I finally got around to re-working a patch that I had made a long long time
ago,
based on this thread:

http://sourceforge.net/mailarchive/forum.php?thread_name=1134597643.30692.35.camel%40jason.acf.bnl.gov&forum_name=ganglia-developers

On the meta & cluster views, this patch adds the average (over the selected
time
range) cluster utilization, using the data from the summary rrds to calculate
the average number of cpus and 1 minute load average.

~Jason

Created an attachment (id=141) [details]
Average cluster utilization.

Here is the patch for version 3.0.7.

Thanks for the patch, but could you please rebase against trunk?  That's where
bugfixes should go and we'll do corresponding backports when deemed necessary.

Hi Bernard,

I checked the ganglia trunk out from subversion and tried the patch, it works
just fine, despite a few line offsets.  I cannot test the trunk on a running
server at this time though, but it does work just fine for me on 3.0.7.

Thanks,
~Jason

Hi Jason,

 I like the concept, so definitely "+1" for inclusion in trunk. But, could you
add the average load onto the host view also, in the "Time and String" metrics
section? I definitely would do the checkin for that :-)

 Wheter this needs to be [back-]ported to 3.0.x, I am neutral. IMO, 3.0.x should
really be bug-fixes only by now.

Cheers
Martin 

Created an attachment (id=144) [details]
Updated patch to include host view.

I made a similar change the change to the host view as suggested.

I have taken the latest patch and checked it into trunk

a fix for find_limits that uses the same scheme proposed by this patch (using
rrdtool graph and the PRINT command) and that was reported in BUG37 also
committed for consistency.

if it is possible to consolidate both function to avoid having to do 2 expensive
execs or to enhance the data provided (at least in host view) and interesting
alternative

Committed to 3.1.X r1694

Jason,

 do you still want a backport to 3.0.X? After all, it is new funtionality.

Martin

http://ganglia.svn.sourceforge.net/viewvc/ganglia?view=rev&revision=1706

http://ganglia.svn.sourceforge.net/viewvc/ganglia?view=rev&revision=1708

Hi,

This could be done simpler and not require any execs by
adding the average to the graph itself in the legend.  Just
add GPRINT at the end of the $series:

	$series .="CDEF:util=load_one,cpu_num,/,100,* ";
	$series .="GPRINT:util:AVERAGE:%.1lf%% ";

This automatically works at all levels: host, cluster, grid.  Then you could
even do it for other reports:

cpu:
	$series .="CDEF:util=100,ccpu_idle,- ";
	$series .="GPRINT:util:AVERAGE:%.1lf%% ";
memory:
	$series .="CDEF:util=bmem_total,bmem_free,-,bmem_total,/,100,* ";
	$series .="GPRINT:util:AVERAGE:%.1lf%% ";
for network, add these lines after the corresponding LINE2:
	."GPRINT:bytes_in:AVERAGE:'%.1lf %s' "
	."GPRINT:bytes_out:AVERAGE:'%.1lf %s' "

You could add an "Average" label before the %, but I chose not to
minimize clutter on the image.  It should be fairly obvious that
the number is an average of the graph displayed.  I did this on
3.0.7, but similar should work in 3.1.

We can add similar useful statistics to all metric graphs, though it makes them
a couple lines taller (I couldn't get it to look good on one line).

      $series .= "'GPRINT:sum:MIN:%.1lf%S Min\\r' ";
      $series .= "'GPRINT:sum:AVERAGE:%.1lf%S Avg\\r' ";
      $series .= "'GPRINT:sum:MAX:%.1lf%S Max\\r' ";

Except that it doesn't look good on small graphs, so make it conditional:
                                                                               
                                                   
      if ($size != "small") {
          $series .= "'GPRINT:sum:MIN:%.1lf %S Min\\r' ";
          $series .= "'GPRINT:sum:AVERAGE:%.1lf %S Avg\\r' ";
          $series .= "'GPRINT:sum:MAX:%.1lf %S Max\\r' ";
      }

Created an attachment (id=158) [details]
Put average numbers on all graphs and min/max on metric graphs

Here is a trunk patch to show what I am talking about.	Adds average numbers to
all graphs and min/max to metrics graphs.  I abbreviated the running processes
label when nodes are shown to prevent creating another line that would make the
image taller.	I find it looks better with $strip_domain = true since FQDN
doesn't fit well on default graphs.

Timothy,

 please open a new BUG for the enhancenemts to the graphs, which I basically like.

 The original idea of having the average load percentage displayed as a single
number should be implemented regradless of your good idea.

Martin

Done as requested: opened new bug206 and included the patch there.