> -----Original Message-----
> From: John Border [mailto:border@hns.com]
> Sent: luned́ 26 novembre 2001 18:44
> To: Michele Milani
> Cc: 'pilc@grc.nasa.gov'
> Subject: Re: "rwnd control" as a PEP mechanism
>
>
>
> Could you elaborate in three or four sentences what you
> mean by "rwnd
> control"?
Sure, I'm sorry, I'm so involved in this matter that it appears me natural
to speak about "rwnd control" without clarifying what i mean.
Our study aims to understand how some equipments sold as "shapers" work.
Packeteer and Allot are the top players in this market. Packeteer declares
to use an algorithm that changes the value of the receiver advertised window
(rwnd) field in ACK packets.
Do you think that a shaper can be a kind of PEP? I think so, but maybe I'm
wrong.
> It looks to me like the post you reference is talking
> about a change to
> the TCP sender and not a PEP.
No, shapers are "inside" the network and the sender and the receiver go on
TCPing without knowing about shapers.
> In any case, RFC 3135 is a "survey" document and, as
> such, is not intended
> to capture every single possible "PEP" idea. It is just
> trying to get across
> the concepts of the kinds of things that are done by a PEP
> (and the reasons
> why they are used and the consequences of using them)...
I understand that and my post was only curious. My intention was not to
criticize RFC 3135, only to know if, according to you, the modification of
the value of the receiver advertised window (rwnd) field in ACK packets can
be an alternative PEP mechanism.
Gorry, I reply to you through the ML as I think all the ML can be
interested.
> -----Original Message-----
> From: Gorry Fairhurst [mailto:gorry@erg.abdn.ac.uk]
> Sent: luned́ 26 novembre 2001 18:33
> To: Michele Milani
> Subject: Re: "rwnd control" as a PEP mechanism
>
>
> Not entirely sure what you mean, so I thought I would check
> before I answered.
>
> By "rwnd", I presume you mean the receiver advertised window.
You're right.
> If so, do you suggest increasing this to a value larger than
> the receiver
> first advertised,
No: in this way I'll lose flow control mechanism and packets will fill
buffers.
> or reducing to throttle the speed of communication?
Yes: I can implement some QoS with different classes of TCP streams that
traverse the same link or I can optimize rwnd value to get better overall
performance.
What do you think about that? Does it make sense?
Cheers,
Michele
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 09:12:29 EST