[ns] Re: network simulation software?
Tue, 27 Mar 2001 13:55:04 +0100 (BST)
On Mon, 26 Mar 2001, Scott Michel wrote:
> On Fri, Mar 23, 2001 at 06:27:05PM -0800, Ping Pan wrote:
> > Try ns-2: http://www.isi.edu/nsnam/ns/. It's quite easy.
> Uhhh... yeah. Right. Sure. This statement should be taken with a great
> grain of salt and requires proper investigation of the ns-users mailing
> list for newbie FAQs.
of which there is just one, saying 'read the webpages' for the most
> Depends on what level you want to simulate. If you're at layers 3 and
> below, it's good. If you're above layer 3, it can be a righteous burr
> under the saddle.
Other way around, surely?
Over the past few years ns has been growing down the stack and
gradually fleshing out some very limited and abstracted MAC/physical
support. You still can't do multicast over wireless, for example -
nothing over layer 3 in that - and the split between disparate sets of
C++ routing code and OTcl routing code is unfortunate.
The strength of ns lies in its one-way TCP suite - and that's layer 4.
Many academic ns simulation papers have the complex basic topology of:
^ some fancy queuing thing here
and don't touch anything below layer 3; links are presumed serial.
Segmentation and reassembly and header/packet size fudgings throw off
stuff below layer 3 entirely... the amount of adjustment needed to
e.g. generate meaningful BER curves for TCP traffic from propagation
effects upward is significant.
You only think ns is difficult for above layer 3 because that's all
you're trying to use it for. Let me assure you that if you try to use
ns for work above -and- below layer 3 you'll have a much more
if you want to use ns for TCP stuff across serial links, you'll
probably be fine. Beyond that, all bets are off; ns has a lot of
built-in abstractions that get in the way of using it for engineering
with real physical quantities at the lower layers, and you have to
appreciate its limitations to get useful results out.