[ns] Possible scheduler granularity problem at large-scale simulations

[Andreas Alexelis <alexelis@acr.atr.co.jp>]

    Hi everybody,

I am working on a QoS scheme over diffserv. I was testing scalability
issues, and in order to do that I tested it on large scale networks.
Until arount 500 end-hosts, injecting a mixture of CBR-TCP traffic in
the network from 1Mbps access lines, everything was going OK. So I put a
bash script for gradually increasing simulations to collect some results
and go home to get some sleep ( one needs to from time to time ;-).
Today I come back and find my 1000-node-topology log file, overflown by
"Scheduler scheduling events with negative delay" messages, but the
simulation was continuing, no segfaults-nomemory leak. Of course, the
resaults taken re not trust worth so I had to throw them, but I intend
to run it again tonight so:

-Is it possible to have hit granularity limits, or should I check my
code once again?
-Does anyone have any experience with such problems? I remember I saw a
simmilar question in the mailing list a while ago .
-Is there a quick way to fix this or I have to make do with less?

Thanks in advance

--
Andreas Alexelis, Lab.1 ACR,
Advanced Telecommunications Research, Japan
E-mail: alexelis@acr.atr.co.jp