SCSI resets on SC10 disks

[Mark Ferraretto]

Hello,

I have a cluster of L-class boxes. We have a couple of SC10 disk arrays attached. We're running HP-UX 11.0 and MC/ServiceGuard. We keep on receiving one or two SCSI reset messages every day in the syslog file. We can't tie them to a specific time or after a specific command - they seem to be random. We've had these for about 4 months now. They don't seem to affect the system's performance but the system isn't live yet so it's hard to tell.

We replaced the SC10 backplanes, the BCC's, we installed self-terminating SCSI cables (the black ones - they helped a lot but we're still getting the messages), SCSI controllers. But we still get these resets. I'm concerned that they will impact on the I/O performance once the system goes live as it will be heavily loaded.

Incidentailly, an ioscan works fine.

Anyone else getting these? I installed another cluster on D's with storage enclosures and I don't get the resets. I'm wondering if it's an SC10 problem. Or if it's a problem at all.

[Bill McNAMARA_1]

Can you post up the info about the dial switch settings on your SC10 and the ioscan output.

As you mentioned SCSI is very touchy about correct termination, make sure the terminator is correct for the SC10.

An swlist or a what /stand/vmunix should show the SCSI driver revision, which should be updated to latest level. I do recall fixes in scsi reset notification.

I have often seen these errors appear on my system, typically related to power and/or lvm but rarely hardware defects.

When was the last time you patched your systems, and what firmware is on the SC10 bccs?
mstm - tools - info - run.

You can find patch info here on the itrc.

Later;
Bill

[Bob Inglis]

The D-series and JAMAICA's use fixed termination.

The N & L series have Auto Terminating Host bus adapters. In a service Guard configuration Auto-Termination MUST be turned off and self-terminating cables MUST be used. If two auto-terminating interfaces are attached to the same SCSI bus they get confused about whether to terminate or not.

Also make sure the Host bus adapters are set to different SCSI addresses (7 & 6 for example) I've seen a lot of people forget to do this.

On Fast wide differential, if you forget it won't even boot, but with LVD and the SC10 it will boot and then you get resets as the two interfaces contend for ownership of the bus.

While you are turning Auto-termination OFF and checking the HAB?s SCSI id, check the interface speed and make sure it?s set to UNLIMITED (for the SC-10)

[geeta]

For Bob,

How would be check if Auto-termination is turned off? I am experiencing a similar problem.

Thanx :)