[Ns-developers] GSoC 2012: PackMime-HTTP project discussion

Yufei Cheng yfcheng at ittc.ku.edu
Wed Mar 21 08:38:45 PDT 2012

Hello Dizhi,

I have implemented one HTTP traffic generator in ns-3, and you can refer 
to code review # 4940041 for more details.  The distribution functions I 
used to generate connection involved parameters are based on the results 
from "Stochastic Models for Generating Synthetic HTTP Source Traffic" 
and PackMimi-HTTP project.

Like Dr. Weigle mentioned, the distribution generation model is based on 
trace results from Bell lab about ten years ago, which means the traffic 
generated mimic the traffic characteristics at that time.  I would be 
very interested in current tracing results and tune the distribution 
functions somehow to represent current web traffic.

Please refer to the paper below for more details about the traffic model 
we implemented, and we are very open to any feedback or cooperation from 
you for additional features.  Also refer to ResiliNets Wiki at 
https://wiki.ittc.ku.edu/resilinets_wiki/index.php/Main_Page if you want 
to know more of our works.

	Address = {Las Vegas, NV},
	Author = {Yufei Cheng and Egemen K. \c{C}etinkaya and James P.G. Sterbenz},
	Booktitle = {Proceedings of the International Telemetering Conference ({ITC})},
	Month = {October},
	Title = {Performance Comparison of Routing Protocols for Transactional Traffic over Aeronautical Networks},
	Year = {2011}}


On 03/20/2012 09:58 PM, Dizhi Zhou wrote:
> Dear professor Michele,
> Thanks for your relay to my former question ( 
> http://mailman.isi.edu/pipermail/ns-developers/2012-March/010082.html )
> I carefully read the PackMime-HTTP related contents in the NS-2 
> manual, paper "Stochastic Models for Generating
> Synthetic HTTP Source Traffic" and "Tmix A Tool for Generating 
> Realistic Application Workloads in ns-2". I have some
> questions and rough ideas now and want to discuss with you.
> 1, Based on the manual and "Stochastic Models" paper, it seems that, 
> currently, PackMime-HTTP does not support pipeline
>     in HTTP/1.1.  I think we can add this feature in NS-3 version by 
> modifying distribution of inter-exchange time gaps and
>     server delay.
>          In your "stochastic model" paper, you mentioned that 
> "pipeline is largely supported only in server implementation".
>    However, in recently three years, more and more browsers support 
> pipeline, such as Chrome 18 and 
> Mozilla[http://en.wikipedia.org/wiki/HTTP_pipelining#Implementation_in_web_servers]. 
>    Do you think that supporting pipeline is necessary for NS-3 version 
> of PackMime-HTTP?
> 2, I also read some background knowledge of Tmix, in paper "Tmix A 
> Tool for Generating Realistic Application Workloads
>  in ns-2", ADU epoch does not include server delay which exist in 
> PackMime-HTTP. It seems that server will reply response
>  directly after receiving request. Are there any reasons for this 
> consideration in this paper?
> 3, Also in above Tmix paper, section 3, you mentioned that the pattern 
> of ADU exchange can have overlapped parts. If
>     I understand currently, is this the feature of pipeline in HTTP/1.1?
> 4, the last question is about the research method: currently, I'm 
> reading the background paper on PackMime-HTTP and Tmix.
>   After understanding most details, I will focus on the structure of 
> PackMime-HTTP in NS-2. At last, I will try to figure out  how
>   to migrate the PackMime-HTTP structure in the framework of NS-3. The 
> whole work may need around two weeks.
>   Here, my question is: do we need to write implementation details in 
> the proposal? For example, class structure, variable and
>   method in each class, etc....
>   Also, in your opinion, which kinds of specific feature in NS-3 can 
> we utilize to implement PackMime-HTTP here? I'm familiar with
>   NS-2, but I just complete the NS-3 tutorial now.
> Looking forward to your apply.
> With best regards
> Dizhi

Yufei Cheng
Graduate Research Assistant
Information and Telecommunications Technology Center
The University of Kansas

More information about the Ns-developers mailing list