charm AT lists.siebelschool.illinois.edu
Subject: Charm++ parallel programming system
List archive
- From: Eduardo Rocha Rodrigues <errodrigues AT inf.ufrgs.br>
- To: Gengbin Zheng <zhenggb AT gmail.com>
- Cc: "charm AT cs.uiuc.edu" <charm AT cs.uiuc.edu>
- Subject: Re: [charm] [ppl] Distributed LB
- Date: Tue, 12 Jul 2011 17:22:27 -0300
- List-archive: <http://lists.cs.uiuc.edu/pipermail/charm>
- List-id: CHARM parallel programming system <charm.cs.uiuc.edu>
Gengbin,
As for the AtSync mode, do I have to wait for migrations to complete
in order to invoke resumeClients?
I was looking at the NeighborLB (actually, the parent class,
NborBaseLB) and it seems to use the AtSync mode, right? This LB seems
to wait for all migrations to complete before resuming execution
(MigrationDone).
I'd like to know also what is the meaning of the method
NotifyMigrated? Does it register a function to be called when a
migration arrives?
Eduardo
On Mon, Jul 11, 2011 at 5:49 PM, Gengbin Zheng
<zhenggb AT gmail.com>
wrote:
> Eduardo,
>
> Have you looked NeighborLB for an example of distributed LB?
> There are two types of load balancing scheme, AtSync and non-AtSync.
>
> If an object is in AtSync mode (which is default for chare array
> elements), it will wait for resumeClients call to resume.
> Otherwise, use non-atsync mode, where you assert an object is capable
> of any-time migration, and you don't need to resume it explicitly.
>
> Gengbin
>
> On Mon, Jul 11, 2011 at 2:26 PM, Eduardo Rocha Rodrigues
> <errodrigues AT inf.ufrgs.br>
> wrote:
>> Hi,
>>
>> I think what I was looking for doesn't make much sense in a
>> Distributed LB. I mean, I don't need to wait for the migrations to
>> complete, right? I can just call theLbdb->Migrate and, immediately
>> after that, call theLbdb->ResumeClients. A migrating object may resume
>> execution in the same processor, but eventually it will get to its
>> destination. Is that correct?
>>
>> Eduardo
>>
>> On Mon, Jul 11, 2011 at 11:32 AM, Eduardo Rocha Rodrigues
>> <errodrigues AT inf.ufrgs.br>
>> wrote:
>>> Professor Kale,
>>>
>>> I am calling theLbdb->Migrate (theLbdb is a property of BaseLB) to
>>> migrate objects. Does this call generate an event when it is done? I
>>> mean, when the object is on its destination. If so, how can I wait for
>>> such an event?
>>>
>>> Regards,
>>> Eduardo
>>>
>>> On Mon, Jul 11, 2011 at 1:11 AM, Kale, Laxmikant V
>>> <kale AT illinois.edu>
>>> wrote:
>>>> I haven't looked at CentralLB:MigrationsDone, but I think a simple use of
>>>> a future (or just a flag) can make it so that the user just has to make a
>>>> blocking call to wait for the event. Am I missing something?
>>>>
>>>> I will be happy to look at the code,and discuss, ma be with Celso and you
>>>> on skype with the code in front of us.
>>>>
>>>>
>>>> --
>>>> Laxmikant (Sanjay) Kale http://charm.cs.uiuc.edu
>>>> <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
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> On 7/10/11 1:06 PM, "Eduardo Rocha Rodrigues"
>>>> <errodrigues AT inf.ufrgs.br>
>>>> wrote:
>>>>
>>>>>Hi,
>>>>>
>>>>>In order to make sure that all migrations have finished, I wrote a
>>>>>callback routine, similarly to what was done in CentralLB
>>>>>(CentralLB::MigrationDone).
>>>>>
>>>>>It seems to work, but I think this is too complex. I want something
>>>>>simpler for a tutorial.
>>>>>
>>>>>Can anyone help me?
>>>>>
>>>>>Thanks in advance,
>>>>>Eduardo
>>>>>
>>>>>On Thu, Jul 7, 2011 at 2:31 PM, Eduardo Rocha Rodrigues
>>>>><errodrigues AT inf.ufrgs.br>
>>>>> wrote:
>>>>>> Hi Abhinav,
>>>>>>
>>>>>> I was looking for something simpler for a tutorial I am going to
>>>>>> present. I got as an example the NullLB and change it so that I
>>>>>> perform some load balancing decisions, instead of doing nothing.
>>>>>>
>>>>>> I managed to do some migrations, but sometimes I resume execution
>>>>>> before some objects migrate. I mean, I call the method
>>>>>> LBDatabase::ResumeClients before some objects arrive to their
>>>>>> destination. Is there any way to guarantee that all the objects that I
>>>>>> ordered to migrate (with LBDatabase::Migrate) had finished migrating?
>>>>>>
>>>>>> Thanks,
>>>>>> Eduardo
>>>>>>
>>>>>> On Sun, Jul 3, 2011 at 4:49 AM, Abhinav S Bhatele
>>>>>> <bhatele AT illinoisalumni.org>
>>>>>> wrote:
>>>>>>> Hi Eduardo,
>>>>>>> Are you referring to writing a hierarchical load balancer? Is yes,
>>>>>>> HybridLB.C in src/ck-ldb is an example.
>>>>>>> - Abhinav
>>>>>>>
>>>>>>> On Fri, Jul 1, 2011 at 11:28 AM, Eduardo Rocha Rodrigues
>>>>>>> <errodrigues AT inf.ufrgs.br>
>>>>>>> wrote:
>>>>>>>>
>>>>>>>> Is there any tutorial on how to write a distributed load balancer on
>>>>>>>> Charm? I mean, something similar to what can be found on:
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>https://charm.cs.uiuc.edu/manuals/html/charm++/3_13.html#SECTION0003133
>>>>>>>>00000000000000
>>>>>>>>
>>>>>>>> Or an example would be helpful.
>>>>>>>>
>>>>>>>> Thanks in advance,
>>>>>>>> Eduardo
>>>>>>>> _______________________________________________
>>>>>>>> charm mailing list
>>>>>>>> charm AT cs.uiuc.edu
>>>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/charm
>>>>>>>> _______________________________________________
>>>>>>>> ppl mailing list
>>>>>>>> ppl AT cs.uiuc.edu
>>>>>>>> http://lists.cs.uiuc.edu/mailman/listinfo/ppl
>>>>>>>
>>>>>>>
>>>>>>
>>>>>_______________________________________________
>>>>>charm mailing list
>>>>>charm AT cs.uiuc.edu
>>>>>http://lists.cs.uiuc.edu/mailman/listinfo/charm
>>>>
>>>>
>>>> _______________________________________________
>>>> charm mailing list
>>>> charm AT cs.uiuc.edu
>>>> http://lists.cs.uiuc.edu/mailman/listinfo/charm
>>>>
>>>
>>
>> _______________________________________________
>> ppl mailing list
>> ppl AT cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/ppl
>>
>
> _______________________________________________
> charm mailing list
> charm AT cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/charm
>
- [charm] Distributed LB, Eduardo Rocha Rodrigues, 07/01/2011
- Re: [charm] [ppl] Distributed LB, Abhinav S Bhatele, 07/03/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/07/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/10/2011
- Re: [charm] [ppl] Distributed LB, Kale, Laxmikant V, 07/10/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/11/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/11/2011
- Re: [charm] [ppl] Distributed LB, Gengbin Zheng, 07/11/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/12/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/11/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/11/2011
- Re: [charm] [ppl] Distributed LB, Kale, Laxmikant V, 07/10/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/10/2011
- Re: [charm] [ppl] Distributed LB, Eduardo Rocha Rodrigues, 07/07/2011
- Re: [charm] [ppl] Distributed LB, Abhinav S Bhatele, 07/03/2011
Archive powered by MHonArc 2.6.16.