AMD64 chipset Linux support .... Elementary costs in CFD and other, distributed apps ...

[Bryan J. Smith]

On Wed, 2006-07-12 at 01:58 -0400, Bryan J. Smith wrote:
> Consider your time swapping out a board/chipset that was specifically
> designated as "does not test to 24x7 tolerances" (e.g., i865 v. i875) in
> that cost estimate.

Or the cost of your engineer's time and/or deadlines for a failed run.

> **NOTE:  I (among others on this list) rolled out such grid computing
> for CFD and other, distributed applications with Linux in the late '90s.
> Even back then, a single 32-bit PCI bus didn't cut it.  The options are
> a little better now with PCIe x1 channels being more standard -- but the
> data rate has always increased as well.

If cost is really paramount, then get the cheapest _refurb_ mainboard
and Celeron or Sempron you can find, run diskless (if possible -- it
does make management easier) keep the cost under $200/unit.  The only
issue is if and when the I/O becomes a bottleneck for the DTR --
especially when you have masters in the distributed system (which is why
you should _never_ skimp there).

You also need to ensure your applications -- especially if engineering
runs take days at a time -- can handle failover and restart an
interation/process if a node fails.  That right there is typically the
biggest issue and cost (far, far more than hardware and IT time) -- code
that is non-interruptable/restartable when runs take days.



-- 
Bryan J. Smith          Professional, technical annoyance
mailto:b.j.smith at ieee.org    http://thebs413.blogspot.com
---------------------------------------------------------
The world is in need of solutions.  Unfortunately, people
seem to be more interested in blindly aligning themselves
with one of only two viewponts -- an "us v. them" debate
that has nothing to do with finding an actual solution.