anyone running with large server parms on linux afs servers?

[Lyle Seaman]

Russ Allbery wrote:

> Nathan Neulinger <nneul@umr.edu> writes:
>
> > Why the -nojumbo? Do you actually have routers that couldn't handle it,
> > or some other reason?
>
> We had major problems with it originally, but "originally" in this context
> may be somewhere around 1992 to 1994 (it was before my time).  Now it's
> there for "if it's not broken, don't fix it" reasons.

It was there because Stanford had (and may still have, given the way these
things go) routers that couldn't handle it.   You had Cisco routers
configured with shallow forwarding queues, and they would always drop one
frag of a jumbogram.  Newer Rx-es do jumbograms differently,  but I think
they also eliminate much of the benefit of using jumbograms, too, so it's
probably a wash.  I won't swear to that,  as I haven't measured it, but it
would be informative for someone to compile the testclient and testserver and
profile them.