[Tsvwg] Re: [e2e] Are you interested in TOEs and related issues

[Marko Zec]

On Sunday 07 March 2004 23:06, Alexandre L. Grojsgold wrote:
> > The problem is that in today's routers the buffer sizes are
> > typically accounted in packets, not in bytes. So looking at your
> > example, and supposing that the line card buffer is limited to 100
> > packets, a burst of small frames will instantly consume 13% of
> > available buffer slots, while jumbos will only use 2%. This is of
> > course no problem if the line card can transmit all those frames
> > instantly, but what if it cannot?
>
> So what?
>
> Small packets get more buffer frames, but are also more rapidly sent.
> The gueue grows fats, but shrinks fast too. It means that a burst of
> small packets will not increase the packet loss, no mather the length
> of the buffer (in packets).


This theory is unfortunately wrong, because it ignores the fact that 
when congestions do occur, traffic flows consisting of small packets 
(regardless whether they are bursty or smooth) will use more buffer 
slots then the flows carrying the larger ones, leaving less remaining 
buffer slots available for queuing further packets / bursts.


> Ok, if the router gets a large amount of packets, representing more
> bytes per second than the capacity of the output line, packets will
> be lost. But it is a function of the amount of data conveyed by the
> packets, not of their individual length.


Can you define the function you are talking about more precisely? What 
does it represent (I assume packet loss ratio)? Which argument(s) does 
it take? If this function is as simple as it might sound from your 
statement, why don't you share it with us?


>
> > > >> BTW, I asked a few transport folks in Minneapolis IETF about
> > > >> how "evil" is traffic burst in today's enviroment, but did not
> > > >> get any concrete answer. Perhaps this topic should be
> > > >> discussed in tsvwg or tcpm.
> > > >
> > > >Because queues in todays routers have finite maximum lengths,
> > > > and this model is unlikely to change in the forseeable future,
> > > > excessive traffic bursts will be more likely subject to
> > > > drop-tail policing than other kinds of more smoothly shaped
> > > > traffic. More than that, the bursty traffic will not only have
> > > > less chance of reaching its target with all fragments in place,
> > > > but it will also most probably do much harm to
>
> I think many people tend to thing about IP routers like old circuit
> switched protocols switches, like X.25.
>
> In circuit switched networks, flow control is generaly implemented in
> each link, each pair of switches exchanging acks and naks, trying to
> keep data at the right pace.
>
> With IP, the flow control is end-to-end, performed by TCP. If a
> router queues too many packets, it will break TCP stability. It is
> not an issue of memory or memory price.


Agreed. And that's precisely the reason why bursty traffic is bad - 
since routers are generally tuned to queue only as few packets as 
feasible for normal TCP operation, excessive traffic bursts will have 
more chance of being tail-dropped at router queues.

Marko