28 octombrie 2010 09:35
I have a problem with a hpc cluster with 1 head node and 1 compute node. The OS is Windows HPC Server 2008 R2.
In cluster manager I go to heat map and I see the CPU usage on compute node continuously changing from 0 to someting about 30%. Every refresh shows different value and one of every 2 values is always 0, the other can be something >0 and usually close to 30. Remotely accessing the node I can't see anything in task manager performance view.
Someone told me this could be a marginal activity of system processes that task manager is not considering but let's be serious, 30% is not marginal in a 8 core system.
I tried to reimage the node and even to reinstall the head node system. The cluster seems to work ok, but this ghost activity is very strange and I really want to get rid of it.
Thanks in advance
26 martie 2011 00:19
Sorry for the delay.
I'm not sure what might be causing the spikes. In these case, if you're using Task Manager on the CN you might like to also try choosing the "Processes" tab; click the "Show processes from all users" button; and then sort the CPU column to see if any process is consistently showing a higher-than expected CPU usage.
BTW Which version of HPC Pack are you using?
7 iunie 2011 18:35Was this ever resolved? I have the exact same behavior on a cluster with 3 compute nodes.