fostel@cscadm.ncsu.edu (Gary Fostel) (03/15/89)
We are about to purchase a low end parallel system, almost certainly
a hypercube or similar system. There seem to be only 2 or 3 vendors
offering such systems: NCube, Intel and possibly Meiko. We need to
hear feedback from people who have actually used these systems; if
you have opinions but no experience, please respect our need for
direct experience.
The scale of the system we are going to buy is whatever can be bought
on University discounts for not much more than $100k. A bit more money
is available for good graphics attachments to the basic computing
element.
If you have used, or are using an NCube/7 or Intel iPSC2 with small
nodes (memory-wise) or a Meiko Transputer based system, please
send me some mail summarizing your opinions of the system. Of
particular interst is the ability to do cost effective hi-res
graphics, support more than one concurrent user, and be relatively
bullet-proof (students will be the primary users!). Sadly, the
funding source will not support maintanence so any observations
about the reliability of the hardware and the vendor are relevant.
As everyone always says at this point, thanks in advance, and if I
get a significant number of responses, I will post a summary of the
responses. Please respond by sending mail to "fostel@cscadm.ncsu.edu".
----GaryFostel----eugene@eos.arc.nasa.gov (Eugene Miya) (03/16/89)
Gary points out some low cost systems. I would like to point out there are many more companies than this producing various types of "toy" Hypercubes, most not on the net. Such companies include CSA (Computer Systems Architects) in Provo, Utah, Levco in San Diego, etc. Many of these people use Transputer and other chip sets. These things go into the back of PCs and Macs. Their software tends to be in a much worse state, but the real question is whether information learned on these machines can appropriately scale to the more respectable Cube makers (frequently not, but...). It's a matter of pride on the parts of some Departments (sort of like if your school teaches BASIC or COBOL, or if you use Apple IIs and IBM PCs in your CS departments). I am not endorsing the purchase of this hardware only noting that lower cost things exist. Yes, its a bit of a lark if you can only afford an 8 or 16 node cube. You really want much larger scale systems, and most of these small systems don't have H/W support for virtual processors. But you will learn a bit about scaling. Believe it or not, there are a few rich folk who go and buy these things for home use. In the late 1970s, it was the mainframe and mini people who snubbed the micro computer people. I only need look out my window to see where we are now. I hope the same thing doesn't happen in parallelism. ;-) When I was growing up the thing to do was build a laser (JHS and HS). (It was actually Marvin Minsky in a written letter back then convince me not to.) A couple of years ago, a newly found friend and her husband wanted me to talk to their HS age son about a project he wanted to do: build a parallel processor. Gave me to whole design. (2 MC68Ks, shared memory, the works). My only comment after listening was, "Have you mother tell you about operating systems." In the end he didn't do it, but some day, some kid (probably happening right now) is thinking about building their own Cube computer for a HS age project. Another gross generalization from --eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov resident cynic at the Rock of Ages Home for Retired Hackers: "Mailers?! HA!", "If my mail does not reach you, please accept my apology." Domains, the zip codes of networks.
dave@VAX1.CC.UAKRON.EDU (David Stoutamire) (03/18/89)
In article <4784@hubcap.UUCP> fostel@cscadm.ncsu.edu (Gary Fostel) writes: > > We are about to purchase a low end parallel system, almost certainly > a hypercube or similar system. There seem to be only 2 or 3 vendors > offering such systems: NCube, Intel and possibly Meiko. We need to > hear feedback from people who have actually used these systems; if > you have opinions but no experience, please respect our need for > direct experience. > > The scale of the system we are going to buy is whatever can be bought > on University discounts for not much more than $100k. > This about describes our situation at the University Of Akron. We have funding for a parallel system in the $100k range and need to obtain information about available systems. Shared memory type machines need not apply, as we will soon be obtaining one of that sort and wish to diversify. Of particular interest are hypercube architectures. This post may be interpreted by vendors as a request for informational literature. As in the above post, we also welcome any comments pro/con from people who have actually used these systems. Respond on the net, or literature may be sent to: Dr. Hui Tan Akron University Dept. of Math Akron, OH 44325