[Ns-developers] Further ideas to improve my GSoC proposal (ns-3 in the cloud)

[Ruhao Yao]

Professor Henderson,


After reading through Peter Barnes¡¯s paper and ppt, I have some ideas to
improve my application.



Parallel Example Adjustment

Comparing to Peter Barnes¡¯s benchmark model, I think my first example is
too nai"ve. So I would like to focus more on implementing Peter¡¯s
benchmark model (specific Random Watts-Strogatz graph). I think I can
implement this model on AWS, but there few challenges to overcome:

  * Critical Path Implementation

Trigger CP computation at configure time

Aggregate m_currentPathLength, m_nEvents to nodes.

Hook Send Callbacks in Netdevice (ReceiveCallback already exists.)

  * The specific random Watts-Stogatz graph model

Setup this model with pre-configured parameters.

Check the correctness.

Partitioning. Peter noted 3 ways of partitioning (Random, Caterpillar
and Optimized), I will consider to choose one or more proper ways to
implement partitioning.

  * Benchmarking of ns-3 parallel performance with this model

Evaluate the parallel performance compared to serial performance.

Evaluate P2P link delay (LA time) impact to parallel performance

Evaluate the synchronization impact to parallel performance

About the internet topology generator

Thanks to Lalith¡¯s help, I learnt Ns3_BRITE_Integration. Due to the
following reasons, I think it¡¯s not appropriate for my benchmark model.

  * The model should be partitioned to different LPs. After reading
    through the BRITE_Integration configuration wiki, I think it can not
    support partitioning.

  * The Random Watts-Stogatz graph is much more complex. For my
    benchmark model, I need to configure many parameters to make the
    graph work. So I think BRITE can not help with it.

I think my second parallel example can make up the disadvantages of
Peter¡¯s benchmark model:

  * With CSMA and Point-to-Point link support, it can add router support.

  * Packets are not only transferred to neighbor node.

  * Evaluate the CSMA and Point-to-Point link performance on parallel
    simulation, it may be a start point to add the CSMA and wireless
    distributed support.

Sorry for response late, I just got back to school 3 days ago. If you
have any confusion, please let me know. Thanks so much!


Regards.

-- 
Ruhao Yao
Beijing Key Laboratory of Internet Security Technology
School of Electronics Engineering & Computer Science
Peking University
Beijing 100871
China