[e2e] How many TCP flows fit in the Internet?
Detlef Bosau
detlef.bosau at web.de
Sat Mar 30 07:29:15 PDT 2013
After some days of thinking, I think this question is a huge part of the
buffer bloat discussion re-framed.
I think it is one of the sacred cows in our community, that a connect()
call will not fail with an "Insufficient Ressources Error" and a TCP
flow once it is established will hardly be terminated for an actual
resource shortage.
So basically, many of us expect an infinite number of flows being
possible on the Internet - though we (me) are not always aware of this
fact, and then wonder about the consequences.
Of course, I can fit 10^10 TCP flows on an 10 MBit/s Ethernet link.
However, I should not expect too much throughput (or goodput
respectively) for the individual flows.
Among all strategies of congestion control, in VJCC I miss the real
establishment of the two by far most obvious.:
In case of congestion
1. reduce the rate of existing flows. (In VJCC and actually existing TCP
Implementations we hardly can reduce a flow's rate beyond 1MSS/CWND.
Exept perhaps by employing a pacing scheme, however I'm not quite clear
yet about possible consequences.)
2. reduce the number of flows. "Soft" version: Block out new flows.
"Hard" version: Terminate existing ones.
(Just to make this clear: We don't want to deliver the Internet from
individual packets. In case of overload, we want to deliver the Internet
from sources of load. When your classroom is overcrowded, you first shut
the door that nobody will come in any more - and if this does not
suffice you will ask some persons to leave the room. In this picture,
AQM appears to me as if we opened the window - and the persons in the
room throw out their handkerchiefs.)
The hard version rises the question, which flows should be terminated
and - even more important: How do applications deal with flow
termination / flow re-routing and of course, with respect to re-routing,
how to we achieve a load sensitive routing in the Internet?
Sometimes I think, one of our sacred cows is the "perpetual motion
machine cow": "There is infinite capacity in the Internet and we all
can use it each and every day."
Actually, the Internet has limited, if huge, capacity and the issue is
to share this in a reasonable manner.
Detlef
(asking for more capacity for his mailbox. It's interesting - and
actually inspiring - to see how many people lurk on this list and hardly
write anything here, but are often interested in interesting discussions
off list.)
--
------------------------------------------------------------------
Detlef Bosau
Galileistraße 30
70565 Stuttgart Tel.: +49 711 5208031
mobile: +49 172 6819937
skype: detlef.bosau
ICQ: 566129673
detlef.bosau at web.de http://www.detlef-bosau.de
More information about the end2end-interest
mailing list