[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Condor-users] Amount of ports used by condor
- Date: Mon, 13 Mar 2006 15:12:51 +0800
- From: <Greg.Hitchen@xxxxxxxx>
- Subject: Re: [Condor-users] Amount of ports used by condor
Thanks for the replies everyone.
The "problem" to do with the amount of ports used by condor
was not a problem at all. I was misinterpreting error messages
in the log files of the type:
failed to bind to a port in the range ****
The problem was in fact caused by the condor service starting
before the PC had grabbed an IP address via DHCP. We seemed to
worked around this OK by making the condor service dependent
on the dhcp client service.
> -----Original Message-----
> From: condor-users-bounces@xxxxxxxxxxx
> [mailto:condor-users-bounces@xxxxxxxxxxx] On Behalf Of Todd Tannenbaum
> Sent: Friday, 24 February 2006 10:14 PM
> To: Condor-Users Mail List
> Subject: Re: [Condor-users] Amount of ports used by condor
> On Fri, Feb 24, 2006 at 01:10:44PM +0800, Greg.Hitchen@xxxxxxxx wrote:
> > This doesn't seem to be what happens though, unless I'm
> > misunderstanding something. Our current test pool has of
> the order of
> > 50-60 machines in it (windows machines). We have used the HIGHPORT
> > and LOWPORT options to restrict the port range to
> 9000-10000. I used
> > tcpdump on our linux central manager and it shows only 445 unused
> > ports in that range, i.e. about 555 unique ones used (~ 55
> machines X
> > 10 ports each (1 vm)).
> Strange. We currently have 1257 machines in our pool, and
> our central manager is currently using less than 50 ports total.
> How do you use tcpdump to report port usage?
> I use lsof. For example, lets say you condor_collector was
> running on pid 8871 and the IP address of the machine is
> 128.444.444.444. Then enter the following command:
> lsof -a -i @128.444.444.444 -p 8871
> to view all ports in use by that process. Do this for the
> condor_collector and the condor_negotiator.
> There are only a few settings in the condor_config that could
> significantly impact the port usage on the central manager.
> They are (off the top of my head):
> UPDATE_COLLECTOR_WITH_TCP (ours is FALSE)
> COLLECTOR_SOCKET_CACHE_SIZE (ours is 0)
> NEGOTIATOR_SOCKET_CACHE_SIZE (ours is 16)
> Perhaps you are updating your collector w/ TCP and have a
> large COLLECTOR_SOCKET_CACHE_SIZE?
> Condor-users mailing list