[Iccrg] Proposal for ICCRG operations

[Wesley Eddy]

On Fri, Feb 03, 2006 at 09:32:27AM +0100, Michael Welzl wrote:
> So, the index should contain:
> 
> * normalized throughput
> * loss
> * fairness (let's use the Jain Fairness Index)
> 
> It could then look something like this:
> 
> WI = (T*t + L*l + F*f) / 3
> 
> where WI is "Welzl Index" (tm)   ;-)
> T is normalized throughput
> L is normalized loss
> F is the Jain fairness index (which is already a value
> in the 0..1 range)
> 
> and t, l, f are weighting factors which decide how
> important the parameters are to us.
>
> ...

It would be nice to have a simple scheme, but I don't think that a simple
summation-based scheme is going to suffice.  In addition to the problems
Michael described with this simple scheme, here are a couple more:

1) no notion of how stability factors in
2) no notion of convergence rate
3) no notion of what happens to a mix of different types of flows
4) ranking is ambiguous in that it gives identical results for very differnt
   inputs

There are ways to cobble in or correct for these, but I don't know how to
meaningfully do it all at once.  In previous work on RED, Mark Allman and
I started with an index based on summation like the one you propose, and
we found problem 4 irritating.  We were only looking at two inputs to the
rating function -- utilization (U) and Jain's index (F), both from 0 to
1.  We decided that an index computed using  U^alpha * F^(1-alpha)
worked much better for our purposes, with alpha used to weight preferences.
This is described in a pretty obscure paper called "A Comparison of RED's
Byte and Packet Modes" from Computer Networks volume 42, issue 2, in June
2003.

-Wes

-- 
Wesley M. Eddy
Verizon Federal Network Systems
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://oakham.cs.ucl.ac.uk/pipermail/iccrg/attachments/20060203/8f702c62/attachment.bin