Hi all, This is a quick email more reporting a "gotcha": an effect that can cause gmetric updates to be lost. This may be "well known" and, in retrospect, should be expected. Apologies if this is already known, but I've not seen it documented, so I thought I'd risk sending an email.
When a program is "listening" for UDP packets (i.e. has bound [via bind(2)] a socket to some part of the udp-namespace), the kernel will queue only a finite number of packets. Once that limit is reached, the kernel will not queue any additional packets so the application will never see them. I've thrown together some code that tests this. To see this for yourself, please download the source from: http://ppewww.physics.gla.ac.uk/~paul/RndCode/pitch-catch.c Here are some test results: Machine Kernel Mem Max queued UDP SL 3.0.8 2.4.21-47.ELhugemem 1GiB 221 SL 4.4 2.6.9-42.0.10.EL 1GiB 337 Debian sid 2.6.21-2-686 1GiB 406 SL is "Scientific Linux", a rebuild of the corresponding RedHat Enterprise Linux. The consequence is that, when sending a batch of ~400 metric update messages from MonAMI to gmond (running on an SL-3.0.8 machine) there's a fairly good chance that the number of queued UDP packets will hit the 221 maximum. When this happens last few packets are lost. I guess forwarned is forarmed, but following NickG's suggestion of allowing the XDR content for multiple gmetric updates to be packed into a single packet would greatly reduce the likelihood of this problem being encountered. This is what ApMon (part of the MonaLisa project) does. However, I guess switching to a TCP transport is the better long-term solution. Cheers, Paul. ------------------------------------------------------------------------- This SF.net email is sponsored by DB2 Express Download DB2 Express C - the FREE version of DB2 express and take control of your XML. No limits. Just data. Click to get it now. http://sourceforge.net/powerbar/db2/ _______________________________________________ Ganglia-developers mailing list Ganglia-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-developers