3D torus with dynamic routing

[吴振伟]

Hi all, 

I wonder to simulate a Cluster Interconnection with 3D torus network topology and adopting some kind of dynamic routing method. I am trying to build a 3D torus network through coding, but I have not find an appropriate routing method. Currently, I adopt the Nix-Vector routing provided by ns3, but it is static rather than dynamic.

Thank you for your time!

[Tommaso Pecorella]

Hi,

try RIPng (it's for IPv6, but it will do the job).

You can also use GlobalRouting and periodically have it recalculate the routes. However... Nix is dynamic, not static !

As a consequence, the problem is solved at its root. Nix is dynamic.

Cheers,

T.

[吴振伟]

Thank you for your kindly answer! One more question is, if Nix is dynamic, why the usage example given in the document (shown below) using Nix along with static routing?

API and Usage

The Nix-vector routing protocol must be added to a list of routing protocols. It is important that list routing is utilized.

Example:

Ipv4NixVectorHelper nixRouting; Ipv4StaticRoutingHelper staticRouting;

Ipv4ListRoutingHelper list; list.Add (staticRouting, 0); list.Add (nixRouting, 10);

InternetStackHelper stack; stack.SetRoutingHelper (list); stack.Install (allNodes);

ref:https://www.nsnam.org/docs/release/3.18/doxygen/group__nixvectorrouting.html

在 2016年1月5日星期二 UTC+8上午3:29:59，Tommaso Pecorella写道：

[Tommaso Pecorella]

Hi,

I'd suggest to always use the latest documentation and the latest ns-3 release, we're at 3.24.1 right now, and you linked 3.18 documentation.

About the statement in the doc (it's the same also in the latest version), it is just stated that Nix can not be used "alone", and it must be used along with the StaticRouting.

You should consider that ns-3 nodes can have more than one active routing protocol: they'll work sequentially (the first one finding a route wins).

Anyway, I'll check the documentation and (eventually) I'll upgrade it.

Cheers,

T.

[Tommaso Pecorella]

Hi,

I just checked. The reason is that NixRouting doesn't handle the local delivery. It does assume that there's another protocol (Static in this case) with an higher precedence that will handle the local packet delivery.

That's why it needs StaticRouting.

Cheers,

T.

[吴振伟]

To evaluate the efficiency of dynamic routing, I call the IPv4::SetDown function to simulate the link failure. The dynamic routing method routes the packet successfully under the failure, however, the throughput stays same with the no failure situation. What may be the problem? Thank you!!!

在 2016年1月5日星期二 UTC+8上午9:10:28，Tommaso Pecorella写道：

[Tommaso Pecorella]

Hi,

Nix is quite fast at recovering (let's say almost immediate).

The changes in throughput are (in your case, like in most cases) dependent on the link bottlenecks and the routing recovery time. If the routing convergence time is fast - and for Nix is immediate - and there's no congested link in the "new" configuration, you'll hardly see any change in throughput.

Since routing efficiency seems to be important for your experiment, I'd strongly suggest to check what model you want to use. Perhaps Nix is too ideal for you. Chances are the you'll have to develop a new model.

Cheers,

T.

[吴振伟]

Thank you for your kindly help:) 

In my case, I set down the shortest path, any rerouted path provided by the adaptive routing algorithm should be longer than the failing path. Part of the network flows indeed shows delaySum increasing. However, there is also certain part of flows, the delaySum of which just dropped down under the link failure I injected. Until now, I have no idea to explain this phenomenon. 

Cheers,

WU ZHENWEI

在 2016年1月5日星期二 UTC+8下午7:14:52，Tommaso Pecorella写道：

[Tommaso Pecorella]

Hi,

well, that's what simulations are for... good luck in analyzing the results !

Cheers,

T.