gamiddle@maytag.waterloo.edu (Guy Middleton) (10/09/90)
In article <41981@mips.mips.COM> len@mips.COM (Len Lattanzi) writes: > In article <1990Oct9.040440.23066@watcgl.waterloo.edu> idallen@watcgl.waterloo.edu (Ian! D. Allen [CGL]) writes: > :Aha! We see this behaviour between real MIPS machines and our Ultrix > :3.1C DECsystem 5400's. BSD Vaxen, Suns, other machines, have no > :problem talking to the 5400's, but connecitons from those three MIPS > :machines hang on large packets. And it's asymmetric -- we can rcp *out* > :of the DS5400, but not in. > > The first M/120's did not support trailers on enet packets and silently > hung connections to other hosts that sent them. (just a guess but try > turning off trailers on all interfaces -- I have no idea if Ultrix > 3.1/4.0 enet hangs have similar causes) Aha. That seems to fix the problem. I'll disable trailers on all our mips boxes; they are evil in any case.
lgy@phys.washington.edu (Laurence G. Yaffe) (10/10/90)
gamiddle@maytag.waterloo.edu (Guy Middleton) writes: >Aha. That seems to fix the problem. I'll disable trailers on all our mips >boxes; they are evil in any case. Why? -- -------------------------------------------------------------------------- Laurence G. Yaffe Internet: lgy@newton.phys.washington.edu University of Washington Bitnet: yaffe@uwaphast.bitnet
bin@primate.wisc.edu (Brain in Neutral) (10/10/90)
From article <lgy.655493851@newton>, by lgy@phys.washington.edu (Laurence G. Yaffe): > gamiddle@maytag.waterloo.edu (Guy Middleton) writes: >>Aha. That seems to fix the problem. I'll disable trailers on all our mips >>boxes; they are evil in any case. > > Why? I don't know the "official" answer, the the practical answer is: precisely because they cause the kind of problems seen in this thread. Not all implementations of TCP/IP know about them, and some handle them badly, and you get crashed, hung or broken connections of all sorts. Such as (this is what I've seen here, communicating with, e.g., 3b2's or IBM mainframes): SMTP sessions that hang at the end, resulting in mail being sent over and over... NNTP sessions hanging forever etc. -- Paul DuBois dubois@primate.wisc.edu "Was all of this because I wore a big man's hat?"
grr@cbmvax.commodore.com (George Robbins) (10/10/90)
In article <3246@uakari.primate.wisc.edu> bin@primate.wisc.edu writes: > From article <lgy.655493851@newton>, by lgy@phys.washington.edu (Laurence G. Yaffe): > > gamiddle@maytag.waterloo.edu (Guy Middleton) writes: > >>Aha. That seems to fix the problem. I'll disable trailers on all our mips > >>boxes; they are evil in any case. > > > > Why? > > I don't know the "official" answer, the the practical answer is: > precisely because they cause the kind of problems seen in this thread. > Not all implementations of TCP/IP know about them, and some handle them > badly, and you get crashed, hung or broken connections of all sorts. > Such as (this is what I've seen here, communicating with, e.g., 3b2's On the other hand, Ultrix 3.1 trailer negotiation seems to work, and if Ultrix 3.1 and Ultrix 4.0 can't talk TCP with and without trailers, then there is something wrong. I hope someone having this diffuculty follows up with the Support Center or an SPR, in this day and age, networking problems shouldn't be left lingering around till the "next release"... -- George Robbins - now working for, uucp: {uunet|pyramid|rutgers}!cbmvax!grr but no way officially representing: domain: grr@cbmvax.commodore.com Commodore, Engineering Department phone: 215-431-9349 (only by moonlite)
djl@mips.COM (Dan Levin) (10/11/90)
lgy@phys.washington.edu (Laurence G. Yaffe) writes: > gamiddle@maytag.waterloo.edu (Guy Middleton) writes: > > >Aha. That seems to fix the problem. I'll disable trailers on all our mips > >boxes; they are evil in any case. > > Why? Basically because they are non-standard, not documented, and don't do any good for modern machines. The former is a general no-no in networking. The latter because that means many new machines don't bother implementing them, which can open them up to all kinds of errors. You have to recognize and reorganize trailer packets, even if you never send them. Since you never send them, however, they tend to lose mind- share. The receiving implementation, which is important, can get broken by other changes - and since you aren't thinking about them (or likely testing vs. them often) you may not catch the bug. If you have pages larger than 1500 bytes (and we almost all do these days), then you actually pay a penalty on every incoming packet to deal with trailers. They do you no good at all. In short, trailers were a quick hack for machines with small page sizes. They are now badly outdated, it isn't clear they did any good to start with, and they are a danger to any ethernet because they are non-standard and very poorly documented. -- ***dan {decwrl,pyramid,ames}!mips!djl djl@mips.com (No, Really! Trust Me.)