[Simgrid-user] questions for Simgrid

Arnaud Legrand Arnaud.Legrand at imag.fr
Thu Jun 16 18:42:57 CEST 2005


Hi,

Zhang Yuanyuan wrote:

> 1. In MSG, you provide a file named "surfxml.dtd" for platform
> description. Is there such file for creating tasks?

No, it's the user code that creates tasks on the fly. Think of MSG as a
virtual distributed computing environment in which you can run some code.
platform.xml describes your platform. Your C code is the application you
want to run on top of it and deployment.xml is the way you deploy your
application on top of this environment. I don't know what is a task... :) it
depends too heavily on your code, on what you want to do with it. The best
thing to do is to describe your tasks on the command line that is going to
be given to your process (the <arg> tag in deployment.xml) and to let them
create those tasks.

> 2. In file "surfxml.dtd", I think you provide the function to input trace
> files for dynamic cpu availability, network bandwidth, and so on. Could you
> tell me what's the format for such trace files?

Sure. Have a look in testsuite/surf/. Here is a simple example of cpu
availability trace:

PERIODICITY 1.0
0.0 1.0
11.0 0.5
20.0 0.9

It means that at time 0, the cpu availability is equal to 1.0. At time 11.0,
it changes to be equal to 0.5. At time 20.0, the new value is 0.9. At time
21.0, the new value is 1.0. At time 32.0 the new value is 0.5, and so on.
Get rid of the first line if you don't want a periodic trace.

I know I have to give more detailed examples about it. I even have opened a
bug for that:

   https://gforge.inria.fr/tracker/index.php?func=detail&aid=66&group_id=12&atid=165

> 3. In the earlier versions of SimGrid, the prediction functions are
> provided in SG. Are there such functions in MSG in latest Ver. 2.94?

No. As I already explained in previous emails, the models used in SimGrid
are somehow complex and do not allow to easily get bandwidth and latency
values. Those prediction functions could work in SG because we were dealing
with very simple models/DAGs and very simple simulations. It is not possible
anymore. Moreover, the goal of doing a simulation with SimGrid is to make it
as realistic as possible. In real life, you would never get accurate
predictions. So make your own measurements and guess some values... If you
really want predictions, you still can make your own measurements and use
the prediction library of NWS. But I really think it is a little bit
"overkill"... You should first try to use the last measured value and see
what it gives.

Best regards,

    Arnaud




More information about the Simgrid-user mailing list