[2627] in linux-net channel archive
Re: SNLE driver (Slow Network Link Emulating Driver)
daemon@ATHENA.MIT.EDU (Miquel van Smoorenburg)
Tue Apr 23 08:21:45 1996
To: submit-linux-dev-net@ratatosk.yggdrasil.com
From: miquels@drinkel.ow.org (Miquel van Smoorenburg)
Date: Mon, 22 Apr 1996 19:08:25 +0200 (MET DST)
In article <199604190834.JAA19507@snowcrash.cymru.net>,
Alan Cox <alan@cymru.net> wrote:
>> > for the IP's we want to slow down. Every netdevice has to tell the kernel
>> > when its ready to send new packets, so after accepting a packet for
>> > sending we calculate how long it would take on the limited bandwidth to send
>> > the packet, set a timer so we can ack the transmission later, and hand the
>> > packet to the real device. (ofcourse we must also check if the real device
>> > is accepting packets at that time).
>> >
>> Exactly like the delay module I implemented for my ISDN driver.
>> Code reuse is a Good Thing. ;-)
>
>Only a delay isnt what you want. A delay has no effect on bandwidth usage
>until you are running out of window, then it starts to cause slow starts
>and increase your network usage if you are not careful.
No, I wasn't planning on just delaying, but on keeping the packets
in the queue, accepting only for example 8000 bytes worth of packets
per second. That's exactly how for example the serial drivers behave.
> Done properly (with
>a traffic shaper) you act correctly for a low speed link, you queue
>and effectively sample the incoming data stream at lower speeds and will
>lose some frames under load (giving correct flow control behaviour).
Yup exactly what I had in mind.
Mike.
--
+ Miquel van Smoorenburg + Cistron Internet Services + Living is a |
| miquels@cistron.nl (SP6) | Independent Dutch ISP | horizontal |
+ miquels@drinkel.ow.org + http://www.cistron.nl/ + fall +
