Clifford; Thanks for your input.

We are a smaller company than you realize (3-4 people); since we survive on contracts that last as short as 6 months and as long as one year; we wince at every new purchase of hardware. If we could get a long term contract that lasted 5 years, I would agree that getting a PC cluster of 10 CPUs or more would be a very good choice. We have 3 PC with a total of 4 CPUs. I have two networked and will try to get the third into the network. Maybe we can buy another one or two 'bare-bone' boxes with sufficient hardware to be number-crunchers only.

As to getting non-brand PC, I think this is only feasible for people who are exprerienced in PC hardware and software installation and bug-fixing since problems will occur and for those with little experience, its worthwhile to have the support of the brandname manufacturers. I have problems getting timely support from Polywell who uses third party tech support. I've found that timely support is probably more important than price and as important as software and hardware performance.

You say that you used Linux on your PC cluster. How much problems have you had with Linux in getting your peripherals functioning (modems, printers, tape drives, etc)? I have heard that installing hardware is easier with Windows 9x/NT than Linux. Windows 98 is much friendlier than Windows NT. If I could, I would avoid Windows NT but the grid gen software requires NT. Hopefully Windows 2000 will make NT as friendly as 98.

We will be forced to use WIND because we need its variable gamma capability in our next contract. WIND is better than NPARC in that it does not require grid block overlapping and allows grid sequencing where the grid can have various degrees of coarsening. Also WIND comes in a parallel processing version in Windows NT. WIND is supported by AECD and NASA/Glenn.

(1). I have no comment on the size of the company. (it is a very small one indeed.) (2). It is AEDC ( Air Force Arnold Engineering and Development Center), not AECD. (2). NASA/Glenn used to be called NASA/Lewis Research Center.

wow that is small. as for using linux it's not difficult to set up peripherals. perhaps you should get in contact with the people at redhat who sell one of the more widely used versions of linux for help. also you'll find that linux users tend to help each other out so you can find a lot of help online and in magazines. linux gives you the power of unix at a lower cost and allows you more performance in multiprocessor operations than windows. try the link i sent in the last message i think you'll find it easier than you anticipate to link the machines you currently have as well as others you might get in the future. it may also be useful for you to seek advie wrt computers in newsgroups dealing with the subject such as the comp.sci threads else if you have a buddy who works in that business you can buy him a few beers for some advice. it seems that you have to work hard to do your engineering work as well as the anciliary computer stuff that goes along with it. i guess i don't fully appreciate having a systems admin dept behind me here at school doing the dog work. i guess you have to find some backdoor approaches to doing that. you could possibly hire a college computer wizkid to come in a few hours a week to work for you since college kids come cheap.

Clifford,

I should have said that we are a tiny company

I have been given Redhat Linux as a gift and will install within a dual boot system on the Polywell which now has Windows NT, as soon as our present contract expires and I get our CFD solutions out.

I have used the Internet, including this forum, and the newsgroups and the bribing friends with beer, for help with my system

I guess that I could view this job as good training in PC support as in my previous jobs, I worked on SGI, HP, Cray, VAX, UNIVAC and IBM systems with their vast hordes of support.