kwe@BU-IT.BU.EDU (08/08/89)
I am interested to find anyone running NFS across p4200s and how you go about doing same. I am looking for actual performance war stories. Our environment is PRonet-80 backbone with Ethernet subnets. All our p4200s have the Proteon processors and Proteon Ethernet boards. We do not use access list checking. Our workstation environment is typically diskless Sun 3/50s and newer going to Sun 3 and 4 file servers. I am curious to know whether you turn on checksumming and when you see performance hits with NFS checksumming on running across one or more p4200s. I assume no one is running NFS across p4100s. :-) Kent England, Boston University
ron@HARDEES.RUTGERS.EDU (Ron Natalie) (08/08/89)
I've run NFS across a p4200, sort of. Last year at INTEROP I NFS mounted my disk at Rutgers from a Sun 3/60 at the show. As near as I can tell it went through A cisco box, at least one p4200, probably several NSFNET NSS's, two JVNC Net VAX's (one 780, one 750), and at least one cisco box at Rutgers. It was slow, as you might expect, but it didn't break. -Ron I'd expect that checksumming to be an independent issue, that only gets checked at the end points and not by the routers themselves.
CLIFF@UCBCMSA.BITNET (Cliff Frost {415} 642-5360) (08/08/89)
>I'd expect that checksumming to be an independent issue, that only >gets checked at the end points and not by the routers themselves. Unfortunately, Sun makes it almost (if not totally) impossible to turn on UDP checksumming for NFS mounts. This means that your NFS mounts depend completely on the various pieces of network hardware and software between the two end points. Fortunately this is OK a very high percentage of the time. Unfortunately it isn't always OK and the symptoms the user sees can be very mysterious. Cliff
medin@NSIPO.NASA.GOV ("Milo S. Medin", NASA ARC NSI Project Office) (08/08/89)
Actually, in 4.0.3, I believe that's been fixed. I recall even seeing in documentation! Thanks, Milo
ron@HARDEES.RUTGERS.EDU (Ron Natalie) (08/08/89)
We run NFS through CISCO routers all the time. Our major problem right now is NFS screwing up on the ends, rather than packets getting damaged in transit. However, I agree, it is stupid that no checksumming is done. Right now there is a flakey link in JVNC land that randomly trashes packets with certain bit sequences. We haven't noticed this with NFS, but certain UUCP packets always cause get lost going through this link. Since the UUCP is encapulated in a checksummed TCP stream, I assume that they appear damaged at the receiving end and are discarded, which would be fatal if it happened to a checksumless NFS packet. -Ron