Problems with W2K3 and SQL2K in MS cluster

[DragomirAtea]

Hi folks!

I've run into a veird problem at a customersite...

Here comes the HW setup.

Servers
2 DL740 G2 each server configured as follows.
8 * 2,7GHz 2MB Cache CPU
32 GB RAM (Usable)
2 * 72 GB (internal RAID1) for OS
2 * FCA 2408
Windows 2003 Enterprise
Microsoft SQL 2000 Enterprise
Configured in a Active/Passive MS Cluster

Storage
HP EVA 3000 2C4D (VCS 3.020)
ZDBDiskgroup (16 * 146GB)
DataDiskgroup (24 * 300GB)
2 * HP StorageWorks 2/16 EL switches
SecurePath 4.0C SP1
BusinessCopy 2.3

All vdisks presented is VRAID 1

Now to my problem...

Lets say that nod1 is active, it holds all cluster resources. everything works just fine, no performance troubles what so ever...

But!!!

If I power down the passive node (node2) the performance on the active node drops lika a rock. I can barely log on to it and the users complains about responstimes.

How the h*ll can a powerdown of the passive node effect the active node like that???

Any ideas, because I'm lost...

Regards
Dragomir Zekic

[Andrzej Kowalik]

Hmm, first could you explain what kind of performance problem do you have? Disks? Network? Processors?
I installed similar configuration without problems. Configuration differences - I used FCA2214 and MPIO drivers.
I had little problem with NICs - sometimes I lost one "external" NIC from team member, but new drivers resolves that.
I can suspect FCA drivers - could you update it on "bad" node? Even if all looks good.
Supported version with EVA3K:
HP:FCA2408 SCSI mini-port DriverSet1 - Bios: 1.63a1;
Driver: 5-4.82a16; FW: 1.01a2
HP:FCA2408 SCSI mini-port DriverSet2 - Bios: 1.70a1;
Driver: 5-5.10a10; Driver: 5-5.10a9; FW: 1.81a2
HP:FCA2408 StorPort DriverSet1 - Bios: 1.70a1; Driver:
5-1.02a5; FW: 1.81a2

Other possible reason: maybe you have wrong configured FC zones?

regards,
Andrzej

[DragomirAtea]

The performance problems i got is mainly CPU.

On the FCA:s I got the latest drivers and FW.

The zones i've configured have been verified by HP, but here goes...

I got four zones in my config.

Managementzone (SMA and HSVports)
Backupzone ( NSR and relevant FCA on hosts)
Clusterzone (clusterports and HSV ports)
Defaultzone (Other attached host that's not clustered, all windowshosts)

What do you think?

[Andrzej Kowalik]

Regarding processor performance - if it happens in all cases, it can be cache which address processors. If this cache is damage server still work but slower and increase processor utilization.
Regarding zone configuration - first of all in zone should be only one initiator (for example: one HBA and two EVA ports - one from each controller). Otherwise, if you have more initiator, failing one of them can cause restart other host or SAN disk disconnecting.
Second main rule - you cannot mix in one zone different device type (disks and tapes) or even disks from different arrays type (for example EVA and MSA).
In your environment - you should create more zones for clustered servers and different zones for other windows hosts. I know that, such solution produce "million" zones, but is much more secure and reliable.
Other thing that you could check is Fibre Array Information Agent. Should be disabled.
But I don't think that this could be main processors utilization reason.
regards,
Andrzej

[DragomirAtea]

I dont understand how cache problems should affect the active nod when i power down the passive nod...

I repeat, there is no failover...