*BSD News Article 89993


Return to BSD News archive

Path: euryale.cc.adfa.oz.au!newshost.carno.net.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.ecn.uoknor.edu!feed1.news.erols.com!cpk-news-hub1.bbnplanet.com!cam-news-hub1.bbnplanet.com!news.bbnplanet.com!howland.erols.net!swrinde!news.uh.edu!bonkers!web.nmti.com!peter
From: peter@nmti.com (Peter da Silva)
Newsgroups: comp.os.linux.misc,comp.os.linux.networking,comp.os.linux.setup,comp.unix.bsd.bsdi.misc,comp.unix.bsd.misc
Subject: Re: User-space file systems. 	(Re: Linux vs BSD)
Date: 24 Feb 1997 06:52:21 GMT
Organization: Network/development platform support, NMTI
Lines: 34
Message-ID: <5erdr5$rgk@web.nmti.com>
References: <32DFFEAB.7704@usa.net> <5e6qd5$ivq@cynic.portal.ca> <5ea4g4$r6e@web.nmti.com> <330DF6FD.41C67EA6@cuillin.org.uk>
NNTP-Posting-Host: sonic.nmti.com
Xref: euryale.cc.adfa.oz.au comp.os.linux.misc:160717 comp.os.linux.networking:69735 comp.os.linux.setup:99306 comp.unix.bsd.bsdi.misc:6085 comp.unix.bsd.misc:2596

In article <330DF6FD.41C67EA6@cuillin.org.uk>,
Guy Dawson  <guy@cuillin.org.uk> wrote:
> Peter da Silva wrote:
> > Since a user-mode application on a 68020 can saturate a 10 Mbps ethernet
> > (observed behaviour) I didn't originally believe that the user-mode NFS
> > server on Linux was a bottleneck...

> User mode is fine IFF you don't mind using the extra CPU resource. On
> a multi-tasking box, a user mode nfsd will be competing with other
> programs for the CPU (and net etc).

Yep. It'll use up a whole 68020's worth of performance out of the 10 or 12
68020's worth of performance the average Pentium box has available. I don't
think it'll turn out to be CPU bound after all, nope...

> Is efficient programming a good thing?

Yes, but let's apply that efficiency in the right place. Like, how about
fixing the deal in NFS where it runs namei() over the net, so to access
/foo/bar/baz/src/main.c takes 5 network transfers (one for each component
of the name) just to find the file to open it. If you passed the whole thing
in one chunk, and got back the resolved portion of the name (which would be
the whole thing if you didn't have to backtrack over the root because of ..
or symlinks) then what would that do for, say, compiles?

	open("/usr/local/include/Xpm/tinyfile.h");

(I kinda like the AFS approach for those things... you'd only have one network
 transaction (and latency) for the whole include)
-- 

             The Reverend Peter da Silva, ULC, COQO, BOFH.

                  Har du kramat din varg, idag? `-_-'