Necdet,
Excuse me, simRes3 shows graphs for the
station just before station 0.
Here are the graphs for station 0:
I have a new STQsize formula that is very close
to the simulations. I still need
to test it more, but when the formula is
accurate in all cases, the duration
of the priority
inversion can be calculated.
Jon
----- Original Message -----
Sent: Monday, April 28, 2003 7:14
PM
Subject: Re: [RPRWG] Simulation
results
Jon,
stqSize(0) in this graph is a lot smaller than the one that you sent
earlier (simRes2.pdf). Is there a conversion problem here (byte vs bits etc)?
Another information that could be very valuable is how long is the priority
inversion when we fix the stq size to the values given in the draft. A delay
jitter histogram would be very useful. I am especially interested in jitter
values exceeding 2 x 256 x MTU time.
Thanks.
Necdet
Jon Schuringa wrote:
No problem, the
new graphs are here: http://www.ikn.tuwien.ac.at/~nthuma/simRes3.pdf Jon
----- Original Message -----
Sent: Monday, April 28, 2003 10:00
AM
Subject: RE: [RPRWG] Simulation
results Jon,Thanks for your interesting work. A couple of questions (if it is
not too much..:):Do you have the values of the
localFairRate for station 0 ? Can you make a graph that shows how this
value changes with time for the aggressive and the conservative modes
?Leon
Jon,
Thank you for the good work. It seems like our formulas in Annex G is
off quite a bit specially for conservative mode. Did you use the models
from AT&T? If there is no modeling error, we need to find out the
root cause of it and modify our formulas.
Thanks.
Necdet
Jon Schuringa wrote:
Yes, it is better to change only one parameter at a time. New
results with fixed 500km ring are here:http://www.ikn.tuwien.ac.at/~nthuma/simRes2.pdf
Jon
----- Original Message -----
Sent: Friday, April 25, 2003
12:10 AM
Subject: Re: [RPRWG] Simulation
results Jon,
Thank you for running these simulations. These are very valuable.
The only think that is not clear to me from the result is whether
the buffer requirement is linearly increased with number of nodes or
with the size of the ring. Can you please run the same simulations
for a fixed size ring say for 500km with varying number of nodes as
in your graphs? After that we can determine whether the number of
nodes affect the buffer requirements or not and based on that we can
modify the formulas if necessary.
Thanks.
Necdet
Jon Schuringa wrote:
Dear all, I finished the simulations about the
maximum stq buffer occupancy withfixed values for
stqLowThreshold and stqHighThreshold
andstqFullThreshold set to infinity, as Necdet proposed. You can get the results here:http://www.ikn.tuwien.ac.at/~nthuma/simRes.pdf
-
Jon
|