Hi all ,
I'm using Windows HPC 2008 R2 Sp4 .
We have a client software that submits jobs to Windows HPC via C# API's .
Most of the jobs are paramtric sweep - each job can contain 50 to 400 tasks .
Latetly we saw somthing strange - all the tasks must end the calculation time at the same time (40-50 sec) - but a few tasks finish thier jobs 10 - 20 sec later .
After fast investigation we saw that one of the nodes cores (Core#15) 20%-30% busy when the compute node is idle .
When that core computes the task it finish 10 - 20 sec later than each other core .
I have the following questions :
- What make that core busy ?
- Why on all the nodes it core #15 ?
- Can I set my tasks affinity not to use core #15 ?
- Can I make the scheduler not to allocate taks for that core ?
Thanking you in advance ,
Shai .
