Date: Fri, 25 Nov 2011 17:47:12 -0800 From: Jeremy Chadwick <freebsd@jdc.parodius.com> To: George Mitchell <george+freebsd@m5p.com> Cc: freebsd-stable@freebsd.org Subject: Re: TCP Reassembly Issues Message-ID: <20111126014712.GA58776@icarus.home.lan> In-Reply-To: <4ECFD8D2.4030604@m5p.com> References: <CAPNZ-Wq38=F3o2hYuYF_unBj3SZQ52XhVhdcwQ8PE_vU9xc2YA@mail.gmail.com> <CABLqceRyohFFYtdnW%2Bt%2B63kwk00cbkJVz3oXWqrR=xijkg14CA@mail.gmail.com> <4ECE9914.6020502@turing.b2n.org> <CAPNZ-WrQdjFEErt6K1uEndSc082q8Q48VHvtbP=5cNXFw7H7Rg@mail.gmail.com> <1A5B3A48-7DF3-4018-A244-152BDE96299A@lassitu.de> <CAJ-VmokMrJipptQWjuhY5-41p=-P2-xc3ze1p4JKTXWz3MgtzA@mail.gmail.com> <CAPNZ-Wr8r%2Bz7u%2BT3xgY3dfwhEZkN27NebrXzaXmXmneWpMw8xA@mail.gmail.com> <20111125020004.GA36109@icarus.home.lan> <4ECFD8D2.4030604@m5p.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Nov 25, 2011 at 01:05:06PM -0500, George Mitchell wrote: > On 11/24/11 21:00, Jeremy Chadwick wrote: > >[...] > >If none of this solves the problem, then I consider this a priority 0 > >blocker (read: "all hands on deck") issue with the IP stack in FreeBSD > >9.x and will need immediate attention. > > > >I would strongly recommend a developer or clueful end-user begin > >tracking down who committed all of these bits and CC them into the > >thread. I would start by looking who implemented the > >net.inet.tcp.reass.cursegments sysctl, because that isn't in RELENG_8 at > >all. > > > > I've tried out the 9.0 release candidates, and what I notice is that for > a few minutes after the system starts, I get wonderful NFS read > throughput (7+ MB/s over a 100 megabit interface) -- more than twice as > fast as 7.n or 8.n on the same hardware -- quickly degrading to abysmal > (less than 0.5 MB/s). Is this possibly related to the problem under > discussion? -- George Mitchell > > P.S. A lot of other 9.0 features look very nice indeed! You could try forcing UDP NFS (assuming this is possible; I would assume on the server side "nfsd -u" is needed and on the client side use of the mntudp option would be needed in /etc/fstab; see mount_nfs(8)) description that others have given indicate the problem being discussed affects purely TCP. Regarding NFS performance in general -- and this is in no way shape or form a slam against Rick -- it would be good to get some actual Linux vs. FreeBSD numbers when it comes to NFS performance, including what protocols are used (TCP vs. UDP) and NFS versions are used (3 vs. 4). I have a gut feeling NFS on Linux is significantly faster, and it would be really helpful to find out how/why. -- | Jeremy Chadwick jdc at parodius.com | | Parodius Networking http://www.parodius.com/ | | UNIX Systems Administrator Mountain View, CA, US | | Making life hard for others since 1977. PGP 4BD6C0CB |
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20111126014712.GA58776>