Thanks for your snapshot. This behavior is not a bug, the page is
designed to do this. All graphs are locked to the same [min,max] range,
where min,max are the minimum and maximum value over all graphs during
the time period chosen.
Unfortunately, when a node goes ballistic it causes graphs like yours
which are not very useful. Most of the time, however, this behavior is
exactly what you want since you are interested in the relative
performance of one node vs another.
Matt knows how to use rrdtune to "get rid" of the errant value, which
will make everything look ok. Otherwise, your short-term graphs (hour,
day) should go back to normal soon of their own accord.
-federico
On Tuesday, September 23, 2003, at 01:44 AM, Dr Jason J Hogan-O'Neill
wrote:
Hi Federico,
Thanks for the reply. Please find attached a snapshot of the cluster
in png format (120kb I'm afraid - sorry).
The offending machine in the cluster is heisenberg.
The graph that shows the total cluster load as a whole should not
really have its y-axis be greater than 28 or 30
The y-axis for each individual machine should be around 2 or 3 max.
Many thanks,
Jason
The graphs should automatically scale to the new value. This could be
a bug in the display system. Could you send me a screen-shot of your
cluster in this state.
-federico
On Thursday, September 18, 2003, at 02:46 AM, Dr Jason J
Hogan-O'Neill wrote:
Hi,
Hope you can help - would really appreciate it.
One of the machines on our cluster went ballistic and the load on it
went above 50! Is it possible to rescale the y-axis so that it
doesnt show the load all the way up to 50? I want the axis to
cut-off at say 2 or 2.5 as this is what it is on all the other
machines. Our statistics for the month are presently not
meaningfully visible at the moment.
Many thanks in advance for any help
Jason
PS I have told all the beowulfs admins to get ganglia on their
systems too.
--
Dr JJ Hogan-O'Neill
High Performance Computing
Dept of Mathematics
University of Bristol
BS8 1TW
0117 928 9000 (switchboard)
0117 928 9125 (direct)
-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Ganglia-developers mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ganglia-developers
Federico
Rocks Cluster Group, San Diego Supercomputing Center, CA
--
Dr JJ Hogan-O'Neill
High Performance Computing
Dept of Mathematics
University of Bristol
BS8 1TW
0117 928 9000 (switchboard)
0117 928 9125 (direct)
<gangliasnap.png>
Federico
Rocks Cluster Group, San Diego Supercomputing Center, CA
