Discussion:
[gridengine users] what is IO in qstat
Lars van der bijl
2013-03-20 14:58:35 UTC
Permalink
hey everyone,

a few weeks ago we where having issue with users submitting jobs the did
massive IO to the file system. each task was writing out about 4 GB of data
and reading in about 2GB. as this was happening in parallel from our farm
it brought our server to it knee's and cause people to moan like crazy.

I'm I right in thinking that the IO output of a job would help me detect
the tasks disk IO?
I can't seem to find any good information on what the qstat IO output
means. I remember seeing that it has something to do with Gb/s ?

any help would be appreciated.

Lars
Reuti
2013-03-20 19:15:37 UTC
Permalink
Hi,
Post by Lars van der bijl
hey everyone,
a few weeks ago we where having issue with users submitting jobs the did massive IO to the file system. each task was writing out about 4 GB of data and reading in about 2GB. as this was happening in parallel from our farm it brought our server to it knee's and cause people to moan like crazy.
I'm I right in thinking that the IO output of a job would help me detect the tasks disk IO?
I can't seem to find any good information on what the qstat IO output means. I remember seeing that it has something to do with Gb/s ?
In case you refer to the field which is also listed in `qacct -j <jobid>`: `man accouting`

http://arc.liv.ac.uk/SGE/htmlman/htmlman5/accounting.html

-- Reuti
Post by Lars van der bijl
any help would be appreciated.
Lars
_______________________________________________
users mailing list
https://gridengine.org/mailman/listinfo/users
Dave Love
2013-03-24 21:50:02 UTC
Permalink
Post by Reuti
Hi,
Post by Lars van der bijl
hey everyone,
a few weeks ago we where having issue with users submitting jobs the
did massive IO to the file system. each task was writing out about 4
GB of data and reading in about 2GB. as this was happening in
parallel from our farm it brought our server to it knee's and cause
people to moan like crazy.
Ensure the (network file?) server is tuned properly. We cope with array
jobs doing that sort of thing to NFS.
Post by Reuti
Post by Lars van der bijl
I'm I right in thinking that the IO output of a job would help me
detect the tasks disk IO?
Local or networked?
Post by Reuti
Post by Lars van der bijl
I can't seem to find any good information on what the qstat IO output
means. I remember seeing that it has something to do with Gb/s ?
It should be labelled "GB".
Post by Reuti
In case you refer to the field which is also listed in `qacct -j
[I.e. cross-referenced from qstat(1). It probably needs better
explanation of other bits of output.]
Post by Reuti
`man accouting`
http://arc.liv.ac.uk/SGE/htmlman/htmlman5/accounting.html
--
Community Grid Engine: http://arc.liv.ac.uk/SGE/
Loading...