UCX small buffer depletion

[Bill Hill_1]

In the past we had problems depleting UCX small buffers, then the problem mysteriously disappeared. Peak small buffer usage on one of our VMS servers has been 668 for over a month, with current usage fluctuating somewhere below that number. Last Thursday morning (I check this every morning) it was suddenly up to 1763 (max is 1800). I stopped and restarted UCX and by late afternoon small buffer usage was only up to 287. It became a problem again over the weekend causing the on-call person to stop and restart UCX several times. When I came in this morning I checked the status and small buffer usage was at 963 peak. That was about 8:15 am. Now small buffers are at 1047 peak and 1039 current. Most of those seem to be used for routing, 839. The server is running VMS ver. 6.2, we can't upgrade due to some applications that corporate runs on the partner on this dual host node pair. What does the "routes" in small buffer usage indicate to me? What causes the redirects and is this my real problem? Is there a configuration change I can make that will help alleviate this problem? Can the reappearance of this problem be due to network changes made to our WAN?

[Steve Nimr]

Bill,

In a former life the programmers shipped out a new version of the app. Small buffer useage went from 400 to 3500 immediately and steadily increased. I ended up writing a script that increased small buffers by 10% if the current value was within 10% of the max. It won't fix your problem but it could be a workaround.

[Bill Hill_1]

Similar "patch" to what I am working on. I intended to simply stop and restart UCX when small buffer usage reaches about 90% of max. That would cause brief interuptions for users so I may try your solution instead. Thanks!

[Bill Hill_1]

Steve, did you ever discover what the cause of the problem was at your facility? I suspect changes corporate in our WAN routing structure but don't know for sure.

[Steve Nimr]

Bill,

The project was an electronic toll system. The developers rolled out an almost totaly re-written application and had changed how the incoming reads were processed. Of course they never conferred with the sysadmin or gave any heads ups notice.

Steve

[Ian Miller. · Accepted solution]

I guess this is UCX V4.2 or similar?

Try setting pmtu_enabled = 0
you can do this in UCX$SYSTARTUP.COM
Parhaps by
UCX SET PROTOCOL command or
syconfig -r inet pmtu_enabled=0

with it enabled (=1) you can get a routing entry for every client.

[Bill Hill_1]

Ian,
Yes, it is UCX 4.2. This sounds like it could be the culprit. How can I see what the current value of "pmtu_enabled" is? I can't find this setting anywhere in UCX?

[Ian Miller.]

with 4.2 I don't know if pmtu is enabled.
Are there lots of routes shown by
UCX SHOW ROUTE
and is dynamic routing (RIP) enabled?

[Steve Nimr]

Bill,

Another thing you can do is modify the sizes of the buffers. I don't have UCX here so this is from memory. But you can set the small buffers to include packets of size n to o, medium buffers from p to q, large buffers r to s. Doing that you could shift the burden from small buffers.

Steve

[Ian Miller.]

Re-reading your question I notice you mention redirects and network changes on your WAN.
Does this mean that your route table is getting lots of dynamically created host routes?
Can you look at the flags on the route table entries shown by SHOW ROUTE and say how many are dynamic and how many are hosts. Since the WAN changes have you modified the static routes on your VMS system to correspond with the changed WAN?