[GE users] SGE, multi-site, and globus

aaron at cs.york.ac.uk aaron at cs.york.ac.uk
Wed Mar 16 10:25:49 GMT 2005


    [ The following text is in the "iso-8859-1" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Two other options are

* CSF (Community Scheduler Framework). This is due to be part of Globus
  Toolkit 4, although an implementation exists for Globus 3.2

  The default implementation for Globus 3.2 submits jobs only as the
  container user, not the submitting user, although patches exist
  (mostly from TACC at University of Texas at Austin) to modify this.

  You have to write your own additional scheduling plug ins for the top
  level.

  There seem to be issues with the Globus 3.2 based version, so it is
  probably best to see how the version that works with Globus 4 develops.
  When a truly stable version of Globus 4 and CSF will be available I
  don't know.

  Globus 3 is a minimum requirement for CSF.

* Condor-G

  Again this is based on Globus but AFAIK can work with a wider range
  of Globus versions than CSF. Support for a wider range of globus
  versions can simplify things if certains other tools you have require
  particular versions of globus. You can work around having
  multiple globus versions, but if you can keep it simple it leads to
  less problems. We are looking at this ourselves this week, and it seems
  to be somewhat easier to deal with than the Globus 3.2 based version
  of CSF.

  One advantage of Condor-G is that the GlideIn submits the job via a
  mechanism that virtually ensures that the user is submitting in a
  condor checkpointing environment. Whilst user-created status information
  based checkpointing is more ideal in terms of efficiency (amount of
  data required to be saved for status) having things submitted
  automatically against a checkpointing environment improves the chance
  of checkpointing being possible, and offers (ultimately) the possibility
  of making migration of jobs for load balancing easier (although there are
  still a lot of hoops to jump through to ensure that this works, not
  least the issue of differences in environments between different machines,
  which is quite a problem with Grid in general, and I feel that further
  virtualisation at a server level may be the way forward here.)

  Aaron Turner, Manager, White Rose Grid (York)


> Hello,
>
> I am looking to link to different sites (geographically) so as to be
> able to use both computing clusters.  Due to bandwidth between the
> locations, I have decided run each site as a stand-alone SGE install
> (60u3).  My question is how to connect the two sites to be able to
> submit jobs from one to the other (maybe even schedule jobs depending on
> wieghting factors)?  I read through JOSH, but noticed that it is based
> on the older SGE and Globus.
>
> Any recommendations?
>
> TIA.
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net
>
>



---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list