BTW, there are other timeout values which can and should be changed
for clusters using TCP_NODELAY sockets (e.g. the full second
retransmit timer when sh->users is set and another seldom used 30
second partial packet timer).
The partial packet timer is a 2.0.x vestige, it no longer exists in
2.2.x
Look, face it, parallel processing and clustering are still
"uninteresting" in the standard kernel. And I think such a small
group of users can live with special local TCP/networking changes in a
"beowulf" type package, in fact they already do.
We try hard not to put changes in which handle a special small group
of users who have special needs that no-one else has, this is the road
to bloat and unmaintainable code. I am, however, not arguing that at
some point clustering will be more central and mainstream
(ie. common), but this isn't the case right now.
On the other hand, I know that many of your observations will help us
improve Linux/TCP generically and for all users, not just cluster
folks.
Later,
David S. Miller
davem@redhat.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/