[etherlab-users] Jitter + Timed Out datagrams
chg at alpenjodel.de
Thu Feb 26 17:34:29 CET 2009
I was trying to do some benchmarking with my master and came across a behavior that i couldn´t explain. Usually my master has a jitter of some microseconds, but every approximatly 750 cycles i have a peak in the jitter that is about ten times higher. First i tried to switch of any unnecessary services and daemons that could cause the problems. But the result was the same. Then i tried to play a bit with the cycle time, made it faster, than made it slower, but still i have the same behaviour independent from the cycle time. So i thought that it can´t be linux that is causing the problems and must have something to do with the master or my application. But in fact it did not change a thing even when i strip of most of my application an just have the basic example running.
The last thing to be changed was the switch that is between the master and the slaves, and my hope was that the switch was the thing causing the trouble. But in fact i the get really a lot of messages about timed out Datagrams. If i put the switch in again ist works again, wich is not what i expact since the latence should be lower without the switch. I don´t want to set up the datagram timout since 500 micro seconds seem to bei quite a lot form me at 1KHz and as one would expect that this is enouph a ist is ok witz the switch.
There is no difference if i use the stable 1.4 Version, oder the development branch 1.5 (Release 1667). The master runs on a debian with kernel 2.6.24, has an intel Pro100 nic on a pentium m without XServer. I use --enable-cycles when configuring. Skipping this does not change a thing.
Can anybody give me some hints what causes the fequentlich returning high jitter, and even more important, what to do against the timed out datagrams.
Psssst! Schon vom neuen GMX MultiMessenger gehört? Der kann`s mit allen: http://www.gmx.net/de/go/multimessenger01
More information about the Etherlab-users