[3454] in linux-net channel archive
Re: shaper or whatever
daemon@ATHENA.MIT.EDU (Alan Cox)
Mon Jun 24 12:12:23 1996
From: Alan Cox <alan@cymru.net>
To: smurf@smurf.noris.de (Matthias Urlichs)
Date: Mon, 24 Jun 1996 15:06:27 +0100 (BST)
Cc: linux-net@vger.rutgers.edu
In-Reply-To: <96Jun24.125151+0100met_dst.2226-574+9@work.smurf.noris.de> from "Matthias Urlichs" at Jun 24, 96 12:51:50 pm
> Why do you need to set the time when you add to the queue? It seems to me
> that it'd be simpler to calculate that when sending, i.e. jiffies +
> length-of-packet / bandwidth, with fixed-point arithmetic (or whatever) to
> allow for rounding errors and small packets.
You have to know when to kick the frame out in terms of time beyond jiffies.
So I have a timer saying this packet starts at about time t, and one saying
this packet is t ticks long. The next packet then starts at time jiffies+t
> > this to its latency. Frames exceeding the latency boundary are dropped. Its
> I'd carry a counter for the bytes in the queue and limit that.
That can lead to long delays unless you compute the byte count for the
queue based on the speed.
> Advantage: You can easily experiment with dropping a random packet in the
> queue, instead of the last one, without going through all the calculations
> again.
I don't want to drop a random packet. The computation already tells me
what to drop and when.
Alan
