[Swift-devel] pilot job paper

[Daniel S. Katz]

On Jul 17, 2011, at 3:11 PM, Mihael Hategan wrote:

> On Sun, 2011-07-17 at 15:59 -0500, Ketan Maheshwari wrote:
>> In section IV.C I did not understand this about Swift Coasters: 
>> 
>> "" there is no possibility to manage this resource pool
>> programmatically ""
> 
> That statement is silly by construction. The resource pool is managed
> programmatically. It's not like coasters are implemented in a mechanical
> device with gears, levers, pulleys, and belts.

I think the point here was that the program that creates the pilot job cannot manage it - the spawner spawns and then goes away.

A couple of the authors are German, and occasionally write in a way that doesn't sound clear in English.

But I'm just guessing - the reviewers might comment on this too...

>> 
>> Which seem to be slightly contradicting with the last statement in the
>> same para:
>> 
>> ""SWIFT and Coaster supports various scheduling mechanisms, e. g. a
>> FIFO and a load-aware scheduler.""
> 
> Swift is separate from coasters. The fifo and load aware scheduler are
> from swift. The coasters have the block allocation scheme and its
> respective scheduler.

Right, which is why these two sentences are not contradictory, I think.

>> 
>> Also, Nimrod/G is mentioned as a pilot-based framework, but no details
>> are provided. As of 2008, when I used Nimrod/G, it did not have a
>> pilot-based execution framework.
>> 
>> A bit surprised that MOTEUR's interface with pilot-based system DIANE
>> and its many interesting recent results are not discussed provided
>> that Marc Santcroos is one of the developers.
> 
> They mention DIANE I thought.
> 
> 

-- 
Daniel S. Katz
University of Chicago
(773) 834-7186 (voice)
(773) 834-6818 (fax)
d.katz at ieee.org or dsk at ci.uchicago.edu
http://www.ci.uchicago.edu/~dsk/