RE: Squid's use in proxying satellite users

Hello Dax,

Check out the following RFC:

Enhancing TCP Over Satellite Channels (rfc2488)

Barry

> -----Original Message-----
> From: David J Woolley [mailto:djw@bts.co.uk]
> Sent: Tuesday, February 02, 1999 8:28 PM
> To: Dax Kelson
> Cc: squid-users@ircache.net
> Subject: Re: Squid's use in proxying satellite users
>
>
> > Specifically I need to 1) turn off TCP slow start, 2) have TCP not wait
> > for acks, and finally 3) a min TCP window size of at least 64Kb.
> >
>
> I think you mean "want", not "need". I would be wary of doing this
> especially in your case, as you are in an environment where these
> features are probably particularly beneficial to the network. You
> may find the result of starting with a huge windows is simply that
> the sending end gets source quenched quickly and you lose a lot of in
> transit packets.
>
> If your connectivity to important sites is good enough, you would
> probably be better off trying to maximimise the ability of Squid to
> do a proxy connection keep, so that it is running with connections
> which are tuned to the real throughput of the network. Even then, a
> large idle time might result in an overload when things restart.
>
> Certainly running with huge packet windows to small sites is simply
> likely to overload their routers and result in either a fast source
> quench or generally annoying them by taking a disproportionate amount
> of bandwidth at start up.
>
> As pointed out elsewhere, Acks are a fundamental feature of TCP. I
> think that the bit about Nagle was wrong; I think Nagle is to do with
> sending packets before they are full.
>
> Disabling slow start creates an illegal TCP implementation, see the
> following from RFC 1122, which is also something like STD 2 (internet
> standard). MUST means that the implementor has no choice.
>
>
> 4.2.2.15 Retransmission Timeout: RFC-793 Section 3.7, page 41
>
> The algorithm suggested in RFC-793 for calculating the
> retransmission timeout is now known to be inadequate; see
> Section 4.2.3.1 below.
>
> Recent work by Jacobson [TCP:7] on Internet congestion and
> TCP retransmission stability has produced a transmission
> algorithm combining "slow start" with "congestion
> avoidance". A TCP MUST implement this algorithm.
>
>
>
> --
> David Woolley - Office: David Woolley <djw@bts.co.uk>
> BTS Home: <david@djwhome.demon.co.uk>
> Wallington TQ 2887 6421
> England 51 21' 44" N, 00 09' 01" W (WGS 84)
>
>
Received on Wed Feb 03 1999 - 11:00:37 MST