AW: [e2e] Queue size of routers

[Alexandre L. Grojsgold]

On Fri, 17 Jan 2003, Raghurama 'REDDY' wrote:

> Date: Fri, 17 Jan 2003 15:41:41 -0500
> From: Raghurama 'REDDY' <rreddy at psc.edu>
> To: michael.welzl at uibk.ac.at
> Cc: MINSHALL at ACM.ORG, AVG at KOTOVNIK.COM, END2END-INTEREST at postel.org
> Subject: RE: AW: [e2e] Queue size of routers
>
> Here is my understanding -
>
> BW*Delay is an end-2-end layer-3 concept (for example in TCP) that need
> to have that much buffering, primarily in the end hosts, in order to do
> "reliable" transport.


That's my understanding too. I completly agree!!

>
> Buffering in backbone routers is necessary for a different reason.  They
> may have multiple igress and multiple egress interfaces.  At any point
> in time it is possible to have:
>
> 	sum(input-rates) > capacity of the output link
>
> So backbone routers also need buffering.  Even though this does not
> explcitily have anything to do with BW*Delay product, it has to be
> proportional to BW in order to proive buffering for a certain length of
> time.


I more than agree, again.

But the original Long's question still holds: what is a typical router
queue length?

I believe that this is a field where assumptions should not be necessary
at all! Router queues are implemented by pieces of code, written by real
programmers and engineers. Why it is so hard to get a simple and objective
answer to such a precise question?

Ok, if the answer is "it depends on the particular implementation, or
interface speed, or layer 2 technology", I would be more than happy to
read just a few real life examples.


My personal assumptions:

a) the router software assigns, for each outgoing interface, a relatively
   small buffer space, just enough to hold N packtes in queue ( N a number
   between 10 and 100);

b) the buffer length is constant and independent of the traffic load. No
   attempt is made by the router to increase the buffer length under heavy
   load or when it senses (?) a long link with large RTT;

c) the maximum queue length probably independs of the interface
   speed or other particular characteristic;

d) No attempt is made to avoid packet loss on congested links by putting
   more buffer space for queueing on it.



Alexandre.