[GSA-RG] Minutes GSA-RG Telecon Nov 17, 2006

[Ramin Yahyapour]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1


Hi Thomas,

thanks for the comments:

>> - Negotiation Protocol
>>   o the protocol will support the gathering of time availability
>> information
>>     specific for a request.
> 
> There exist several approaches for this. For example, a mechanism
> developed within the Askalon system (paper at SC'06), work on placements
> of reservations (by Krzysztof Rzadca, me and other colleagues at ZIB;
> papers in CPE and @EuroPar-2006), and more (by Ramin?).

Yes, there are several approaches to this.
Thus, I assume it should not be too difficult for Oliver.
As far as I understood, he already had something for this.

>>   o two-phase commit, extended state-diagram for pre-reservation (to
>> support e.g. co-allocation)
>>   o problem: failure of brokers during comittment in pre-reservation
>>     currently not further investigated,
>>     for a later stage: "Paxos" seems like a suitable protocol to deal
>> with reliable committments (see work by Jon MacLaren)
> 
> I don't think 2PC or Paxos is sufficient. Paxos might help you if a
> broker crashes, but not if one of the selected resources is not
> responding. What you need is a decision within a certain period of time?
> If you're interested in our (TU Berlin & ZIB) findings, we could discuss
> this (on 27th Nov.).

Not sure about "decision within a certain period of time".
The problem is actually on the broker level.
Common example: You try to co-allocate two resources and have to issue the final
commit that you get them both or none. To this end, we have a reliable
pre-reservation in a two-phase commit. This usually works fine.
However, if in the 2PC the second server (or the communication to him) fails during the final
commitment you end up with an unwanted committed agreement for the first resource.
Even a three/four-phase commit etc will lessen but not resolve the problem.

Here, Paxos seems reasonable as it can cope with broker failure during the
commitment phase. Jon showed some results to this extent.


Ramin

> 
> Cheers,
> 
>  Thomas
> 
>> Attached are some minutes from today's telecon.
>> Can also be found here:
>> http://forge.gridforum.org/sf/go/doc14029?nav=1
>>
>> Philipp, I do not know whether you also took notes.
>> Feel free to update the minutes.
>>
>>
>> Ramin
>>
>>  
>> ------------------------------------------------------------------------
>>
>> -- 
>>   gsa-rg mailing list
>>   gsa-rg at ogf.org
>>   http://www.ogf.org/mailman/listinfo/gsa-rg
> 


- --
Dr.-Ing. Ramin Yahyapour         | mailto:Ramin.Yahyapour at uni-dortmund.de
University of Dortmund           | phone: (+49) 0231/755-2735
IRF-IT                           | mobil: (+49) 0179/5261973
44221 Dortmund / Germany         | fax:   (+49) 0231/755-3251
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.4 (MingW32)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQFFXZ4gVmlOotJWGRYRAvtUAKCz/1TPZggEU7ZbdpmmJ+uInqZ7ZwCfY4kd
1aHntt9IV1YV+bIiH5hJJ3c=
=PDXz
-----END PGP SIGNATURE-----