[e2e] Open the floodgate - back to 1st principles

[Guy T Almes]

David,
  A very good question.
  I will attempt an answer, but those who were active in this area about 
ten years ago should chime in.
  My impression is that, at least during the early 1990s and probably since 
then, there was a rule of thumb that a router should have a delay-bandwidth 
worth of memory per output port.  This was understood to be friendly to TCP 
in that it would allow the buffer to drain while the TCP sender recovered 
itself from a stumble following the bursting of the queue.
  This was during the time when high-speed wide-area meant T3 across the 
country.
  This is not adequate to achieve the purpose then intended, at least not 
in what would now pass as a high-speed wide-area path.  But it does add to 
router cost.  I am honestly not sure if this rule of thumb is being 
remembered correctly or if router designers examine it critically.

  Regards,
        -- Guy

--On Sunday, April 25, 2004 21:44:57 -0400 "David P. Reed" 
<dpreed at reed.com> wrote:

> At 03:03 PM 4/25/2004, Guy T Almes wrote:
>> Router designers are told to provide big buffers to allow these queues
>> to  get big.
>
> Why in the world should we encourage these queues to get big?
>
> There is a big difference between putting memory in so the queues can get
> big as the result of transient congestion, and encouraging routers to
> operate with these big queues full!
>
> Any control theorist will tell you that if you fill up all of the buffers
> in the network, there is no way to control it stably.   (a simple
> simulation will demonstrate it as well).
>
> It's like trying to drive a car at high speed from the back seat with
> rubber bands tied onto the steering wheel.
>
>