up vote 1 down vote favorite
share [g+] share [fb]

I am using ubuntu 10.04LTS. The Ganglia on all of my nodes is not reporting the network usage correctly. There is a fairly large amount of fluctuation on the actual network, but on ganglia it seems to show a flat rate for days at a time, then it will change to another flat rate.

It is probably worth noting that I have three main nodes, two of which are in the same datacentre, and another one which is in a different datacentre.

Essentially I would like to record all in and out on my ETH0 interface, ignoring the LO interface.

link|improve this question
feedback

1 Answer

up vote 0 down vote

aha! I'll bet you've hit the issue I'v been talking about for years. I don't know how it's configured in your environment but a lot of tools monitor the systems very infrequently, on the order of minutes. The results can be mush...

it can even be worse with sar which had a default interval of 10 minutes!

collectl, like sar, focuses on a single system but grabs data every 10 seconds. the result is much more accurate information that shows most fluctuations. of course fluctuations below the monitoring level will not be seen so if you want to see them you need to reduce the interval. collectl's interval can be specified to fractions of a second but monitoring at sub-second intervals can eat up a lot of CPU.

-mark

link|improve this answer
Thanks. My ganglia is monitoring every 15 seconds - It works perfect for cpu, memory, and load, just that disk usage is wonky and the network usage is wonky. I will try reducing the interval, but I don't know if it will fix it :( – softbagel Feb 2 at 12:50
hmm, I guess the obvious next question is what is your definition of wonky? from my experience virtual ALL tools get their data from the same place which in the case of disks is /proc/diskstats. When I was developing collectl I always verified its data with tools like iostat for disk data. I'd run collectl and (or iostat if you must ;)) and compare their output. If there really is a difference, file a bug with the ganglia team, provided of course you understand why there is a difference. – Mark J Seger Feb 4 at 13:40
feedback

Your Answer

 
or
required, but never shown
discard

Not the answer you're looking for? Browse other questions tagged or ask your own question.