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

Re: [HTCondor-users] Recommended hardware specifications for a condor submit node



Memory is the biggest thing.  I tend to use a rule of thumb that I have 8GB of RAM for every 2000 slots that I am handling.
A clean and fast network interface is important too.
Remember also that the real load on condor is not how many jobs are running, but how many are starting and stopping.
I have two clusters of similar capacity ~5000 slots  and identical head node hardware but one head node is much more heavily loaded than the other because the mean length of job is shorter.
If you are going much above 6000 slots then I would strongly consider an SSD-based device to handle the spool area.

Steve Timm


-----Original Message-----
From: HTCondor-users [mailto:htcondor-users-bounces@xxxxxxxxxxx] On Behalf Of Dimitri Maziuk
Sent: Monday, August 12, 2013 3:43 PM
To: htcondor-users@xxxxxxxxxxx
Subject: Re: [HTCondor-users] Recommended hardware specifications for a condor submit node

On 08/12/2013 03:09 PM, Colin Dewey wrote:
> Hi,
> 
> Are there any general recommendations for the hardware specs of
machines intended to be condor submit nodes, particularly those that will be handling thousands of jobs? I'm assuming that many cores are desired, but am not sure if there are other specs that are relevant.

If the jobs generate lots of output that gets sent back, bandwidth is relevant too. And if you're on 10Gb/s, possibly disk write speeds.

--
Dimitri Maziuk
Programmer/sysadmin
BioMagResBank, UW-Madison -- http://www.bmrb.wisc.edu