We are currently on HPC server version is 3.2.3716.0, which I believed is MS HPC Pack 2008 R2 with an integrated SP2. Please correct me if I’m wrong.
We plan to add new nodes to our current HPC environment – comprising 3 HP generation blade server models. I have noticed that HPC appears to always select the server_01 first if the server01
cores are not utilized.
For example, I have 30 nodes and have jobs scheduled to run at various times. There could be multiple scheduled jobs running but if the jobs have completed their runs within the
hour, and had returned all the cores back to the HPC grid, the next jobs will utilize the same server_01 first – again. The latter servers (e.g., server_20-30), which are newer and faster servers, are hardly utilized! Is there a way to set the node usage
to set as round-robin or trick it to use the latter server first? As mentioned, we plan to add new nodes - with the latest HP G8 blade servers. It’ll be nice to have the HPC selects the faster server/CPU first.
Do your newer servers have more memory/cores than old ones? If so, you can use "prefer nodes with more memory" option when scheduling your job. You may have to upgrade to SP3 to use this feature if you can't find it in SP2.