> I've recently had a number of IBM web pages pointed out to me
> that talk
> about zSeries boxes being useful in grid computing.

390's are useful in grids, but primarily as storage management nodes and
as control nodes. As Mark points out, 390s don't produce enough raw
cycles to be very interesting as computational engines, however, they do
have interesting driver applications that could initiate and employ
high-performance grid farms to great benefit (think about things like
risk analysis that use large data sets, and need raw cycles to chew
things up).


> David Boyes has been active with Globus on Linux/390, and I can easily
> imagine workload that would be a good fit for that.  What I'm not sure
> about, and perhaps David can comment on this, is if grid
> computing workload
> and grid nodes can be characterized properly to make the appropriate
> workload get sent to Linux/390 nodes.

With the current gatekeeper/broker and "clumping" of machines and
resources (for lack of a better word), workloads can be directed to what
is effectively a group of resources with particular attributes.
Applications can specify the attributes necessary for execution of a
particular task, and part of the job of the broker process is to locate
available resources that possess the necessary attributes and direct the
task to the appropriate place.

If configured correctly, the Linux Globus code registers itself as
biased toward I/O intensive applications and indicates that it is not a
good choice for computationally intensive tasks. The real interest is
allowing a 390 application to initiate and control processes running in
the Intel grid.

Reply via email to