[Linux-cluster] Too large load on the login node

[Nicolas Ferré]

Steven Whitehouse a écrit :
> Hi,
> 
> On Thu, 2009-10-01 at 11:36 +0200, Nicolas Ferré wrote:
>> Hi,
>>
>> We recently installed a new cluster composed of 1 login node and several 
>> computing nodes running CentOS. These nodes share a GFS2 fs made of two 
>> partitions.
>>
>> A strange thing is that on each node, the activity load (as monitored by 
>> the 'top' command) is always larger than 1. After some googling, it 
>> seems someone already reported this problem but I can't see any solution.
>> Moreover, on the login node, the load is even larger:
>> top - 11:34:17 up 1 day, 22:07,  1 user,  load average: 16.19, 16.20, 16.12
>> while there is no cpu-intensive running processes.
>>
>> Do you have an explanation?
> 
> Processes in uninterruptible sleep are counted in the load average. One
> of the gfs2 daemons (in early versions) was set to sleep in this way. I
> suggest that you should upgrade to a more recent version (simply because
> a number of other bugs have been fixed since then) although the
> uninterruptible sleep is harmless aside from its effect on the load
> average,
> 

As far as I know, our system is up-to-date.
 > uname -a
Linux slater.up.univ-mrs.fr 2.6.18-164.el5 #1 SMP Thu Sep 3 03:28:30 EDT 
2009 x86_64 x86_64 x86_64 GNU/Linu

 > rpm -qa|grep gfs
kmod-gfs-0.1.31-3.el5_3.1
gfs2-utils-0.1.62-1.el5
gfs-utils-0.1.18-1.el5