[3369] in linux-net channel archive
Re: Traffic shaper
daemon@ATHENA.MIT.EDU (Mike Kilburn)
Thu Jun 20 18:02:09 1996
Date: Thu, 20 Jun 1996 18:15:19 +0200 (SAT)
From: Mike Kilburn <mike@lserv.conexio.co.za>
To: "Theodore Y. Ts'o" <tytso@mit.edu>
cc: Alan Cox <alan@cymru.net>, alan@lxorguk.ukuu.org.uk, roque@di.fc.ul.pt,
linux-net@vger.rutgers.edu
In-Reply-To: <9606201610.AA08351@dcl.MIT.EDU>
On Thu, 20 Jun 1996, Theodore Y. Ts'o wrote:
>
> Well, consider what a "slow link" do when it receives too many incoming
> packets --- or more accurately, what a router will do when it has a T-1
> line on one end, and a 14400kbps line on the other, and it's receiving
> too many packets down the T-1 that are destined for the 14400kbps: it
> will start dropping packets. That's how the data source will figure out
> that there is a slow link in the path.
>
> You're suffering from the misunderstanding that TCP measures "slow"
> links by doing some sort of timing measurement between when a packet is
> transmitted and when it is received. Although that may be the intuitive
> way to think about it, that's just not how it works. (It's also
> completely impractical to do things that way, think about how you would
> do this time measurement....) How TCP implementations really measure
> "bandwidth" is by seeing how many bytes/second they can stuff down a
> pipe before it starts complaining.
>
> - Ted
Ok. TCP does not do what I thought it did. No problem. What I am
proposing is no worse than a slow link, if fact thats what I am after.
I dont care about anything above the IP level. I must write some code
now to convince myself what I want to do will not work, because it
seems like it should to me.
