Questions and thoughts w.r.t. reported CPU-related resources
Processors resource
So I'm a little confused about some things with regards to CPU resources, especially the way Fountain reports them.
I can see Configured/Processors being useful. And I could see Utilized/Processors and Available/Processors also being useful, if reasonably handled.
It seems to me, from what I've seen, that Fountain takes Utilized CPU * # of configured procs, rounds to an integer, and calls this the utilized processors. Is my understanding correct? And is this correct behavior?
It seems that "Available Processors" is relevant on the context of how many processors are allocated - if a job has 1 proc on a 2-proc node, and can share its node, then I'd think that the Processors resource should list 2 configured, 1 available - regardless of current CPU usage. Or is this already what's being done?
CPU utilization in multi-processor systems
Update: This is being handled in ticket #20.
Further, in the interest of administrative knowledge, would it be useful to add a mechanism for obtaining the CPU usage of each processor on a multi-proc node? Yeah, a node may have 75% CPU usage, but is that because the load is distributed, or is 1 of its 8 processors lying dormant? I, at least, could see that as useful. This information can be obtained, at least on Linux systems - see the sources for the wmsmpmon WindowMaker? dock applet if an example is needed.
This data could be modeled after the Network element, i.e.
<Processors>
<Processor>
<ProcessorNumber>0</ProcessorNumber>
<Utilization>0.730</Utilization>
</Processor>
</Processors>
I (Michael) do not believe that this should replace the existing Processors and CPU resources - those should remain, with current behavior for the CPU resource (Processors resource could use a little revamping). This element could appear alongside, just like the Network element used for the topography viewing.
Comment by samm on Fri Mar 24 18:55:33 2006
The CPU usage per processor is probably of no interest to Maui since AFAIK the process manager has no way to single out a specific processor to start a parallel job. Perhaps Brett can clarify further? However, you are correct that, in terms of administrivia, the CPU usage per processor would be of interest. It should not be too difficult to implement, the current usage information is gathered from the /proc file system. I'll see what it would take to split this information into individual usage per processor.
Comment by mekstran on Mon Jun 5 11:23:16 2006
Closing out - this has been implemented.
