is anybody here working on the udp performance problem in freebsd6?

Rick Jones rick.jones2 at hp.com
Thu Sep 14 22:55:06 UTC 2006


Paul Vixie wrote:
> re: http://unix.derkeiler.com/Mailing-Lists/FreeBSD/net/2006-09/msg00129.html

modulo differences in the default UDP SO_RCVBUF size (assuming that the 
bind code doesn't make a smaller than default on all platforms 
setsockopt() call), and modulo differences in acutal path length (which 
I would think would appear easily in say netperf UDP_RR tests with CPU 
util enabled - although watch those CPU assignments) I would wonder if 
maybe there were differences in the interrupt coalescing between the 
platforms?

ftp://ftp.cup.hp.com/dist/networking/briefs/nic_latency_vs_tput.txt

rick jones


More information about the bind-workers mailing list