disadvantages of linux cluster

[Donald Becker]

On Tue, 5 Nov 2002, Stephane Glockner wrote:

> I work in a french university laboratory. We do massive Computational
> Fluid Dynamics at present time on ORIGIN2000, IBM Regatta... The solution of 
> a Beowulf cluster seems very interesting as regards as financial cost and 
> performances. 
> We would like to use it in the following manner :
> - multi users (around 10 or 15 users)
> - multi single-processor, light and heavy applications (<=2Go of RAM)
> - use of a parallel version of our application
> 
> I didn't see anywhere on the web disadvantages of such a solution. What are 
> they ?

The general issues are
  Your parallel applications must use a message passing model, typically
  MPI or PVM, rather than shared memory.  While DSM/NVM implementations
  are available to emulate shared memory, they require more application
  tuning work than converting to message passing.

> Is it simple to administrate day after day ? Is it stable ?

I feel these are the critical issues.

The long-term administration effort vary widely between cluster
distributions.  We have focused on a distribution for commercial
application deployment, thus Scyld has a significant advantage over other
distributions with
  - a consistent architectural model
  - minimal administrative training and effort
  - very easy scalability
  - defined and tested release series
  - binary compatibility between versions and among network layers

Most other cluster distribution are focused on developers and discard
an architectural model, testing and binary compatibility in favor of
pulling in the latest version of each subsystem.  That's great for
computer science departments, which really want to see and learn about
(and sometime debug) the latest changes.

> Does it support heavy load of users and applications as defined above?
> Isn't it too much an experimental solution ?

You can definitely "go experimental" here -- most cluster distributions
are in this category.  Most whole-system testing work requires
repetitive work, attention to detail and doesn't result in the
appreciation of the community, thus you usually find it only in
commercial distributions.

[[ BTW, keep in mind the adjective "commercial" is orthogonal to Open
Source / Free Software.  ]]

-- 
Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993