RE: Some performance measurements on the FreeBSD network stack


>=0A=
>From previous tests, the difference between flowtable and=0A=
>routing table was small with a single process (about 5% or 50ns=0A=
>in the total packet processing time, if i remember well),=0A=
>but there was a large gain with multiple concurrent processes.=0A=
>=0A=
=0A=
Yes, that sounds about right when we did the tests a long while ago.=0A=
=0A=
>=0A=
> Removing flowtable increases the cost in ip_output()=0A=
> (obviously) but also in ether_output() (because the=0A=
> route does not have a lle entry so you need to call=0A=
> arpresolve on each packet). =0A=
>=0A=
=0A=
Yup.=0A=
=0A=
>=0A=
> So in revising the route lookup i believe it would be good=0A=
> if we could also get at once most of the info that=0A=
> ether_output() is computing again and again.=0A=
>=0A=
=0A=
Well, the routing table no longer maintains any lle info, so there=0A=
isn't much to copy out the rtentry at the completion of route=0A=
lookup.=0A=
=0A=
If I understood you correctly, you do believe there is a lot of value=0A=
in Flowtable caching concept, but you are not suggesting we reverting=0A=
back to having the routing table maintain L2 entries, are you ?=0A=
=0A=
--Qing=0A=
_______________________________________________
freebsd-current@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@freebsd.org"