Date: Wed, 23 Aug 1995 10:49:07 -0400 (EDT) From: Christopher Sedore <cmsedore@mailbox.syr.edu> To: "Amancio Hasty Jr." <hasty@rah.star-gate.com> Cc: freebsd-hackers@freebsd.org Subject: Re: Making a FreeBSD NFS server Message-ID: <Pine.SUN.3.91.950823091036.2184B-100000@rodan.syr.edu> In-Reply-To: <199508221805.LAA00928@rah.star-gate.com>
next in thread | previous in thread | raw e-mail | index | archive | help
The Network Computing article test were done over FDDI. I got ~3MB/sec write and ~2MB/sec read while doing the tests. Note that the ~2MB/sec read was from a disk which only yielded ~2MB/sec when using IOZone locally, so a 7200 rpm drive may make the number go up. As I understand some of the NFS protocol issues, you probably won't see more than ~2MB/sec read from a single process anyway, so you'd have to use multiple IOZones or whatever other benchmark you use to break that barrier. These tests were sequential (though I did others for the article), and don't really represent a full picture of NFS. Results for operations other than long sequential reads or writes were less impressive (as discussed in other posts). The other issue to watch carefully with regard to FreeBSD NFS vs NT NFS is that NTFS journals filesystem metadata. I can hit reset on my NT box anytime and have my filesystems come back without trouble. I can't say this for *BSD. This journalling costs something, perhaps more than sync updates for metadata in ufs. What bothered me most about the NFS products is the rather poor design decisions that appeared to be made for some products-this reflects poorly on NFS implementors rather than NT. If you really want to do an NT vs FreeBSD comparison, build equivalent boxes with NT and FreeBSD+Samba, this will better compare NT vs FreeBSD. (I've actually got this setup in my lab, but haven't had the time to run benchmarks. I am using Samba on FreeBSD to serve video files to client workstations, because I could get the video to play w/out skips more reliably from FreeBSD/Samba than NT, using the same hardware). I will say that for high performance multithreaded programming, NT is a wonderful place to develop. I wish that I could say the same for FreeBSD, since I'd much rather make a home for my projects there. -Chris On Tue, 22 Aug 1995, Amancio Hasty Jr. wrote: > >>> Brian Tao said: > > On Sun, 20 Aug 1995, Amancio Hasty Jr. wrote: > > > > > > Curious then, where is the time being spend in the NFS code? > > > > > > Given that we can drive the ethernet at near capacity and that the > > > disks are very fast . It pretty much leads me to believe that > > > the NFS code or protocol is the bottle neck. > > > > Are you talking about the case of synchronous writes to a FreeBSD > > NFS server? I don't expect the bandwidth in the other cases to climb > > any higher (already in the 800K/sec to 900K/sec range over 10Mbps > > Ethernet). > > > Should be interesting to find out the NFS performance numbers with > your configuration using fast ethernet. > > If they are very high, I suggest sending the performance figures to > Networking Computing 8) > > Cheers, > Amancio > > > > > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SUN.3.91.950823091036.2184B-100000>