<html>
<head>
<meta content="text/html; charset=windows-1252"
http-equiv="Content-Type">
</head>
<body text="#000000" bgcolor="#FFFFFF">
I've changed MinRAM to 3GB. I think actually, on second thought,
that 4GB might prevent machines that only have 8GB from taking jobs
when they should be able to. I'll watch and tweak as necessary.<br>
Wow the cluster is getting hammered. Fun :)<br>
-Josiah<br>
<br>
<br>
<div class="moz-cite-prefix">On 4/22/15 7:54 PM, Thomas Helmuth
wrote:<br>
</div>
<blockquote
cite="mid:CABgVVjf98O1WRpSq1hDH5TPnP-xm5uLCMpcwY-M48=eponJHHQ@mail.gmail.com"
type="cite">
<div dir="ltr">Ok, good, glad it's not something in Clojush doing
the crashing! I'm guessing somewhere in the range of 4GB would
do the trick. So, let's try that?<br>
</div>
<div class="gmail_extra"><br>
<div class="gmail_quote">On Wed, Apr 22, 2015 at 7:52 PM, Wm.
Josiah Erikson <span dir="ltr"><<a moz-do-not-send="true"
href="mailto:wjens@hampshire.edu" target="_blank">wjens@hampshire.edu</a>></span>
wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">I have
rebooted compute-1-17. That should fix that. (Apologies to
Piper, who had a maxwell job running. You'll have to retry
that one)<br>
<br>
The other problem is that some nodes are actually out of
memory because of the insanely intensive rendering jobs that
are running on them. We could play with minimum free memory
required to take new jobs.... how much RAM do you think is
necessary? Right now it's set to 1GB. Maybe I should try
changing it to 2GB or 4GB?<span class="HOEnZb"><font
color="#888888"><br>
-Josiah</font></span>
<div class="HOEnZb">
<div class="h5"><br>
<br>
<br>
On 4/22/15 7:22 PM, Thomas Helmuth wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
Hi Josiah,<br>
<br>
I just started my first runs on fly since the hard
drive switch. I'm getting some weird errors. The first
is that every run started on compute-1-17 crashed with
the following message:<br>
<br>
====[2015/04/22 19:06:03 /J1504220051/T10/C10/thelmuth
on compute-1-17 ]====<br>
/bin/sh: line 0: cd: /home/thelmuth/ClusteringBench/:
Not a directory<br>
/bin/sh:
/home/thelmuth/Results/clustering-bench/determin-decim/ratio-0.25/replace-space-with-newline/tourney-7/logs/log9.txt:
Stale file handle<br>
/bin/sh: line 0: cd:
/home/thelmuth/Results/clustering-bench/determin-decim/ratio-0.25/replace-space-with-newline/tourney-7/csv/:
Not a directory<br>
...<br>
<br>
It sounds like 1-17 for some reason cannot access my
homedir. When I tried to SSH to compute-1-17, it asks
for my password, which it doesn't for other nodes. So,
sounds like there's something amiss there.<br>
<br>
I had some other runs, it looks like all on rack 2
nodes, that crashed printing the following to the
output log files:<br>
<br>
Java HotSpot(TM) 64-Bit Server VM warning: INFO:
os::commit_memory(0x00000007e5500000, 447741952, 0)
failed; error='Cannot allocate memory' (e\<br>
rrno=12)<br>
#<br>
# There is insufficient memory for the Java Runtime
Environment to continue.<br>
# Native memory allocation (malloc) failed to allocate
447741952 bytes for committing reserved memory.<br>
# An error report file with more information is saved
as:<br>
# /home/thelmuth/ClusteringBench/hs_err_pid21904.log<br>
<br>
This looks like an error we used to get with Java but
I haven't seen recently. I checked, and I don't think
anything has changed in our code regarding memory
management.<br>
<br>
Is it possible that one or both of these errors are
caused by something in the move, or an upgrade to
tractor or something? I can go back and look for
similar errors to the second one to see if we figured
out what went wrong there.<br>
<br>
Tom<br>
</blockquote>
<br>
</div>
</div>
</blockquote>
</div>
<br>
</div>
</blockquote>
<br>
</body>
</html>