[Swift-devel] Re: Question for running swift with 2 psets on BGP

[Ioan Raicu]

Zhao,
There should be a way to set a maximum number of tasks to have 
outstanding at any point in time.  You should set this to 256 per site, 
as we have 256 CPUs on each P-SET.  This is a setting from Swift that 
needs to be set.  Perhaps, if this number was set higher (say infinity), 
Swift might have submitted all 512 tasks to 1 P-SET before the Swift 
scheduler was able to submit any to the second  one.  What happens if 
you make the task lengths 60 seconds, and you send 5120 tasks?  Do they 
all queue up on 1 service?  Or do they eventually start load balancing 
across the two services?  Ideally, on the BG/P, with multiple sites 
(services), you don't want to queue anything up, you just want to send 
enough tasks to keep all CPUs busy, but no tasks in the queues.

Mihael, Ben, where would Zhao set this parameter that will allow us to 
limit the number of outstanding tasks per site?

Ioan

Zhao Zhang wrote:
> Hi, Mihael
>
> I run a sleep_10x512 workload with swift + falkon on BGP. Swift send 
> all tasks to the first pset,  but never to the second.
> The plotted log are at 
> http://www.ci.uchicago.edu/~zzhang/report-sleep-20080722-1456-3fnc42b1/
> The sites.xml is at http://www.ci.uchicago.edu/~zzhang/sites.xml
> And the tc.data is at http://www.ci.uchicago.edu/~zzhang/tc.data
>
> Thanks for help
>
> zhao
>

-- 
===================================================
Ioan Raicu
Ph.D. Candidate
===================================================
Distributed Systems Laboratory
Computer Science Department
University of Chicago
1100 E. 58th Street, Ryerson Hall
Chicago, IL 60637
===================================================
Email: iraicu at cs.uchicago.edu
Web:   http://www.cs.uchicago.edu/~iraicu
http://dev.globus.org/wiki/Incubator/Falkon
http://dsl-wiki.cs.uchicago.edu/index.php/Main_Page
===================================================
===================================================