly and it can perform failover correctly. The only problem is that it stops responding every 4 hours which occurs 10:30am 2:30pm 6:30pm... regardless of startup time. When this happens, explorer and/or applications that are using the shared disk freeze both on server and clients and it will be back to the normal state in a few minutes. I noticed that the similar lockups happens when the disk access is very busy e.g. when we perform file search in entire shared disk or when Backup Exec starts pre-processing backup jobs. Therefore I checked perfmon and found that Disk Queue Length spikes when the issue occurs.

I understand that PERCs in clustering mode i.e. with write-back cache disabled could perform much slower than non-clustering mode so I can schedule tasks that require intensive disk access in midnight etc but cannot understand why the lockups happens every 4 hours. I have checked the server settings but found no periodic task running at those times. There's nothing in the event log, either. I have tried removing the anti-virus software and Backup Exec but still got the same issue without those software.

I have already called Dell technical support and have sent DSET log but they said no hardware issue indicated. Can someone shed light on this?

kodchiang · Answer

I found a suspicious descritption in Dell document:

http://support.dell.com/support/edocs/software/svradmin/5.1/en/omss_ug/html/cntkpatm.html

It says "For example, on some controllers the Patrol Read runs every four hours." so I disabled the Patrol Read function and will see how it goes.

dharold · Answer

I have the same configuration as yourself and have the same issue occur.  I disabled Patrol reading, updating the system bios, perc 4e/dc firmware, and the harddrive firmwars.  I was able to see some improvement but nothing great.  After talking with Dell and doing some research online it seems to be simply a limitation of the Perc card in clustering mode.  Clustering mode disables the cache on the cards causing the I/O to the disks to be much slower than normal.  If you come to some other resolution I'd be very happy to hear it.

kodchiang · Answer

Disabling Patrol Read function resolved the issue in our configuration, too. I don't think we can dramatically improve PERC 4e/DC performance as long as it disables write-back cache in clustering mode - we may need to get rid of bus-based RAID controllers and migrate to external RAID box or SAN.

PowerVault

Performance issue with PERC4e/DC and PowerVault 220S

Was this post helpful?