🚀 go-pugleaf

John Navas (spamfilter@navasgrp.dublin.ca.us) wrote:
>
> In <95sflb$8at@nfs0.sdrc.com>, scjones@thor.sdrc.com (Larry Jones)
> wrote:
> >
> >I think you're being unduely pessimistic (or do you know something I
> >don't?).  You seem to be assuming that the CMTS will allocate upstream
> >minislots on a first-come first-served basis, but if I were writing the
> >code, I'd certainly use a priority-based scheme where small requests
> >take precedence over large requests.
>
> I think that would take (1) special code in the modem or (2) quite a bit
> of FIFO buffering in the CMTS.

Why?  The modem already has to know how many minislots to request in
order to send the data that it has pending, and the CMTS has to buffer
requests for upstream bandwidth until such time as they can be
satisfied.  All that's required is to make that buffer a priority queue
that sorts the requests by number of minislots rather than a pure FIFO.

> I see no evidence of either, either in
> the specs or in the available implementations.  (The CMTS doesn't know
> whether the modem wants to send an ACK or a data packet; for that
> matter, the modem doesn't know either.)  More to the point, real world
> behavior works as I've described.

The point is that ACKs are (usually) small, data packets are (usually)
big, and any constant stream of upstream data that would saturate the
link will, assuming a reasonable amount of buffering the CM, cause the
CM to be requesting a large number of minislots for upstream data. 
Since the CMTS already has to allocate the upstream bandwidth and deal
with oversubscription, it seems like this would be an obvious way to
keep "bandwidth hogs" from screwing everyone's performance.  If the real
world doesn't work that way, I wonder why not.

-Larry Jones

I don't think math is a science, I think it's a religion. -- Calvin