[Beowulf] Re: Parallel Programming with MPI, anyone read the book?

[Peter Pacheco]

Hi Kyle,

I seem to recall reading a book of that title.  If my memory serves,  
it was an excellent book. ;)  Alas, pretty much anything written  
about machines of 10 years ago is going to be very out of date:    
when I wrote the book, the teraflops barrier had yet to be broken.    
On the other hand I think the material on MPI is still useful.

Best wishes,
Peter Pacheco

> Date: Thu, 24 May 2007 14:06:26 -0400
> From: "Kyle Spaans" <3lucid at gmail.com>
> Subject: [Beowulf] Parallel Programming with MPI, anyone read the
> 	book?
> To: "Beowulf List" <beowulf at beowulf.org>
>
> I've found this Mogran Kaufmann Publishers, Inc. book by Peter S.
> Pacheco in my University's Computational Mathematics Club. It's (c)
> 1997, but I figure it should still be mostly relevant.
> Has anyone else read this, and/or have some comments on it's
> validity/applicability?
>
> So far, the only thing I've been disappointed with is this sole
> reference to Beowulfery that I could find in Chapter 2, under the
> headings Distributed Memory MIMD, and "Bus-based Networks" where they
> say:
>
> "The last, and probably the simplest, network is a bus. A cluster of
> workstations on an ethernet provides a popular example. Of course,
> busses tend to be fairly slow, and even worse, busses, especially
> ethernets, soon become saturated if there are more than a few nodes or
> more than absolutely minimal communication. Thus, although they are
> very useful for program development, currently available bus-based
> system don't show much promise for very large-scale applications."
>
>
> This paragraph sounds a little dated to me. Isn't Ethernet more
> prevalent than that?