Date: Mon, 22 Jan 2007 18:06:19 -0500 From: Randall Stewart <rrs@cisco.com> To: Andrew Gallatin <gallatin@cs.duke.edu> Cc: freebsd-net <freebsd-net@freebsd.org> Subject: Re: kern_mbuf.c patch Message-ID: <45B5436B.7090502@cisco.com> In-Reply-To: <17844.51894.773943.99076@grasshopper.cs.duke.edu> References: <45B0D2E3.9050203@cisco.com> <17841.6943.770698.707214@grasshopper.cs.duke.edu> <45B345FD.7080001@cisco.com> <17844.51894.773943.99076@grasshopper.cs.duke.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
Andrew Gallatin wrote: > Randall Stewart writes: > > Andrew Gallatin wrote: > > > Randall Stewart writes: > > > > nmbclusters = 1024 + maxusers * 64; > > > > + nmbjumbop = 100 + (maxusers * 4); > > > > > > The limit on page-size jumbos seems far too small. Since the socket > > > buffer code now uses page-sized jumbos, I'd expect to see its limit be > > > the same as nmbclusters. > > > > > > > > > Drew > > > > > Drew: > > > > Let me re-visit this .. I started real small on purpose.. so > > folks would complain ;-) > > > > How about if I calculate the number of pages the > > nmbclusters use (I will go look in the UMA structures) and > > then make it so the limit is the same number of pages > > (scaled like nmbclusters) for each of the larger clusters.. > > That sounds reasonable to me, at least for nmbjumbop, but I'm not sure > that the larger 9k and 16k clusters are used outside of drivers, so > the nmbclusters limit may be too large for them. But I suppose some > limit is better than none :) > > Drew > SCTP uses the best fit size for user data.. thus 16k gets used for large messages :-) R -- Randall Stewart NSSTG - Cisco Systems Inc. 803-345-0369 <or> 803-317-4952 (cell)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?45B5436B.7090502>