Friday, 1 May, 2009 2:37 PMWe upgraded the RAM in the headnode of our small 5 node cluster from 4GB to 8GB.The server is 64bit Windows Server 2008. HPCC 2008.Under node management it shows all 8GB of RAM, on the heat map it shows 8GB of RAM.However, when submitting a new job, under resource selection the server only shows as having 4GB of RAM. So, in theory if jobs were submitted to systems with less RAM or more RAM the metrics would be incorrect, because changes in RAM are not reflected there. Is there a way to update that information?
Friday, 1 May, 2009 3:37 PMHello Andy
Just so you don't think you're alone, I've checked this out on a dev cluster & see the same behaviour. Thinking about this a little, and I'm just guessing here, maybe the Node management list and heat map views query the node itself, but the resource selection options query the cluster database?
Have you tested whether a job complies with the new memory amount despite the resource selection report error?
Friday, 1 May, 2009 3:46 PMYes, the jobs seem happy with the additional RAM and it is being utilized. We are not currently submitting jobs where we are configuring them to run on systems with specific hardware types. I assume others might be, in which case a solution or workaround would be useful. Thanks for verifying the behavior.-Andy
Friday, 1 May, 2009 4:23 PMI agree this is useful to know.
I dug a little deeper and tried a few jobs configured to run on nodes which only have > xMB memory, and the submission did honour the actual memory available rather than the memory reported by the resource selection screen. So, it appears to be a reporting issue rather than one which affects submission operations...
Monday, 4 May, 2009 8:28 PMModeratorCould you try something and let me know how it goes?
after the job is done running on the node you updated the RAM on, can you take the node offline and then bring it back online ( from the management console) and tell me if that fixes the memory that is being reported on the admin console?
- Marked As Answer by parmita mehtaModerator Wednesday, 6 May, 2009 12:10 AM
Tuesday, 5 May, 2009 12:32 PMHi Parmita
Yes, an offline / online operation on the node seems to sort out this issue in my case.