Re: [GridSim-users] Problems with communications

[Héctor B. de F. <hec...@di...>]

I designed the environment as a node of a multicluster corresponds to a
unique GridResource, which has only one PE. For example, If I have 100
nodes, I have 100 GridRresources, so I think it is not problem of the
allocation policy, as I'm assigning the gridlets to the nodes specifically
I want.

The coomunications works in this manner: When a GridResource receives a
communication request from the user, it pauses the execution of its
Gridlet. It will resume the gridlet when he have received the responses of
the GridResources which it is communicating. This is done at the
beginning, so the big part of the execution (processing) is done after the
communications part.

Resuming the story of the general problem, the processing execution time
is correct. The part that does not have a correct behavior is related to
the communications.

Hector.

> If I did get your question right I think you need to check how the
> resources deals with your gridlet,
> does the execution pause while the communication is going on or does the
> resource complete the
> execution and sends the gridlet output before sending the packets,
> resources use the FCFS so you
> need to be clear on that
>
>  Dibbydon
>
>
>
>
> ________________________________
> From: Héctor Blanco de Frutos <hec...@di...>
> To: gri...@li...
> Sent: Sun, November 1, 2009 10:23:23 PM
> Subject: [GridSim-users] Problems with communications
>
> Hello.
> Finally I have adapted GridSim to simulate a multicluster environment, and
> implemented some kind of communications between nodes. But, I have a
> problems with the behavior of it. The working model is at follows: The
> user sends a gridlet to every of the required gridresources (working as an
> unique node, meaning that one resource has only one PE), and immediatelly,
> the user sends a notification to the nodes, with the order to communicate
> with the res of the nodes. Then, every one sends a big packet (20-40 MB)
> to the others.
> While the nodes are transmitting, they pause the execution of its gridlet,
> to simmulate a blocking communication of the nodes.
> All seems to work well, until I read the execution time of the gridlets.
> If we have for example two clusters, one with a 100mbps channel, and the
> other with 100mbps, the gridlets last (for example) 1000 seconds to
> finnish. But If I change the channels to (for example, this combination):
> 100mbps, and 10mbps, the finnish time of the gridlets last the same. This
> not changes until I reduce even more the BW of one of the channels, to
> 100mbps and 1mbps. These ranges of BW have not much sense as, changing to
> 100mbps and 20mbps has no real impact on it. I revised the event traces,
> and all seems correct. Only the finnish time seems to do strange things.
>
> I use RIPRouter, Simplelink, and FIFOScheduler as the network elements.
> and the topology is the following: a main router, where the user and the
> clusters (with a router for every cluster) are connected to it. The
> gridresources are connected to the router of its corresponding cluster.
>
> Anyone has a idea of why the simulation has its strange behavior? Maybe
> I'm doing something incorrectly, but in one week revising the code and the
> network examples I can't figure which is the problem.
>
> Sorry for a too big message, but I wanted to give as much detail as
> possible.
>
>
> Best regards,
>
> Hector.
>
> --------------------------------------------
> Hector Blanco de Frutos
>
> Dept. d'Informàtica i Enginyeria Industrial
> Escola Politècnica Superior
> Universitat de Lleida
> --------------------------------------------
>
>
> ------------------------------------------------------------------------------
> Come build with us! The BlackBerry(R) Developer Conference in SF, CA
> is the only developer event you need to attend this year. Jumpstart your
> developing skills, take BlackBerry mobile applications to market and stay
> ahead of the curve. Join us from November 9 - 12, 2009. Register now!
> http://p.sf.net/sfu/devconference
> _______________________________________________
> Gridsim-users mailing list
> Gri...@li...
> https://lists.sourceforge.net/lists/listinfo/gridsim-users
>
>
>
>


--------------------------------------------
Hector Blanco de Frutos

Dept. d'Informàtica i Enginyeria Industrial
Escola Politècnica Superior
Universitat de Lleida
C/ Jaume II, núm. 69
25001 Lleida

http://www: http://gcd.udl.cat
http://www.neopontec.com

email: hec...@di...

Tel: +34 973 702 730
Fax: +34 973 702 702
--------------------------------------------