I'm running the commands:
sinfo -o %C
and
scontrol --oneliner show nodes
and trying to sum up numbers in the second command to make the numbers
in the first command. I'm doing this mainly for my understanding of what
each number in the first command *really* means.
The easy ones are:
total CPUs: in the first command it's the 4th number
in the second command it's the sum of all the 'CPUTot' values
allocated CPUs: in the first command it's the 1st number
in the second command it's the sum of all the 'CPUAlloc' values
But the other two are not so easy:
other CPUs: in the first command it's the 3rd number
in the second command it's *almost* the sum of the 'CPUTot'
values
for all nodes in state DOWN, DOWN* or IDLE+DRAIN
idle CPUs: in the first command it's the 2nd number
in the second command it's *almost* the sum of the 'CPUTot'
values
for all nodes in state IDLE
The cluster I'm looking at also has some nodes in state MIXED and I'm
wondering if, if I could break their CPUTots down then I'd get all
my sums to balance.
So my questions are:
1) would adding the broken down MIXED CPU counts make the sums balance?
2a) if no, then what am I missing?
2b) if yes, then how can I break down the MIXED CPU counts, preferably using
'scontrol show nodes' with some extra options?
I'm not sure if this is relevant, but I have some hidden partitions and
some nodes that are (temporarily) not allocated to any partition. I mention
this because I see 'scontrol' has a '--hide' option.
Also, we have 'Shared=EXCLUSIVE' for all partitions, which should make
any formulae simpler, but I'd be interested in more general formulae.
Advice sought, many thanks!
Alexis
