On 11/23/2010 01:25 PM, Sim IJskes - QCG wrote:
Looks like a totally saturated disk subsystem (to me, from a distance). Is there sar data from that machine?
When you have a jobs that prints a timestamp just before the archive stage, you can correlate this with the sar data. You just have to look at the runq size or the waiting/blocked percentage to see if the I/O was the culprit. Is there idle=0% or very low, then its CPU bound.
However, you have to find out how the work is divided between slave and master, maybe all the data is first sent to the master, and archived there.
Gr Sim
