[Hcoop-discuss] Unresponsive Server

[ntk at hcoop.net]

> rob at hcoop.net wrote:
>
>>Has any else noticed that fyodor becomes particularly unresponsive??
>>
>>It seems to happen most noticeably for me at night time (1AM - 4AM PST).
>>
>>
> This is the time when the back-up cron job runs, so that is almost
> certainly what is slowing things down.  As far as I know, it's already
> running with 'nice' (with no priority argument), but perhaps it a lower
> priority would be a good idea?  The job already takes hours to complete.
> :-(

I'm doubt this will help.  I think the niceness property basically
prioritizes CPU time, and the backup process hardly uses any.  The problem
is more likely that it is saturating network and/or disk bandwidth
(probably mostly the network) and starving out other requests.

I think it would be more effective to throttle the backup process's use of
the network, if that can be done easily.  Obviously this will
proportionally lengthen the backup duration, and it won't be any good if
we're running backups 24/7.

I think we might need to rethink the backup strategy.  How is it being
done now, dumping the whole filesystem tree?  If we could use rsync it
would be better, or if we had a multi-tiered backup schedule, backing up
certain portions on a weekly or less frequent basis and only important
dynamic stuff daily.

-Nathan