<<< Chronological >>> Author Index    Subject Index <<< Threads >>>

Re: "scoring" of results


Jacques,

> Following the discussion at RIPE35 about "scoring" the results, I
> would like make the following suggestions:

Thanks for your suggestions, really appreciated,


> 1. Keep the results for latency and packet loss separate. This gives a 
> finer notation.

Yes and no.  I think we should have 1 number per connection that
summarizes 3 effects: delay, spread in delay and loss, if not, we'll end
up with the same problem as we have today: too many numbers and plots.  
Of course, the individual contributions to the scores will be made
available, so when score looks bad and one wants to investigate a
potential problem, one can take a step back, look at which effect causes
the problem, then look at the individual plot.


> 2. Divide latency by the "real" distance between hosts. Given the fact that 
> they all have a GPS receiver, they should be able to give very precise 
> location, can't they? This would make sure that test-boxes away from the 
> larger groups don't get "bad" scores.

Yes, this is (almost) what we want to do.  We're thinking along the lines
of defining a "best possible delay" equal to something like:

  (d_0 + distance/c) 

with d_0 the delay that one would get for a path consisting of 

  test-box #1 --> router #1 --> router#2 --> test-box #2

with no long cables between anything.  This is typically of the order of a
few ms. Distance is the distance between the two boxes along the greater
circle route, c the speed of light (in fibre/copper?).

This is the best one can do for any given setup.  (One can do slightly
better in a lab setup, just take 2 boxes with a piece of ethernet in
between, but this is not a setup one will ever find in practice).

The delay in the score will be measured in terms of this "best possible
delay", so 2 boxes far away from eachother as well as 2 boxes at nearby
locations, can both get the maximum score.  2 boxes close to each other,
but with routing via a much longer path than necessary (for example, we've
seen a case where 2 boxes in northern Europe 500 km apart, had their
traffic routed via New York), will get a bad score.

Finally, the boxes know their position (after a while) with an accuracy of
about 25m in each direction.

Henk

------------------------------------------------------------------------------
Henk Uijterwaal                    Email: henk.uijterwaal@localhost
RIPE Network Coordination Centre     WWW: http://www.ripe.net/home/henk
Singel 258                         Phone: +31.20.535-4414,  Fax -4445
1016 AB Amsterdam                   Home: +31.20.4195305
The Netherlands                   Mobile: +31.6.55861746  
------------------------------------------------------------------------------

The Committee (...) was unable to reach a consensus that substantial merit was
lacking. Thus, the appeal was deemed meritorious.          (Orlando NABC #19).





<<< Chronological >>> Author    Subject <<< Threads >>>