Subject: One vserver hogging down complete system
From: Markus Fischer <markus@fischer.name>
Date: Thu, 12 Jun 2008 10:26:12 +0200

Hi,

I've yet to support one vserver with a very old mail server installation 
still running which often brings down my whole system to it's knees, 
load rising up to 10 or 15 and rendering the system more or less unusable.

I'm running 2.6.22.19 with 2.2.0.7.

I tried to configure the vserver (mail01) to only use parts of the 
resources, but maybe I failed here.

The host system is a Primergy RX200 S3 Xeon Dual-Core DP with 4GB RAM 
(it's showing four CPUs)

mail01 has rlimits/rss.hard set to 131072 (512MB) and

# cat sched/fill-rate
1

# cat sched/interval
8

The idea is it to give the vserver an 1/8th of the RAM and an 1/8th of 
the available CPUs. We're having about 10 vserver running, most of them 
idle anyway, but some do cron jobs and to vast amount of bzip2ing.

The other server (mngr01) which runs a few cron/bzip2 jobs has the 
following rlimits:

# cat rlimits/rss.hard
524288
# cat rlimits/rss.soft
262144

and scheduling

# cat sched/fill-rate
1
# cat sched/interval
6

So it is given a bit more CPU time. The bzip2 jobs are running with 
niceness 19 already, but I get the impression that it doesn't change a 
thing.

bzip2 jobs always make the system very unresponsive the moment they 
start. Is it maybe just not possible to properly schedule them? We're 
using -9 compression to save disc space.

Thanks for any advice!

- Markus