topic Re: ProCurve 2708 and Gigabit NIC stalling in Operating System - OpenVMS

ProCurve 2708 and Gigabit NIC stalling

Sebastian Bazley — Thu, 23 Mar 2006 08:05:53 GMT

2-node Cluster (ES45+DS25) running VMS 7.3-2

The cluster is connected using:
Gigabit NIC in each
ProCurve 2708 switch

We've been unable to get this working properly - it only seems to work for a very short while and then stalls.

Has anyone else had any problems configuring this?

We're hoping to try the kit out on a test system in the UK shortly.

In the meantime, any advice/experience would be most welcome...

Re: ProCurve 2708 and Gigabit NIC stalling

Volker Halle — Thu, 23 Mar 2006 08:11:43 GMT

Sebastion,

what do you mean with 'stalls' ?

Does the cluster hang ? Any messages on the console regarding the LAN driver and possible duplex mismatch ? Does auto-negotiation work or are both (LAN NIC and switch-port) set to the same speed and duplex mode ?

Volker.

Re: ProCurve 2708 and Gigabit NIC stalling

Sebastian Bazley — Thu, 23 Mar 2006 08:37:56 GMT

Cluster hangs.
No messages.
We have set auto on both consoles.
The switch defaults to auto and has no management interface.

It did work once for 10 minutes, and then it hung.

Re: ProCurve 2708 and Gigabit NIC stalling

Volker Halle — Thu, 23 Mar 2006 08:47:00 GMT

Sebastian,

the systems in the cluster constantly communicate via all LAN interfaces by sending SCS HELLO multicast messages every 3 seconds. Each system in the cluster would output a 'Lost connection to node' message, if it does not receive a HELLO message from the other node within about 9 seconds.

Is BROADCAST enabled on the console terminals ? $ SHOW BROADCAST ? If not, you have to enable it, to see those messages !

You could force a crash, if the cluster seems to hang: Press HALT button, then >>> CRASH

In the crash, you could check LAN counters etc. How do you determine, that the cluster 'hangs' ?

Volker.

Re: ProCurve 2708 and Gigabit NIC stalling

Sebastian Bazley — Thu, 23 Mar 2006 09:00:46 GMT

Sorry, I should have said that the cluster was working fine before the new NICs and switch were installed.

[I believe they have now reverted to the previous hardware.]

It seems to be a problem that is specific to the particular switch and NICs.

Unfortunately, we can't run any more tests until we get access to the test systems in the UK.

In the meantime, we were hoping that someone might have come across a similar problem with that specific hardware.

Re: ProCurve 2708 and Gigabit NIC stalling

Allan Bowman — Thu, 23 Mar 2006 09:30:11 GMT

You might want to disable the switch auto-negotiation and set both ports manually. Be sure to verify your network paths if you have a second NIC on either node - if STP is enabled on the switch, it could eventually shut down your expected path.

Allan in Atlanta

Re: ProCurve 2708 and Gigabit NIC stalling

Sebastian Bazley — Thu, 23 Mar 2006 11:11:11 GMT

As far as I know, the switch does not provide the facility to change anything...

HP describe it as:

Unmanaged: provides plug-and-play simplicity

Re: ProCurve 2708 and Gigabit NIC stalling

David B Sneddon — Fri, 24 Mar 2006 02:51:01 GMT

Should that not be...

Unmanaged: plug-and-pray stupidity

Re: ProCurve 2708 and Gigabit NIC stalling

Colin Butcher — Fri, 24 Mar 2006 04:18:49 GMT

I think you've just discovered why I don't like unmanaged switches (especially in production environments) - you can't get any information out of them, or set them up as you want to.

Which NICs are you using? Assuming they're copper then have you got the appropriate specification cables and distances for GigE over copper?

I generally use fibre and managed switches - less problems on the whole. After all, it's not cheaper to use anything else if it wastes your time and creates unreliability issues. The last thing you want in a production environment is something that doesn't work consistently and reliably.

If the systems really matter then you may want to consider dual NICs and dual switches so that you have two cluster interconnect paths. Don't forget to enable / disable the releveant protocols on all the NICs in the systems either.

Hope this help,
Cheers, Colin.

Re: ProCurve 2708 and Gigabit NIC stalling

Peter Zeiszler — Fri, 24 Mar 2006 10:23:24 GMT

Can you force the NICs to 1000 Full and have the switch then autosense?

We have had issues before when having to use NICs and Switches both in autosense. They both keep changing speeds and duplex which results in dropped packets and NO communication.

Re: ProCurve 2708 and Gigabit NIC stalling

Art Wiens — Fri, 24 Mar 2006 13:38:21 GMT

Theoretically there is no such thing as 1000/half afaik. Auto negotiation of a gig port "shouldn't" be difficult there "should" only be one choice.

Someone at HP should be able to confirm that a 2708 is (or is not) compatible with the nic's in your Alpha's.

Art

Re: ProCurve 2708 and Gigabit NIC stalling

Colin Butcher — Sat, 25 Mar 2006 05:51:58 GMT

In my experience GigE is best left as "autonegotiate" at each end of the link. In practice it should always come up as 1000FDX as GigE is implemented as a point-to-point link between a NIC and a switch port.

I've discovered that some switches using mini-GBICs will not always detect the failure of a single fibre in the pair if the switch is not in "autonegotiate" mode. That can lead to the switch believing that the link to the NIC is still up - which can then create other problems elsewhere.

The OpenVMS documentation section on LANCP may be useful: http://h71000.www7.hp.com/doc/82FINAL/6048/6048pro_033.html