[Beowulf] Customize Comparison between Torque(PBS)/SGE and LSF

Rayson Ho raysonlogin at gmail.com
Fri Aug 5 23:25:54 EDT 2005


On 8/4/05, Sebastian <ChiefShadow at gmx.de> wrote:
> SGE:
> + GUI
> + Support (Mailing List)
> + all demands are meet
> + LAM-Support with Reutis Howto
> + Basics for Checkpointing/Migration are available
> + OpenSource

Gridengine supports checkpoint/migrate, but you may need to configure
a ckpt interface. It may involve some scripting to do to invoke
external commands to do checkpointing (if any provided by LS-DYNA?).
http://gridengine.sunsource.net/nonav/source/browse/~checkout~/gridengine/doc/htmlman/htmlman1/sge_ckpt.html


You can get commerical support from:
http://www.sun.com/software/gridware/partners/index.xml


Webinterfaces are also available... in fact there are 4 or 5 written:
- the simplest one is the GE momitor: http://www.veus.hr/php/gemonitor/
- or this one with XML->html: http://scripps.bos.bioteam.net/xml-qstat/
- or http://gridengine.sunsource.net/gep/GEP_Intro.html
- and there's one from the Rocks cluster guys


> Torque:
> + Integration Torque and LAM/MPI with the tm-Modul
> 
> - requirements only available with MAUI Scheduler
> - no GUI when using MAUI (is that right??)
> - xpbs&xpbsmon isn't good looking

Maui is supposed to be the most advanced scheduler, with backfilling
and advanced reservation...

But with the size of the cluster, I am not sure if it's needed or
not... and in fact, Maui+Torque is slower than SGE in terms of
throughtput:

http://arxiv.org/abs/cs.DC/0506006

>From page 7 of the paper:

                                SGE        TORQUE TORQUE+MAUI OAR   OAR(2)
Available Processors    34              34              34            
 34        34
Jobmix work (CPU-sec) 443340    443340  443340          443340 443340
Elapsed Time               14164      14818    15115            15264    14037
Efficiency                     0,9206     0,8800    0,8627         
0,8543  0.9289

This is kind of expected, since [SGE and Torque] "all the jobs
requiring few processors are scheduled first while all the big
parallel jobs are delayed until the end of the schedule."

Rayson

> 
> This is actually my small list with advantages/disadvantes, are there some
> wrong or are there some missing?
> 
> I read of some problems with PBS when you submit more than hundret jobs the
> same time, because i think we don't have more than 20 jobs in a queue this
> shouldn't be problematical.
> 
> Now i listen what you will tell me and i'm amused of many answers.
> 
> Thx and Kind Regards,
> 
> Sebastian
> 
> --
> GMX DSL = Maximale Leistung zum minimalen Preis!
> 2000 MB nur 2,99, Flatrate ab 4,99 Euro/Monat: http://www.gmx.net/de/go/dsl
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
>

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list