Re: question about flowmon

[Lloyd Wood <L.Wood@surrey.ac.uk>]

On Fri, 19 Nov 1999, Guo, Liang wrote:

> I apologize if this is an old question, but I couldn't find an answer from
> the mailing list archive. I want to monitor a specific flow, say flow 10,
> there are two suggested ways: 1. call trace-all and post process the trace
> file to get flow 10's behavior. 2. attach a flowmonitor on one of the
> links that flow 10 pass through.
> 
> The problem is, the first method consumes too much disk space,

even using trace-all with e.g.:

set tracefile [open "|gzip -c > out.tr.gz" w]

? You don't _have_ to postprocess; you could do something like:

set tracefile [open "fgrep r | {awk -f some script here to narrow it \
   down further} | gzip -c > out.tr.gz" w]

It's not as elegant as a per-flow monitor controlled from Tcl; if I
had to monitor a specific flow I'd probably just cheat by putting a
test of f->flowid() in FlowMon::dumpflows before dumpflow is called,
although that's hardly optimal and you do have to run some simulation
first to figure out which flowid you want. See ns_doc 18.7 (or 19.7,
if someone will update the pre-generated ns_doc.)

hth,

L.


> while the
> second method consumes too much memory because it will record every flow
> that passed throught that link. 
> 
> Has anyone done similar things before? I mean monitoring a single flow. Do
> I need to write up another piece of code to do that? Appreciated for any
> advice.
> 
> Guo, Liang.

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>