How to run ns3 on multi-cores?

[nahid]

Hi there,

I have a http client server simulation which I need to up and down link interfaces dynamically during simulation. The problem is that I need to run simulation with about 1000 client nodes and 100 server nodes. but with this order of nodes delays are really huge. So I tried to run ns3 based on what has been described here: https://www.nsnam.org/docs/models/html/distributed.html. I tried to run my nodes on different processes but that's not working for me cause I have a dynamic topology. can any one help me with that? I really need a good solution for my problem? 

Thanks.

[pdbarnes]

When setting up the distributed simulation just ignore the fact that links will be going down. Construct all links, and partition the graph as normal. Then set down specific links to arrange your starting configuration.

Peter

[pdbarnes]

I should add that if your topology with all links up is dense you’re going to need more memory, since you have to instantiate the entire topology on each compute rank. We’ve managed 700+ nodes in a full clique topology in 128GB. We’re working on an approach which will scale to arbitrary graph sizes, but aren’t very far along.

Peter

[nahid]

Instantiate the entire topology on each rank?? sorry can you explan that more??
An another this is that what about shared parameters my algorithm uses about nodes? how can all processes access the same value? (I can't find any method to broadcast values in NS3).

 Nahid

[pdbarnes]

Please study the manual section on distributed simulation:

https://www.nsnam.org/docs/models/html/distributed.html

For more background, including a brief intro to MPI, please watch latest distributed simulation tutorial, from WNS3 2016:

http://www.nsnam.org/tutorials/consortium16/slides/WNS3-Distributed-Simulation-2016-06-14.pptx

The prior year tutorial session is here (with video):

http://www.nsnam.org/tutorials/consortium15/WNS3-Distributed-Simulation-2015-05-13.pptx

https://vimeo.com/album/3480129

[nahid]

Thanks alot. That really helped me. I have run a topology of 800 nodes and it seems to be ok. but when I run it with 1000 nodes it terminate and gdb dosn't tell me what's the problem. can you help me with that please?

Thanks

[pdbarnes]

Please provide a lot more information. Does it exit normally? With an error code? Error message? Can you add some logging/printing to narrow down where the problem is? Initialization? Running? Shutdown? What’s your infrastructure? How many compute nodes? Ranks? MPI version? What’s the nature of your topology? How are you partitioning/assigning ranks?

Peter