Re: [e2e] opening multiple TCP connections getting popular

From: Bob Briscoe <rbriscoe(at)jungle.bt.co.uk>
Date: Sat Sep 01 2007 - 05:02:37 EDT

Joe,

I composed responses yesterday to each of your points, but I've realised there's no purpose in sending them until the underlying terms of reference are mended...

At 08:00 31/08/2007, Joe Touch wrote:
>Bob (et al),

---snip---
> >> (and, as others have noted, it's not
> >> clear we _need_ to do anything).
> >
> > Tell that to the CEO of any network operator. I think you're saying
> > there's not an engineering problem. But that's because resource
> > allocation problems are economic problems not engineering problems...
>
>No - I'm saying it isn't a problem. Whether it's economic or otherwise,
>as others have noted, not being 'fair' depends on an agreed concept of
>fairness. We have one now - per-TCP connection, relative to RTT. That's
>not perfect, but it is a definition, and it does work.

Well, most people who have entered this debate (mosly on tsvwg) have tried to distance themselves from ever having been definite about per-TCP connection/RTT as the agreed measure.

Of course there's not a problem... if you're measuring it the way you are... but that's because the measure you're using isn't a relevant measure - fairness is a social science issue so you need a social or economic measure. Otherwise your judgement of whether 'it works' is circular...

Imagine a country had a tax system based on the number of transactions entering and leaving a person's bank accounts. * I'm saying such a metric doesn't produce a fair taxation system (e.g. high earners stuff more money through less transactions). * By analogy, by sticking with transaction count as a measure, you're led to argue that there's not a problem. You can cite surveys of the transaction count into people's bank accounts that show people are being taxed in fair proportion to this count, and further, you're led to say that there can't be a problem because the spread of the count of transactions between most and least is pretty small.

If you look instead at the different volumes of congestion that users cause against how much each is contributing to the system, you should be able to see there is a huge problem. There's also a far greater spread of congestion volume caused by different users than is warranted by any value they get from causing it - many orders of magnitude between the lowest quartile and the highest, but the spread of contributions is probably within an order of magnitude.

It's not just flow rate. Flow rate must be weighted by the prevailing congestion at each instant and the result accumulated over /time/. Then attributed to each /economic entity/. That's congestion volume per 'user'.

And, if we go straight to congestion volume without passing through flow rate, the resulting accountability system is really simple.

Ditch flows.

Bob

---

Bob Briscoe, <bob.briscoe@bt.com> Networks Research Centre, BT Research B54/77 Adastral Park,Martlesham Heath,Ipswich,IP5 3RE,UK. +44 1473 645196 Received on Sat Sep 1 05:47:05 2007