[gsa-rg] Slide for the steps of Grid scheduler interaction

[Ignacio Martín Llorente]

Dear Ramin,

Ramin Yahyapour escribió:
> Dear Ignacio,
> 
> thanks for the input.
> I already have some questions about your model.
> 
> In your picture, GRAM, RFT, MDS are the key interfaces.

Yes

> 
> However, through which interface do you do the actual scheduling.
> That is, based on which information and which interaction
> do you make a decision when and where an action is performed.
> Is this solely done through the information from MDS?

There is no cooperation between meta-scheduling instances. The GridWay 
metascheduler always performs job submission though GT GRAM and takes 
decisions based on the attributes provided by MDS.  In this way, a 
meta-scheduler interfaced by GRAM can be considered as a local resource 
manager (like PBS, SGE...) and the MDS attributes provided are an 
aggregation or a summary of the underlying grid infrastructure, managed 
by the other metascheduler.

The key idea is to virtualize a whole grid infrastructure by 
encapsulating the metascheduler in a GRAM service. This also provides a 
natural way to achieve interoperability between different Grid 
middlewares (EGEE, GT, Unicore...).
> 
> As the local RMs usually have their own schedules, I am interested on
> how you make the determination when a resource will be available in
> the future. Whether you use any kind of reservations?
> And if yes, how do you interact with a remote scheduler about the
> "timing" of this reservation. This usually considered essential for
> workflows and co-allocation of resources, like data, storage, bandwidth etc.

We currently do not provide support for advance reservation. Such 
functionality will be provided as soon as available by GRAM. We hope in 
few months.

Regards,

Ignacio

> 
> Ramin
> 
> Ignacio Martín Llorente wrote the following on 19.07.2006 10:49:
>> Dear friends,
>>
>> The slide seems very interesting. We have been already working in the
>> interconnection between schedulers. I am attaching one slide with our
>> view. The key point is to communicate meta-schedulers with the same
>> protocols and interfaces used to access local resource management systems.
>>
>> More info at http://www.grid4utility.org (a subproject of the GridWay
>> metascheduler).
>>
>> Regards,
>>
>> Ignacio
>>
>> Ramin Yahyapour escribió:
>> As discussed during the last telecon,
>> I created a slide with the mentioned steps for the interaction
>> between schedulers. I reused the initial slide by Ariel.
>>
>> You can find it at:
>> https://forge.gridforum.org/sf/go/doc13677?nav=1
>>
>> Does this cover correctly our previous discussion?
>> Would it fit with the current approaches as presented
>> by Mathias and Ariel?
>>
>>
>> Next,
>> I will have to work on the requirement forms.
>> Probably, some of you also have not yet done their part.
>> I hope you can find time for it.
>>
>> I am quite busy next week. Thus, I will not be able to
>> conduct a call. I would suggest the week afterwards; e.g.
>> August 2nd or 3rd.
>> It woul be nice if everyone could provide its section for the
>> requirement document, so that we could conclude this activity
>> in the next weeks.
>>
>> Ramin
>>
>>>
> 


-- 
Ignacio Martin Llorente
http://asds.dacya.ucm.es/nacho
----------------------------------------------------------

GridWay, Meta-scheduling Technologies for the Grid! http://www.gridway.org

----------------------------------------------------------
You are invited to visit our research web page http://asds.dacya.ucm.es
----------------------------------------------------------