Re: [Boston.pm] OT Grid Computing

[Charles Reitzel]

If I have it right, Grid Computing is the descendant of academic projects 
such as Condor.  They typically support mixed computer architectures (e.g. 
you can mix different OS and hardware designs in the same grid) - sometimes 
over wide areas (i.e. the occasional slow link).  The defining 
characteristic, is job distribution based on a message passing 
infrastructure.  This usually implies application re-tooling to 
support/exploit the environment.  Simpler schemes to distribute "batch" 
jobs over a LAN (e.g. distributed Make) are not considered Grid Computing, 
per se.  For an example of why not (and to relate this thread to Perl), see 
this: http://www.beowulf.org/pipermail/beowulf/2003-April/006491.html

There is also a Parallel::MPI module (which is an XS binding of the C MPC 
API), although there is little descriptive info.

Side issues: Due to its high performance and low cost, many feel that 
Gigabit Ethernet will win out over more self-important switched fabrics as 
the hardware interconnect.  There is also a huge overlap with the Storage 
Area Networks (SAN) in the high speed system interconnect space.  Again, 
with Gigabit Ethernet showing well for the same reasons.

Looked at from a historical perspective, the traditional thinking about 
network vs. disk is changing. The differences in latency and throughput 
between local disk and network data are approaching 0.  With disk 
striping  becoming the norm, even disk spindles are less and less of a 
factor in I/O calculations.

take it easy,
Charlie
At 03:02 PM 10/13/2003 -0700, Ranga Nathan wrote:
Press is beating up on Grid Computing, no doubt because IBM, Oracle and
others see $$$$ signs.
What do people think?
To me it seems like safe threaded applications that can run on any of the
networked machines. Is it too simplistic?
_______________________________________________
Boston-pm mailing list
[EMAIL PROTECTED]
http://mail.pm.org/mailman/listinfo/boston-pm