*BSD News Article 3570


Return to BSD News archive

Xref: sserve comp.unix.bsd:3614 comp.protocols.nfs:4207
Newsgroups: comp.unix.bsd,comp.protocols.nfs
Path: sserve!manuel!munnari.oz.au!uunet!snorkelwacker.mit.edu!bloom-picayune.mit.edu!athena.mit.edu!eichin
From: eichin@athena.mit.edu (Mark W. Eichin)
Subject: Re: 386BSD: 16550's vs. NFS
In-Reply-To: schneck@Physik.TU-Muenchen.DE's message of Wed, 12 Aug 1992 20:25:59 GMT
Message-ID: <EICHIN.92Aug12193408@tsx-11.mit.edu>
Sender: news@athena.mit.edu (News system)
Nntp-Posting-Host: tsx-11.mit.edu
Organization: Massachusetts Institute of Technology
References: <EICHIN.92Aug9004425@tsx-11.mit.edu> <1992Aug9.083431.5746@BitBlocks.COM>
	<oc27njc@rhyolite.wpd.sgi.com>
	<schneck.713651159@Physik.TU-Muenchen.DE>
Date: Wed, 12 Aug 1992 23:34:15 GMT
Lines: 29

I've gotten a number of useful comments from people on the net; thanks
for reading my note. I'm currently looking at FAS (Final Async
Solution) to see if I can port that to BSD rather than continuing to
hack the mainline driver (after all, FAS is designed to support
multi-port boards.) The interrupts are still not showing up, I'm about
ready to take a logic probe to the board and test it... I've even set
breakpoints in Vcom3 and they don't fire, so I'm pretty sure they're
not getting to the CPU.
	Several people have suggested reducing the block sizes; from
bad experiences with UDP fragmenting on Ethernets, I already knocked
the block size down to 1024 (both rsize and wsize.) I tried it at 128
once but got XDR errors from SunOS on the server.
	Another suggestion:
>>And radically bloat the attribute cache expirations.
	386BSD mount doesn't have an option to control that (the only
place I've ever seen fine control of the attribute cache is in SunOS.)
A quick glance at the source shows NFS_ATTRTIMEO at 5 seconds in
nfs/nfs.h, I'll try compiling that higher (but I doubt that it is much
of a problem at 38,400bps.)
>> ... and use ppp or nfs over TCP or turn on UDP checksums ... you might 
>> loose big with NFS otherwise, even with MNP4/V.42!
	Actually, I'm using a direct wire and *still* losing. I don't
know if I mentioned this before but NFS over TCP (to another 386BSD
machine) works fine, it is the UDP version that fails.
	I haven't heard from *anyone* who is actually *using* 386BSD
NFS over SLIP, so I may be the only having this problem just now :-)
				_Mark_ <eichin@athena.mit.edu>
				MIT Student Information Processing Board
				Cygnus Support <eichin@cygnus.com>