Added FDDI link as backup

[Zeni B. Schleter]

We have a OpenVMS cluster (V7.3-2) over GigE network. We added links for an FDDI ring that is on a separate network for a backup cluster connection. It appeared to function. There now appears to be a problem with jobs that are disk I/O intensive taking much longer than usual. The change appears to roughly coincide with the addition of the ring.

I was going to try to use the LAVC$Stop_Bus on the FDDI device but am hesitating for the warning about a cluster crash. Several years ago we did use the code on a different host without any problems. Anyone know of a reason that I should not use that utility?
It is in Sys$examples.

Another question (bigger question) is why is it using the FDDI? The GigE should be much faster . From the HP Insights Web Interface the FDDI is Half duplex which does not seem correct but the ring was added without shutting down the cluster. That may be another problem.

Thanks for any insights...

[Volker Halle · Accepted solution]

Zeni,

nowadays, you would want to use MC SCACP STOP LAN_DEVICE FWA instead of LAVC$STOP_BUS. Just make sure, you don't stop the only remaining cluster interconnect.

You can also lower the priority of SCS communications via FDDI (SCACP> SET LAN/PRIO=n FWA).

Use SCACP> SHOW CHAN to find out the load classes, packets transmitted etc. via the various channels. SCACP (and SDA> PE ...) offer lots of information about cluster communications...

Volker.

[Zeni B. Schleter]

MCR SCACP show 3 lan devices, FDDI, GigE and local. FDDI was added March 8th at 16:08. Unless there is round over, the FDDI shows twice the packets (S+R) as the GigE. Both have priority of 0. Volker, you suggest just changing the priority . The help say that the priority can be a range of -127 to +127. So should I set it to minus for the FDDI or set the GigE to have a positive.

The load class shows the FDDI as 100 and the GigE as 1000. Why should it use Fddi if GigE is available?

[Volker Halle]

Zeni,

try using priority -2 for FDDI. Maybe FDDI is preferred due to a larger buffer size ?

Could you provide your SCACP> SHOW CHAN output in an attached .TXT file ?

Volker.

[Zeni B. Schleter]

I changed the priority of FWA on one member of the cluster to -2. The packets now appear to be split between the two. The delay for the GigE is much greater than that of the FDDI but that appears to be very dynamic and after a couple of iterations .

Can't post what I am seeing without typing it in. But the GigE device does appear to be getting more of the traffic now.

I have started a job that I noticed had a very extended runs after the FDDI addition. It went from 30 minutes to 2 hours. If it is back to normal I will consider success.

[Zeni B. Schleter]

Success. Traffic over FDDI is much reduced and job completed in a timely manner when compared to pre-FDDI configuration. Setting the Priority with the SCACP utility was a definite solution.
Volker,
Many thanks for the very timely response.

[Volker Halle]

Zeni,

don't forget to add the SCACP command into the system startup. This is not a permanent change of the LAN_DEVICE priority and needs to be re-done after booting.

Volker.

[Zeni B. Schleter]

Thanks, Volker. I had wondered about that but had already closed the thread . I will add it to the startup procedures today.