Re: bibliography

[Aman Singla <aman@cthulhu.engr.sgi.com>]

> I no longer recall why (this was two years ago), but there was something in
> the retransmit behavior on error that made me think, "Oh, this is going to
> suck on packet loss."

Retransmit behvaior is always an artifact of how intelligent can
you make your timeout stuff; ST spec doesn't talk much about how
to best do the timeout for the retransmission. Sophistication in
schemes there can come out of experience with more and more
networks only... something that TCP has; hopefully some of the
experience from TCP can apply to the ST case also.

There was an issue with packet loss/out of order delivery related
to the last block of the transfer not being very efficient and
relying on timeouts for the final ack to successfully complete the
transfer. But that was addressed in a more efficient manner by
the spec sometime in the last two years.

> The explanation in the draft I read also was not clear on the amount of
> pipelining allowed, but I felt that there were going to be occasional
> round-trip pauses.  Is this not true?  If so, I'm happy about that.

That is not true. The amount of pipelining is controlled by the
'advised outstanding number of CTSs' used by the stack on a per
transfer basis. Its best to have the exact number of outstanding
CTSs to keep the pipeline full and flowing all the time - and again
the onus is on the sophistication of the implementation to figure
out that number; defaults values can work only so well.

> Is the Linux code in the main kernel distributions?  I haven't seen it, but
> didn't know to go looking.  If not, got a URL?

The linux code is still not feature complete and alpha quality. I
made the first release of the long message stack last month as a
patch to the kernel. I'll be making the next release with some
additional support related to Gigabit Ethernet tomorrow or early
next week. These are development releases; we're still quite some
ways from the complete implementation of the ST feature set and
as such it might be a while before it gets into the mailine kernel.

The project page is http://oss.sgi.com/projects/stp

:a