[Click] ns2click + performance measures
Michael Neufeld
neufeldm at cs.colorado.edu
Mon Mar 29 08:30:21 EST 2004
The tracefile information you're getting from nsclick should be close to what
you'd get using regular ns-2, the only real weakness being for packets which
are generated from within your click graph. Those packets will show up simply
as "raw" packets without much in the way of annotation, and I'm guessing
that's the problem you're having? If you just want to look at statistics for
data packets, why not just go to using the ns-2 "cbr" traffic generator? That
should play nice with the tracefiles and it should be relatively easy to get
the numbers you want. If you want to grab statistics about control packets
generated by your click graph you're either going to have to write custom
code in ns-2 to dump appropriate information to the tracefile, or go straight
from tcpdump files, as it sounds like you're doing. You shouldn't be seeing
packets delivered before they're sent, the simulated time starts out at UNIX
time "0" and goes up from there. Are you setting the timestamp annotation on
the packets before they get dumped? Also, do you have an unequivocal way of
identifying packets in the tcpdump files, even after they've gone over the
network? Without a sequence number or some kind of absolutely unique
identifier you're going to have problems keeping track of things. We've
usually just stuffed a source-unique sequence number into packet payloads
from each source -- combining that with the packet type and
source/destination IP address were usually enough information to allow
tracking of packets.
Granted, nsclick isn't the absolute best simulation tool around, but I've
found that it does OK at what it does, and I haven't been able to find
something to completely replace it yet. If you've got any specific
suggestions or patches for bug fixes or features that would make it a more
effective tool please let me know - I'm generally pretty happy to fix bugs
and add useful features to it.
-Mike
On Monday 29 March 2004 01:52 am, Michael Voorhaen wrote:
> Hi,
>
> I'm trying to perform some simulations using ns2click and wanted to
> gather some performance measures. I've tried looking at the ns-trace
> file, but all I get are MAC layer statements. I also keep some ethereal
> dumps for each node, from which I can get some info e.g. Packet delivery
> ratio is easy enough to compute, since I just need to compare the right
> files. However end-to-end delay is getting a little bit more difficult
> since some nodes are apparently receiving packets before they are
> actually sent. Is this due to how ns-2 works?
>
> I know that ns2click probably isn't the best simulation tool, but the
> code that I am writing now is going to be deployed on an actual testbed
> later, and I would just like to get some results as soon as possible.
>
> Can anyone maybe give me some pointer on how I should proceed?
>
> Kind Regards,
> Michael
> _______________________________________________
> click mailing list
> click at amsterdam.lcs.mit.edu
> https://amsterdam.lcs.mit.edu/mailman/listinfo/click
More information about the click
mailing list