[mrtg] Intermittent Graphs
s.shipway at auckland.ac.nz
Sat Feb 2 22:51:57 CET 2008
I found that the biggest bottleneck is not CPU, but disk IO speed. Therefore, abosolutely the fasted disks (multiple spindles in a RAID if possible) and the highest speed interfaces - Fibre, SAS, or SCSI, definitely not IDE. This stops the bottleneck. The next problem is to have enough memory to hold the queries.
Next, I run multiple threads of MRTG (since there are 4 CPUs I can have at least 4 threads at once) and spread these over the 5-min window so that it doesnt all happen at the same time (I have a home-grown script that takes care of this scheduling). This allows me to have 4000+ targets on a 4x3GHz CPU box with 4Gb memory and 4 disks in a SAS-attached RAID, and still have lots of free memory and about 70% CPU use. The disk IO is almost saturated, though.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the mrtg