[mrtg] faster processing

[Perry, Duane]

We have a large network that I have been trying to monitor with MRTG for
years now.  As I was given better hand-me-down equipment over the years
I kept running into the wall on the number of Targets I could query in
the 5 minute interval before the next cycle started.  I finally got a
new machine built just for MRTG but found that more processing power was
not the key.  Tobi explained that disk i/o is the bottleneck.  We
already have pretty fast disks so one suggestion was to go to RAMDISK
which would create a virtual disk in RAM and occasionally copy it to the
hard drive so the rrd files would not be lost if the machine went down
for some reason.  I suspect this would speed the process up dramatically
as I have found that we can query about 34,000 Targets and get the
responses back in about 30 seconds.  We have the memory installed but
have not tested the RAMDISK solution yet.  Has anyone else tried that?

 

I have also been looking at Cacti and noticed that there is a plug-in
called "boost" that seems to perform the RAMDISK function.  It stores
the rrd info in memory until it can write it to disk later.  Since it is
performing this for an RRD based system I imagine it would be easy to
adapt for MRTG.  Has anyone tried "boost"?  Adapting boost to MRTG would
be beyond my programming skills but I wondered if anyone knew enough
about this to advise me on whether to ask our programming team to look
at such a project.

 

Duane Perry

University of Missouri

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.oetiker.ch/pipermail/mrtg/attachments/20071101/58c3266d/attachment.html