This post is more than 5 years old
11 Posts
0
1069
CPU usage more than 100%? in isi job reports view
Hello Everybody!
some of our jobs consuming more than 100% of the CPU.
That's looks for me a bit confusing, cause I would expect a value from 0 - 100.
SnapshotDelete[79851] phase 1 (2016-02-18T14:21:01)
---------------------------------------------------
Elapsed time: 763 seconds
LINs deleted: 57474/57474
Snapshots deleted: 7/7
Snapshots truncated: 7/7
Errors: 0
Total blocks: 3189996 (25519968 KB)
Freed blocks: 3189996 (25519968 KB)
CPU usage: max 144% (dev 1), min 0% (dev 2), avg 81%
Virtual memory size: max 128680K (dev 1), min 127656K (dev 1), avg 127992K
Resident memory size: max 15300K (dev 1), min 12864K (dev 3), avg 14093K
Read: 280687 ops, 1750104064 bytes (1669.0M)
Write: 525454 ops, 2843091968 bytes (2711.4M)
Like here in this outpout I got a max CPU usage of 144%.
Is this because the isi job reports view shows me the sum from all 3nodes in the cluster?
And what does (dev 1) & (dev 2) stand for?
Thank for you help!
- Alexander
Peter_Sero
1.2K Posts
0
February 19th, 2016 05:00
The 144% from the example are a "max" value, observed on one node at one point in time.
That was 144%/4(cores) = 36% of the total CPU capacity of that node at that time.
It makes more sense to look at the averaged value, 81% in the example.
81%/4 ~ 20% load on each node, relative to one node's full CPU power.
As the 81% value was averaged over the cluster already, 20% is
also the average load of the cluster relative to the cluster's full CPU power.
Cheers
-- Peter
sluetze
300 Posts
1
February 18th, 2016 08:00
Hi Alex,
regarding the CPU value: if you have multiple cores the CPU value shows a equivalent to one core.
so 2 cores per CPU mean 200% load is possible (per CPU)
Peter_Sero
1.2K Posts
0
February 18th, 2016 08:00
CPU usage: 100% means 1 CPU-core, as with the top command,
which means it could go up to 800% on an 8-core machine
(as an example -- not that the job engine would actually do this).
The CPU and memory values are measured per node,
so it makes sense with min and max to show on which node they were observed:
"dev" means node or device "lifetime" ID, not the "logical" node number.
Both, logical numbers and node IDs are shown with:
isi status -n
hth
-- Peter
Alexander_Pursc
11 Posts
0
February 19th, 2016 05:00
Okay I understand now, that the CPU usage depends on how many CPU-cores I got.
The SnapshotDelete output (isi job reports view) is cluster wide – I’ll got the same results on every of my 3 nodes for this Job-ID.
Let's assume that you got three x200 Nodes which every node got an E5504 CPU - therefore 4 cores per node.
Should I now calculate CPU usage 144% divided by 12 (3nodes x 4cores per node)?