[e2e] tcp in high rate network
David Borman
dab at BSDI.COM
Mon Jun 17 11:58:36 PDT 2002
> From: Rick Jones <raj at tardy.cup.hp.com>
> Subject: Re: [e2e] tcp in high rate network
> Date: Mon, 17 Jun 2002 11:07:42 -0700 (PDT)
>
> Isn't there a presumption in TCP that the sequence space will not be
> wrapped within MSL or 2*MSL? I suppose that enabling timestamps means
> that is "way out there" but even today, timestamps are not universal,
> and we have TCP's going at > Gbit/s and wrapping the classic 32bit
> sequence space fairly quickly.
Yes, PAWS addresses that issue. From RFC 1323, pg 21:
To make this more quantitative, any clock faster than 1
tick/sec will reject old duplicate segments for link
speeds of ~8 Gbps. A 1ms timestamp clock will work at
link speeds up to 8 Tbps (8*10**12) bps!
...
Based upon these considerations, we choose a timestamp clock
frequency in the range 1 ms to 1 sec per tick. This range also
> Date: Mon, 17 Jun 2002 18:20:53 +0100 (BST)
> From: Lloyd Wood <l.wood at eim.surrey.ac.uk>
...
> On Mon, 17 Jun 2002, David Borman wrote:
>
> > The simplistic answer is that there is nothing in the TCP protocol
> > that limits the speed at which it can work.
>
> slow start bounds the speed at which you can reach the speed you want.
>
> L.
But mainly for small files, where small is relative to delay*bandwidth.
As the size of the delay*bandwidth increases, so does the crossover point.
Both of these issues relate to congestion control. And while congestion
control and window management is a critical component in any TCP
implementation, it is not part of the TCP protocol itself. Even Silly
Window Avoidance is not part of the TCP protocol (it's documented in
RFC 813).
-David Borman, david.borman at windriver.com
More information about the end2end-interest
mailing list