[GE users] Condor for SGE users

juby justin.ottley at gmail.com
Tue Jul 7 15:57:21 BST 2009

    [ The following text is in the "utf-8" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some characters may be displayed incorrectly. ]


Last year I completed a relatively extensive internal survey of Condor 
and SGE. At my company, there was a large, long-standing system based on 
Condor that I had been administering for a while, and I wanted to 
improve the custom layer that sat on top. After my survey and a couple 
presentations, we ended up switching completely from Condor to SGE. At 
the beginning I didnt expect this to happen, but boy am I glad we switched.

I could talk about this probably forever, but there a few key points I 
could share (I would need permission to release the entire survey). I 
disclaim that these are my opinions (albeit after researching and using 
both systems for a nontrivial amount of time), and should *not* be taken 
as truth.

I found that:
- Condor's configuration, especially after using SGE, is complicated 
(read: insane). Besides the obvious, there are other more subtle issues 
that make SGE much easier to deal with (one example is getting CLI 
results back over stdout).
- SGE is a "better integrated" system. There are design decisions in 
Condor that I think prevent its features from being well integrated, 
primarily the isolation of ClassAds. These decisions are beneficial in 
some (I think rare) situations, but in practice I find the design is 
more constraining that beneficial.
- SGE is, for the most part, a simpler system (fewer daemons to 
implement the same kind of features means easier troubleshooting, a 
single abstract resource management concept means the features work well 
together and are "predictable").
- SGE's fair share system is more flexible, easier to understand, and 
again, better integrated
- SGE's failover is better (read: actually works)
- SGE comes with useful tools like qmon and ARco (which I have to say, 
in spite of a few bugs, is a rather incredible tool).
- SGE's documentation is quite good (comparatively: searchability, 
easier to read, useful diagrams, more examples, whitepapers, etc).
- Condor does not have native job dependencies (DAGMan is Condor's 
answer, but I find in practice it is an overly complicated solution that 
ends up being less powerful than the simple SGE solution).

If you need a resource management system, I strongly recommend you 
consider SGE.


igardais wrote:
> Condor is mostly used for checkpointing.
> I think you can configure SGE to use Condor checkpointing.
> If they are not yet using SGE, setup a small cluster, validate the use 
> of Condor within and then make a demo of SGE's great power :)
> These are good things for an internship thesis.
> Ionel
> ------------------------------------------------------------------------
> *De :* magawake <magawake at gmail.com>
> *? :* users at gridengine.sunsource.net
> *Envoyé le :* Mardi, 7 Juillet 2009, 12h56mn 50s
> *Objet :* [GE users] Condor for SGE users
> For my summer/fall internship I will be  working for a major oil 
> company as a researcher. The systems teams insists we use Condor 
> because they have some experience with it, but I prefer SGE. Does 
> anyone have a feature breakdown between these two products? I really 
> want SGE :-)
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=205993 
> <http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=205993>
> To unsubscribe from this discussion, e-mail: 
> [users-unsubscribe at gridengine.sunsource.net 
> <mailto:users-unsubscribe at gridengine.sunsource.net>].


To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

More information about the gridengine-users mailing list