[GE users] campus grid vs global grid

Chris Dagdigian dag at sonsorol.org
Thu Jul 17 14:51:52 BST 2008


    [ 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. ]


I'll give you a cynical perspective from someone in industry who cares  
more about usefulness and productivity than hype and press releases ...

But first a question back to you - have you considered the possibility  
of keeping the cluster in a single site and bringing the users and  
groups in over the internet, VPNs or dedicated circuits?

I think when you do a true cost analysis for "global grid" you'll find  
significant downsides including the cost of bandwidth, the high  
overhead needed to keep the system running, the high overhead needed  
to deal with multi-group political issues, the significant security  
management hassles you will find that the cost of centralizing compute  
resources is far easier from a cost, filesystem, security and  
management perspective than trying to distribute dirt-cheap commodity  
server hardware at multiple sites.

If you can't point to an overwhelming and well-justified need for  
multi-site grid computing (examples include: business continuity  
reasons, 24x7 workforce reasons, big science problems, inept  
management who want to put out press releases etc. etc.). The people  
who do this kind of thing just because they think it would be neat or  
interesting are almost certainty going to fail utterly.

With the exception of a single canadian bioscience company I've only  
ever seen two types of organizations actually be successful with multi- 
site capital "G" Grid computing:

  - Organizations funded by national governments or sovereign funds  
(tens of millions of dollars per year or more ...)
  - Academic institutions that benefit from cheap labor (grad  
students) and the PR/reputation benefits of Grid computing

I'm getting so bored and tired of going to the same conferences and  
seeing the same group of rotating talking heads talk about how amazing  
their Grids are. They (of course) don't mention that their projects  
are successful because they have IT budget and staffing resources that  
simply don't exist in a non-academic or non-government-grant-funded  
world.

This goes back to my main point -- a true analysis of costs and  
resources often points to the fact that centralizing compute and  
bringing in the users from remote locations is often the most  
straightforward approach. The price of compute power is just so cheap  
these days compared to the cost of fast storage and networks that you  
really need a massive justification before it makes sense to  
distribute the CPUs across multiple sites.


</cynic> mode off...

-Chris




On Jul 17, 2008, at 9:27 AM, Frédéric Bastien wrote:

> Hi,
>
> I'm working in a lab at the University of Montréal. We are exploring
> the possibility to group the cluster with other labs in other
> university. To do so I taught using SGE, but I have a few question. In
> the user doc, it tell that SGE is made for campus grid but not for
> global grid. I will need the technical difference that this represent.
>
> - Do we need a global file system between labs?
> - Do we need a unique user id? user name?
> - No firewall?
> - Other limits/requirement?
>
> Here is the URL that talk about campus vs global grid:
>
> http://docs.sun.com/app/docs/doc/820-0699/chp1-1?a=view
>
> Also, I have another question, at the end of 2007 their was a PDF that
> suggest that kerberos be integrated in SGE to allow the transfert of
> ticket. Do this have been implemented in SGE 6.2?
>
> thanks for your time
>
> Frédéric Bastien
>

---------------------------------------------------------------------
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