system won't boot

[Brian Street_1]

Hello,

I have 2 N-class systems that I'm trying to set up to eventually become a SG cluster. I have the systems connected as shown in the attached diagram. The SC10s are set up in full-bus mode.

The SC10s have a home (mirrored across controllers on system 1) and local opt filesystems. I want to be able to share the home filesystem with system 2 when I create the SG.

Right now, the first system is up and operational (I'm wondering if what will happen if I reboot though) and the second system will not boot - it panics. My first guess is the cabling is not correct or a setup problem with the SC10.

Any ideas?

Thanks,
Brian Street.

[John Poff]

Hi Brian,

It sounds like there might be a SCSI problem on the second machine. I'd try narrowing it down by unplugging one of the SCSI cards on the second box, terminating it, and trying to boot. If it boots, you'll know the problem is in that chain. If not, plug it in and try unplugging the other one.

JP

[Patrick Wallek]

If you have the SC10s plugged into both machines, then I believe that you will need to change the SCSI-ID on the controllers on one machine so that you don't have 2 devices on the same SCSI chain with the same SCSI ID.

Also, I would not try to have a VG mounted on BOTH machines at the same time. You are really asking for trouble there. In a SG cluster, I don't know that I'd do NFS either. I'd be tempted to just let each machine have it's own /home and sync them up occasionally.

[Steven E. Protter]

Two data collection steps.

ioscan -fnC on both machines, compare results.

dmesg on the machine that acting up.

mstm xstm(X only) exercize all of the scisi devices of the machine.

There could be a cable problem, or a drive cage issue.

SEP

[Steven E. Protter]

sorry, use ioscan -fnC disk or regular old ioscan to check the buses and cards.

SEP

[Michael Steele_2]

You can't share that file systems that are mounted at the same time with LVM in a SG cluster unless you use NFS. Is this /opt you're referring to? Like /opt/local/xyz? And /home?

You can export that sub directory via NFS but think twice about exporting /home via NFS. Its easier to sync them up with a rcp cron every night. Just do a remsh diff and fire off a report.

vg00 is unique to either node and not shared since it would be a SPOF.

Only the cluster lock disk is shared between the two nodes and the cluster binaries control that.

For your cabling, system a is on SCSI target 7, system b is on SCSI target 6, disk take up the remaining targets 5 down to 1 and then 15 down to 8.

You can't have system a and system b both on target 7.

And your disk devices should look like this:
c0t1d0, c0t2d0, c0t3d0 for system a
c1t1d0, c1t2d0, c1t3d0 for system b.

Vg01 is on c0t2d0 and c0t3d0 on system a until it fails over to c1t2d0 and c1t3d0 on system b.

See attached and refer to figure 4-1 and page 72:

http://docs.hp.com/hpux/pdf/B3936-90024.pdf

[Brian Street_1]

Hello everyone,

Thanks for your suggestions; I'll try them all this morning and see how I fare.

The /home filesystem is not really a home filesystem as we know it but a central filesystem where all of the development work is stored. Perhaps it might be better if it was called something else instead - that was decided by others.

I will have a local opt filesystem on the SC10s - one for each system and on different SC10s - and the home filesystem - mirrored across the SC10s - that I want to failover to the second system.

Thanks again for your input; hopefully I'll find the problem quickly this morning.

Brian.

[Michael Steele_2]

Hi Everyone:

Regarding my posting above, this comment could easily be confusing: "...Vg01 is on c0t2d0 and c0t3d0 on system a until it fails over to c1t2d0 and c1t3d0 on system b..."

Let me re-phrase this.

Theses disks will be seen from both system A and system B simultaneously via ioscan.

???Piggy backing??? system???s A SCSI controller atop of system???s B SCSI controller enables disks c0tYdZ and c1tYdZ to reside within vg01 in an exclusively locked state, and only active to one node at a time via LVM, the SG cluster and the enabled package.

[Brian Street_1]

hello everyone,

I've tried to change the scsi id on one of the controllers and have had no luck.

I thought I had to stop the boot, go into service mode and use the scsi init command. The problem is that I only see the internal scsi controllers, and not the pci slot controllers.

any idea how to get there? mfg mode?

Brian.

[Michael Steele_2]

Need the model number. Some of these cards will only work in slots marked twin turbo. Twin turbo is a dedicated bus while the others share a bus. Here are some possibilities

A5149A - Ultra SCSI's
A4800A - FWD SCSI-2's PCI

When installing these cards in an N 4000 it???s important to feel the card SNAP into the slot. Simply inserting and aligning the card isn't enough.

Also, those grey plastic inserts are really there to guide your aim into a vacant slot and lock the card into place. You pull out about half an inch for this. So first pull out the plastic insert to unclamp the vacant slot, and then push in the card until you feel it SNAP into place, (* you???ll use more strength than you feel you should *) followed by pushing in the plastic insert which locks/clamps the card into the slot.

I've attached a WORD doc from and A500 that shows what a twin turbo looks like.

Identify your card, determine if it needs twin turbo.

Your objective is to see it via ioscan -fnkC ext_bus, else, ???..