Charm++ parallel programming system

[Yan Liu <yanliu AT cybergis.org>]

Hi Professor,

I have discussed seed balancer with Yanhua. I plan to extend one of the seed balancers, e.g., random or neighbor. If I can still use LDObjUserData in balancer, I can make better decisions on how to migrate chares.

But at first I need to pass extra bound and branching info (e.g., search type, load estimate, and master search node hash) to the balancer at the time a chare is created using ckNew(). I assume if I compile charm to enable CMK_LB_USER_DATA, I should be able to get the extra info in balancer, right? Or UserData feature is only available for centralized balancer?

Thanks,
Yan

Kale, Laxmikant V wrote on 12/06/2014 07:28 PM:

The load balancers for singleton chares (used in tree-structured computations, such as in branch-and-bound) are different in structure. I assume thats what you are looking for. They are called “seed balancers” in charm documentation. Yanhua Sun will probably be the best person to connect with on this issue.

On Dec 6, 2014, at 7:07 PM, Yan Liu <yanliu AT cybergis.org> wrote:

I'd like develop my own load balancer in this way:
1. when I do ckNew, I'd like to pass lb data to load balancer;
2. in LB implementation, I'll call stats->objData[i]->getUserData() to retrieve the data passed in 1.

I don't know how to do 1. Any suggestions? Or LBObjUserData is used in a very different situation?

Thanks,
Yan
_______________________________________________
charm mailing list
charm AT cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/charm

-----

Laxmikant (Sanjay) Kale         http://charm.cs.uiuc.edu

Professor, Computer Science     kale AT illinois.edu

201 N. Goodwin Avenue           Ph:  (217) 244-0094

Urbana, IL  61801-2302          FAX: (217) 265-6582