[Iccrg] Re: updated CTCP draft
lachlan.andrew at gmail.com
Wed Nov 21 03:56:31 GMT 2007
I've finally started reading your draft -- sorry for the delay.
I'll get back to you once I've read it properly, but here are a couple
of preliminary points:
- I don't think it can be declared "safe" yet, since the "Control Law"
section doesn't specify alpha. A later section says "we use
alpha=1/8", but the draft doesn't say that that is part of the
*requirements* of the algorithm. I could implement this with
alpha=10000 and claim that it satisfies the protocol description. If
you want to say these parameters are tunable, you might want to
specify a range in which they can be varied, and specify a "default"
value in the algorithm description.
- Someone recommended using MUST, SHOULD, RECOMMENDED, MAY etc
according to the normal practice for normative RFCs. I'd say you
"SHOULD" do that (you don't have to if there is a good reason not to,
but otherwise please do it).
- It seems that dwnd can become negative if diff<gamma. In
particular, if dwnd(t)=0, then dwnd(t+1) = -1. Is that what you
mean? In an RFC, I believe it is important to be very precise about
things like clipping.
- I'd much rather see a separation of algorithm/implementation from
justification/motivation. As one of many examples, discussion of
fairness doesn't (to me) belong in the section defining the algorithm.
This will also make clearer whether "automatic setting of gamma" is
part of the algorithm or not. Unlike the other points, this doesn't
affect whether or not it is "safe", so it is up to you, but I think it
will make the document much easier to use.
On 30/10/2007, Murari Sridharan <muraris at microsoft.com> wrote:
> Lachlan you must have seen my mail to iccrg on the updated draft, this had a
> few more cosmetic changes, some more cleanups than the doc I sent out, but
> from a technical stand point it's the same.
Lachlan Andrew Dept of Computer Science, Caltech
1200 E California Blvd, Mail Code 256-80, Pasadena CA 91125, USA
Ph: +1 (626) 395-8820 Fax: +1 (626) 568-3603
More information about the Iccrg