Ip Addressing in fat-tree

[avon yan]

Hi, everybody!

I have some problem about fattree simulation.

1. In fattree topology, each node has its own ip, like 10.pod.switch.ID. However, while simulating, each node has several interfaces, and each interface should have its own ip, so how can I make all the interfaces in one node the same ip, like 10.pod.switch.ID?

2. What should I do if I want to simulate the fattree routing?

Looking farword to your help!

[Tommaso Pecorella]

Hi,

"10.pod.switch.ID" is not an IP number. Mind explaining a bit better your problem and your goal ?

Thanks,

T.

[avon yan]

well, in the paper "A Scalable, Commodity Data Center Network Architecture", it said:

We allocate all the IP addresses in the network within the private10.0.0.0/8 block. We follow the familiar quad-dotted form with the following conditions: The pod switches are given addresses of the form 10.pod.switch.1, where pod denotes the pod number (in[0, k − 1]), and switch denotes the position of that switch in the pod (in [0, k−1], starting from left to right, bottom to top). We give core switches addresses of the form 10.k.j.i, where j and i denote that switch’s coordinates in the (k/2)2 core switch grid (each in
[1, (k/2)], starting from top-left). The address of a host follows from the pod switch it is connected to; hosts have addresses of the form: 10.pod.switch.ID, where ID is the host’s position in that subnet (in [2, k/2+1], starting from left to right). Therefore, each lower-level switch is responsible for a/24 subnet of k/2 hosts (fork < 256).

That is how ip is assigned in fattree.

在 2014年12月11日星期四UTC+8下午6时01分58秒，Tommaso Pecorella写道：

[AncientMoon]

Hello, I have the same problem while simulating the fat-tree. Have you
solved this problem? Would you mind helping me with this? Thank you very
much!

[Tommaso Pecorella]

Hi,

let me get this straight: if I was among the reviewers, that paper wouldn't have been published. Ever.

This is not because the idea is wrong. The terminology is wrong.
In the whole paper the authors calls "switch" what is, for real, a router. Call them routers and I have no objections. Call them switches and I'll happily kick your asses 'til tomorrow.

As a rule of thumb: a switch doesn't have an IP address. If it does have one, it's just for management purposes. It must never, ever, look beyond MAC addresses or MAC-level tags.

The second thing I severely dislike in that paper is that the authors claims that they can use "commodity" switches (and they're routers) and then they add a friggin' extra routing table element. Like if this was an easy and cheap thing. Anyway, it's a scientific paper and one can dream (if the reviewers doesn't wake you up).

Indeed, the authors used Click to simulate / implement their setup... good choice - it's not exactly commodity.

About how to assign the addresses in ns-3, I really don't see the problem. Just follow the scheme the authors used. On the contrary, simulating the "switch" (it's a router) will be quite a job.

Have fun,

T.

[古月永日]

 Well, thank you very much for your reply. I've got the your notion.
 Actually, in my idea, one IP can be only assigned to one unique interface in ns-3 . However, in that paper, they seem to give one "switch",  your mean router, only one IP address.  While it seems not feasible, since each "switch" can have k interfaces. So, I want to know how to assign IP addresses to these interfaces in ns-3 according to that paper.

在 2015年5月5日星期二 UTC+8下午4:06:50，Tommaso Pecorella写道：

[Tommaso Pecorella]

Hi,

one (rather brutal) way is to bridge the interfaces. In this way you'll have just one IP address.

The best way would be to develop a more realistic switch model, or to use...click (yes, there's a click ns-3 module, but I have never tested it).

Have fun,

T.

[古月永日]

OK, thank you very much. I'll try it out.