From: C++ Newbie on
A login/job submit node on a Linux parallel cluster is running Red Hat
Enterprise Linux AS release 4 (Nahant Update 3).

I am curious as to why it is so persistently slow even though users
aren't supposed to run jobs on it (jobs are auto-killed after 15 mins)
and the node is fairly high specced (4 AMD Opteron 854 processors
1799.978MHz, 1024KB cache/core with 3599.39 bogomips, 8GB RAM). The
server is also slow when top does not reveal any processes dominating
CPU/memory usage.

There is a persistent pair of zombie processes called
"SecondaryServer" and "SmallMonitoring", see below:

root 30246 0.0 0.0 0 0 ? Zsl May06 0:49
[SecondaryServer] <defunct>
root 13832 0.1 0.0 0 0 ? Zsl May06 1:02
[SmallMonitoring] <defunct>

They appear upon a reboot but do not use more than 1% CPU so far.

I don't have root privileges (just someone who submits jobs to the
queue) so I appreciate that my curiosity may be dead-ended.

Any other suggestions? Is there a way to check the real time network
latency and free bandwidth (just temporarily) by generating traffic
between the submission node and my desktop PC? Only ssh works to the
server.

Thanks.
From: Bill Marcum on
On 2008-05-07, C++ Newbie <newbie.cpp(a)googlemail.com> wrote:
>
>
> A login/job submit node on a Linux parallel cluster is running Red Hat
> Enterprise Linux AS release 4 (Nahant Update 3).
>
> I am curious as to why it is so persistently slow even though users
> aren't supposed to run jobs on it (jobs are auto-killed after 15 mins)
> and the node is fairly high specced (4 AMD Opteron 854 processors
> 1799.978MHz, 1024KB cache/core with 3599.39 bogomips, 8GB RAM). The
> server is also slow when top does not reveal any processes dominating
> CPU/memory usage.
>
> There is a persistent pair of zombie processes called
> "SecondaryServer" and "SmallMonitoring", see below:
>
> root 30246 0.0 0.0 0 0 ? Zsl May06 0:49
> [SecondaryServer] <defunct>
> root 13832 0.1 0.0 0 0 ? Zsl May06 1:02
> [SmallMonitoring] <defunct>
>
> They appear upon a reboot but do not use more than 1% CPU so far.
>
A zombie by definition shouldn't be using any CPU at all. These
processes probably aren't related to the slowdown. If you're curious
about them you have to find how they were started, and maybe catch them
with top or ps before they die.