[ndnSIM] ndnSIM Digest, Vol 8, Issue 17

[Alex Afanasyev]

Hi Tun,

There is definitely a lot of things to be improved.  Even without enabling NACKs, there could be an implementation of consumerWindow that replicates to some extent more aggressive versions of TCP, such as Cubic or Compound TCP.  Enabling NACKs basically enables explicit notifications from the network about available resources and again, a lot of can be improved here.

---
Alex

On Jul 21, 2013, at 8:57 PM, Tun Tun Oo <htunhtunu at gmail.com> wrote:

> Dear Alex,
> 
> Thanks a lot for your contribution.
> As you mention, by using the enable NACKs the result is improved.
> We have to do some tuning of consumerWindows app for better result as TCP.
> In this sense, some sample model such as limit-window, rtt-estimator, rtt-deviation, etc are required to improve.
> This just my thinking about it and I am not quite sure.
> Could you advice me what I am missing or mis-understanding.
> Thanks again.
> 
> Rgds
> TUN
> 
> Date: Thu, 18 Jul 2013 08:34:53 -0700
> From: Alex Afanasyev <alexander.afanasyev at ucla.edu>
> To: Tun Tun Oo <htunhtunu at gmail.com>
> Cc: ndnsim at lists.cs.ucla.edu
> Subject: Re: [ndnSIM] ConsumerWindow and Congestion
> Message-ID: <46F071A0-EBBA-4B68-83C8-D66FA0B29035 at ucla.edu>
> Content-Type: text/plain; charset="iso-8859-1"
> 
> Hi Tun,
> 
> The behavior you're observing is probably not wrong, it just that ConsumerWindow is not equal to TCP.  ConsumerWindow has very simplistic congestion-related logic implemented, so it causes such an oscillations and weirdnesses.  You may want to consider implementing a more aggressive and more TCP-like consumer, or you can also enable NACKs (EnableNACKs parameter for the forwarding strategy), which may improve (but probably not going to completely solve) your problem.
> 
> ---
> Alex
> 
> On Jul 17, 2013, at 8:07 AM, Tun Tun Oo <htunhtunu at gmail.com> wrote:
> 
> > Dear Alex
> >
> > I did an experiment using the 8-node bottleneck scenario with ConsumerWindows application. As topology, I arranged as similar to the 6-node bottleneck scenario.
> > C1\                                                        /P1
> >      \                                                    /
> >      /-------R1---------R2---------R3-------R4 \
> >    /                                                     \
> > C2                                                        P2
> >
> > All the access link are 10 Mbps/10 ms  and all routers are 1 Mpbs/10 ms.
> > I set the Size Attribute of ConsumerWindow as 2 Mbyte and other setting as default.
> > I found the some strange behaviors from graph-output
> >  First 3 sec simulation two consumers are working as usual.
> > Next 20 sec only one consumer (c1)can send and receive.
> > After 20 sec two consumers can send and receive.
> > From 23 sec to end of simulation (40 sec) only one consumer (c2) is sending and receiving.
> >
> > I would like to know your advice of what  is wrong in this experiment?
> > Thanks in advance.
> >
> > Regards
> >
> > TUN
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.lists.cs.ucla.edu/pipermail/ndnsim/attachments/20130723/00dd322c/attachment.html>