Leonardo Pappagallo wrote: > hi Jan, > I haven't resolve the delay problem (a lot of sincronytation cycles > without transmitting packets), I have tried to implement your > suggestions: > > 1) stack solution with 64KB > >> Mmh, I'm not sure about the locking behaviour of glibs regarding main's >> stack. I think to remember scenarios were, even after mlockall, you >> first have to touch the stack pages, e.g. by creating a large local> >> array (say 8 KBs) and filling it. > > 2) mlockall() solution >> To exclude a problem with the real-time environment, you could also try >> the scenario in kernel space. Should be straight forward: just create >> RT-tasks and put the while loops into them. > > but the behavior doesn't change. > > I thought that this strange behavior was only at the packets sending > startup phase but when i tried to test the situation for more time > about (5 minutes - 600 Kpackets of 256 bytes), I have noticed that in > some instances I have the same situation with delays also about 50 ms. >
Than consider the kernel test and the instrumentation with time stamps next to see where the delay is located. Jan
signature.asc
Description: OpenPGP digital signature
