[SimGrid-user] simdag: platform design - cloud with a common storage center, routing problem

Frédéric Suter frederic.suter at cc.in2p3.fr
Fri Jun 9 19:25:37 CEST 2017


Hi Aurélie,

You might be able to implement the setting you want by writing the right 
platform file. In fact it really depends on whether your storage center 
has "to do something" during the communications or not. If it hasn't, 
you can write a platform such that the routes between every pair of node 
(or cluster) goes through this central "node" (which can be what we call 
a router).

Another thing that has an impact on your platform file is whether 
communications within a cluster also have to go through this storage 
center or not. From what you asked, I understand that it is not the 
case. Have a look at examples/platforms/dogbone.xml, it will give an 
idea of what such a platform file could look like.

hope it helps

F.


Le 07/06/2017 à 07:19, Aurélie Kong Win Chang a écrit :
> Hello,
>
> I am implementing a simulator to do experiments on scheduling of 
> scientific workflows on a cloud-typed platform. I use simDag to manage 
> the dependencies between tasks.
>
> In my model, there is no peer-to-peer, all communications have to go 
> through a storage center. But I don't know how to force the 
> communications to do this. We found two solutions to handle this:
>      - model the cloud as multiple clusters and manually apply a *2 
> factor to all communications (which is what I am using for now) -- 
> This supposes/models implicit distinct yet comparable up and down 
> links between each cabinet and the host
>      - model the cloud as multiple clusters and a separate host (the 
> host will be the storage center, we don't take account of problem of 
> accessibility or time to read/write, we are only interested in 
> communication time). Then create in the workflow, between every pair 
> of tasks with a link of dependence, a false task of amount of work of 
> zero, which we would place on the "storage-center" host
>
> But to us, it's more workarounds there. Is there a better way to do 
> this ?
> Thanks by advance,
>
> Aurélie Kong Win Chang
>
> _______________________________________________
> Simgrid-user mailing list
> Simgrid-user at lists.gforge.inria.fr
> https://lists.gforge.inria.fr/mailman/listinfo/simgrid-user

-- 
One should, for example, be able to see that things are hopeless
and yet be determined to make them otherwise.
                                           -- F. Scott Fitzgerald



More information about the Simgrid-user mailing list