[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [HTCondor-users] swap space

I only started monitoring thoroughly the swap usage from yesterday.

Our ganglia tells us about the swap usage, but doesn't say which process is consuming that.

Since yesterday I am monitoring that, and I see that the tomcat process (which runs the CREAM-CE) consumes 16G RAM, by itself, but no swap space.

I understand the condor_schedd and condor_collector are consuming swap space because they don't have enough memory left.


MÃtyÃs Selmeci <matyas@xxxxxxxxxxx> escribiÃ:

Is that memory usage normal? The CHTC central manager
(collector+negotiator) only has 8G of RAM and 16G of swap, and swap is
barely used at all.


On 11/20/18 10:15 AM, Almudena Montiel wrote:
Thanks Dimitri

Yes, in the worker nodes we do have 3G per core in average, some
nodes have 4G per core.

I was asking for requirements of the server where the htcondor
central daemons are running. We have all the central services of
htcondor running in a single server. I was referring to this server
in my previous mails.

Thanks and regards,


On 20/11/18 17:05, Dmitri Maziuk via HTCondor-users wrote:
On Tue, 20 Nov 2018 14:34:09 +0100
Almudena Montiel <almudena.montiel@xxxxxx> wrote:

As for SWAP, we have increased it to 48G. Hopefully that will be
Well the longer answer is if you have to swap, your performance
will go down the drain even if you gon't get as far as thrashing.
E.g. we run BLAST, it memory-maps its search database in ~2GB
chunks, we tried running it with less than 2GB RAM per slot, it
went to swap and the jobs took days to complete -- too long to suit
our requirements. So all our workers have 2GB RAM per core (slot)
or more now.

Almudena Montiel Gonzalez              e-mail: almudena.montiel@xxxxxx
Systems Manager
Dept. Theoretical Physics. Block 15.
Laboratory of High Energy Physics
Universidad Autonoma de Madrid.
Phone: 34 91 497 4541      Fax: 34 91 497 3936
James Watt 2, Cantoblanco, 28049 Madrid, Spain.