- Stefan
-----Original Message-----
From: Scott Bradner [mailto:sob@harvard.edu]
Sent: Tuesday, August 11, 1998 10:12 AM
To: end2end-interest@ISI.EDU
Subject: Sitara Networks
I talked to the Sitara Networks people this am.
Basically they replace TCP with their own protocol (protocol type 109)
(There is a lot more to their technology than what I was able to note
down
as we talked but here is the gist of what I got.)
The basic flow is one of bursts of packets with the length of burst and
the
inter packet interval modified by the flow control mechanism. Initial
inter packet interval is determined by the server using information from
the client on the client's connection speed (i.e. this is designed to be
used with modem-connected clients.) Each transmission burst has a flag
on
the last packet that is used to request reception stats from the client
end. The stats include a list of what packets were lost (if any) - the
RTT
is calculated by measuring the response to the stats request
The burst length & rate are modified up or down by looking at the ratio
of
packet loss in a burst over the burst length - single packet losses are
generally ignored in figuring what rate to use but multiple packet loss
in
a burst will cause the system to back off.
They are working on a paper to describe the technology that they hope
to
have ready for distribution in a few months.
They are talking to a bunch of companies about installing their servers
and
setting up to download their clients - at least some of the companies
are
quite interested in getting them to involve the IETF in an analysis of
the
technology &/or having the IETF consider standardizing it.
Scott