[Simgrid-user] Simulate a Centralized Broker

Patrick K. maillists at kfpc.de
Fri Jan 27 14:00:56 CET 2012

Hello Jean-Noel,

thanks for your answer.

On 2012-01-27 11:33 AM, imag wrote:
> Hi,
> On Jan 27, 2012, at 10:05 AM, Arnaud Legrand wrote:
>> Hi,
>> On 24/01/2012 16:21, Patrick K. wrote:
>>> - TGs/DAG => I already have a lot of generated TGs for testing in 
>>> simple
>>> flat-files. I think i could load the important parameters from the 
>>> files
>>> with an parser into SimGrid!?
>> Jean-Noël Quintin wrote a dax loader and a dot loader. Look at 
>> SD_daxload and SD_dotload in 
>> http://simgrid.gforge.inria.fr/simgrid/3.7/doc/group__SD__simulation.html. 
>> Also loot at examples/simdag/dax and examples/simdag/dot.

The DAGs/TGs that i should use has the STG-File Format described at:


Also the Resource-Broker breaks down the Tasks in the TGs and handles 
them seperately (gives them an individual start and end time, 
distributes them to several Grid Resources taking account the 
dependencies between the tasks).

Because of this, i think the SimDiag is not the right API for me!? I 
have to do this with the MSG API, programming there the intelligence of 
the Resource Broker and the SLA and the AR between the Grid Resources 
and the Resource Broker. Or is there a way to load the tasks from the 
TGs files with SimDiag and then moves them to a task object from the MSG 



> I just want to add something about the dotloader. I have although add
> the possibility to detail a schedule inside the dot file.
> You could compute your schedule before the simulation and just put
> the information inside the dot.
> For you, it could have one problem with this solution. The tasks will
> be executed as soon as possible in function of the dependencies.
> May be you want to start a task only after some idle time (because
> the task is not already added by the user).
> And if you find some bugs inside this module, let me know I would
> correct it as soon as possible (I was alone to use this, I think).
> Best Regards,
> Jean-Noël
>> If you work on DAG scheduling, you may want to have a look at Daniel 
>> Cordeiro, Grégory Mounié, Swann Perarnau, Denis Trystram, Jean-Marc 
>> Vincent, Frédéric Wagner. Random graph generation for scheduling 
>> simulations. Proceedings of 3rd International ICST Conference on 
>> Simulation Tools and Techniques (SIMUTools 2010), :10, Malaga Spain, 
>> mar 2010. They wrote a webpage (http://ggen.ligforge.imag.fr/) on DAG 
>> generation. That's not a DAG repository but it is complementary to 
>> workload archives.
>> If you have other sources of workload and are ready to share...
>>> - Independent jobs => I could download some workloads traces (from 
>>> the
>>> parallel workload archive for example). Again a parser!?
>> I don't have any but that seems doable in a few lines of C. Maybe 
>> Frédéric Suter and Jean-Noël did this at some point.
>>> - Central Resource Broker => I think i have to write the interna of 
>>> the
>>> central resource broker and of course the logic how it communicates 
>>> with
>>> the grid resources. Could i use here the SimGrid APIs?
>>> - Grid-Resource => Communication part of the grid-resource, that 
>>> provide
>>> me AR and SLA. Yes that also my work, with or without the SimGrid 
>>> APIs!?
>>> - Network-bandwith between Broker and Resources => Could SimGrid
>>> simulate me this?
>> Sure. Writing such things with MSG should not be too painful.
>>> - Platform => The Grid-Resources should have the parameters from a 
>>> real
>>> grid e.g. NorduGrid or EU-DataGrid. Could SimGrid simulate me this, 
>>> if i
>>> provide an corresponding platform file?
>> Yep. Although we do not have nice descriptions of such platforms. 
>> But if you come up with such informations, we'll be glad to help you 
>> expressing it with SimGrid.
>>>> I only know the MSG API, so I won't advice you for MSG or SimDAG.
>>>> I'll let the others give you recommendations.
>>> Okay, let's wait ;).
>> Regarding, simbatch and garsim, we did not migrate them to our git 
>> repos as there was no activity in these projects. I would advise you 
>> to have a look at it to get some inspiration but maybe to take a fresh 
>> new start.
>> Best regards,
>>  Arnaud
>> _______________________________________________
>> Simgrid-user mailing list
>> Simgrid-user at lists.gforge.inria.fr
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/simgrid-user
> _______________________________________________
> Simgrid-user mailing list
> Simgrid-user at lists.gforge.inria.fr
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/simgrid-user

More information about the Simgrid-user mailing list