Re: [savannah-help-public] Slowness of Savannah in the morning

[Bob Proulx]

Karl Berry wrote:
> I added an iotop invocation to vcs:/etc/cron.d/sv, at the same time as
> top, to see if it gets us anything useful.  Also nice-d all the relevant
> cron jobs.

Thanks.  I think your theory about being I/O bound is probably
correct.  But in that case I am not sure what to do about it.

> There was a load average of 10+ on both Saturday and Sunday from 0700
> though 0800 (more or less), so clearly the backup job change was not a
> panacea (as expected).  /var/tmp/toplog shows the usual array of git,
> bzr, cvs, etc., processes, mostly niced, mostly in disk wait.  The
> backup job doesn't even show up any more.

Perhaps it is time to ask for another VM to split the workload?
Because it would be nice to be able to do A-B comparisons between good
and bad until able to zero in on the problem.

If we started moving tasks from the current VM onto a different one
then eventually we would move the "heavy" one, the problem would
follow it, and then point to that task as being the heavy problem one
that needs more power.  I think all of the tasks are tied to their
name.  That is git.sv.gnu.org and bzr.sv.gnu.org and so forth.  It
should be trivial to move individual services from one place to
another.

However if the host system is I/O bound and all of the VMs are on the
same host system then I doubt this would change anything.  But it
might prove that the problem isn't in the VM but in the host system.
I am somewhat grasping at things because there just doesn't seem to be
any obvious pointer to the problem.

> Sigh.

Agreed.

Bob