Why are all my work units "waiting for shared memory" since about two-three weeks?

[SilntObsvr]

For the past two or three weeks (roughly since BOINC Manager 8.2.4 dropped, though I have no evidence this is related) I've been seeing "waiting for shared memory" on every task, and my user averages are dropping toward zero.

I'm running Debian 12.11 on an AMD fx8350 with 32 GB RAM, nVidia RTx2070 and 535 drivers, multiple terabytes of storage with 50 GB dedicated for BOINC tasks.

I run MilkyWay on CPU only, and Einstein on GPU (it requires a fractional CPU for each task, too).  I'm used to seeing 8 CPU worth of Milkyway tasks running simultaneously, and one GPU running Einstein.

The only significant software I've installed semi-recently was Chromium (because Firefox was misbehaving on Foundry Virtual Tabletop), but that was six weeks ago or more, and BOINC ran normally until at most about three weeks ago.

I've tried tweaking various settings as suggested on the Einstein and BOINC forums -- without any improvement.  What can I do?

[David P. Anderson]

The most likely thing is that your system has reached the kernel-defined limit

on the number of shared memory segments.

You can view this limit in /proc/sys/kernel/shmmni

and see the list of segments with the command 'ipcs'.

If this is indeed the problem you can raise the limit.

If that doesn't work, please create a github issue.

- David

--
You received this message because you are subscribed to the Google Groups "boinc_alpha" group.
To unsubscribe from this group and stop receiving emails from it, send an email to boinc_alpha...@ssl.berkeley.edu.
To view this discussion visit https://groups.google.com/a/ssl.berkeley.edu/d/msgid/boinc_alpha/b66ea23e-6303-44d6-97bf-2fac97dfedban%40ssl.berkeley.edu.

[Michael E.]

Two things:

1. SilntObsvr - You have hit a resource limit of some kind. Not sure of the cause but running fewer BOINC tasks might help. Others can likely help more than myself (not a Linux expert). 

2. In general, we have been testing only CPU tasks in BOINC Central. But as Georges pointed out, we should also test GPU tasks as well. I know Georges pointed this need out for Apple Mac silicon GPUs, but is it also needed to test podman with other GPU types? Or is that Phase 2 of BOINC Alpha test?

--

[SilntObsvr]

Well, in a bizarre twist...

I aborted all the work units waiting to be resumed (some up to a couple weeks old, likely expired or close to expiration) and with computing preferences settings nearly identical to what I've used for years (might have raised or lowered one thing or another by a few percent), I told both projects to "Update" -- and when they downloaded new work units, one Einstein and one MilkyWay work unit started crunching, just like nothing had happened.

I can only presume that one project or the other had bitten down on a work unit that for some reason hogged all available resources, and aborting the work units (or updating the projects -- not something I usually have to do manually, AFAIK) fixed the problem.