Ivan Voras wrote: > Dag-Erling Smørgrav wrote: >> Randall Stewart <rrs_at_cisco.com> writes: >>> Sure.. dumb question though.. whats the magic cookie to pin >>> something on a cpu.. is it a system call or is there a "shell" tool >>> that will do it? >> >> Neither. There is a kernel function to tie a thread to a CPU, but it >> is not exported to userland. > > I was thinking about the kernel part, but now, thinking more, it's > probably very non-trivial to do. I though that using sched_bind() could > do it, but this only works if there's a specific thread created for some > task - I don't know how can something like 'a network stack', which > consists of myriad of callbacks and asynchrounsly called functions, be > pinned. Sorry for the noise. :) > > Not noise .. Anything that gives a suggestion on how to tweak things is good.. and I learn more :-D I am going to try LOCK_PROFILING next.. on the most drastic set of differences.. and see what I see.. I have always not liked the sender locks I have in place.. the reader side worked out real cool.. but the sender did not :-( May need to re-think these... or I might find some other surprise ;-D R -- Randall Stewart NSSTG - Cisco Systems Inc. 803-345-0369 <or> 803-317-4952 (cell)Received on Thu Apr 12 2007 - 11:10:14 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:08 UTC