Hi, in my student research program, I analyse the QoS (latency) of an embedded linux system (pc-104 module, 300MHz) with RTAI(rtai-3.2) and RTnet(rtnet-0.8.3). I wrote a ISR (realtime) for the parallel port, which generates an impulse on the data pins for each incoming interrupt before acknowledging the IRQ. In my experiment set-up I generated a (square) input signal (TTL) for the parallel port with a frequency of 10kHz. This signal also started a frequency counter (scaler) and the response of the parallel port stopped the counter. The stored latency-times could be read from the intern memory (64Mb) of the scaler to the PC via a vme bus. With this setup I measured the latency of the system (10mins -> 3.6mio values) with different parameters (without RTnet (idle - load) and with RTnet (idle - load)). idle: network up and running (RTnet or normal networking) load: compiling kernel & analysing network with netperf
Each of this measurements was made twice, once in oneshot mode and once with a periodic scheduler (50.000 ns) -> 8 measurements. The plottet results could be viewed at: http://wwwcip.informatik.uni-erlangen.de/~sirimemb/srp/ And now the question: Why do I have two peaks when the system is idle (at 4.3us and 5.3us)? Look especially at the following plots: http://wwwcip.informatik.uni-erlangen.de/~sirimemb/srp/detailded_periodic.png http://wwwcip.informatik.uni-erlangen.de/~sirimemb/srp/detailded_oneshot.png Thanks in advance, Richard Membarth -- "We need to change that attitude about how prolific we can be with the people's money." George W. Bush March 16, 2001
pgpg16UNUkBYj.pgp
Description: PGP signature
