From: Paul
Date: Thu, 26 Oct 2000 22:53:08 +0100
Subject: [TSSP] Linux, clustering and distributed computing
Richard, Terry I'm glad to hear you're both considering installing linux, I recall Bob Golding mentioned a while ago that he was thinking of doing the same. It does represent a bold move. RMC wrote: > this Beowulf clustering capability: does it lend itself to > packaging work for other machines to do? The idea behind the clustering is that a (local) network of machines appear to the program as a single machine with lots of processors. Generally a high network bandwidth between the 'nodes' is required so that the individual machines can virtualise themselves effectively, so it's not a useful technique for distributed wide-area computations. Using MPI for tsim and tlap doesn't give any computational advantage but it does make it easier to spread the calculations across a lot of machines, so for me its just an administrative thing. I already have the cluster setup for doing stuff in number theory and astronomy, so I use it simply because its already there. Frankly I would forget clustering until you're comfortable with the OS itself - the learning curve will be plenty steep enough. > i have a pair of dual -CPU machines that I could process on for > you if it becomes necessary to use finer quantisation in your > calculations. I don't know about finer quantisation - the precision at the moment seems enough for our requirements. What I can forsee is a need to examine a lot of (not necessarily real) tesla systems, as a step towards tackling the problem of finding an 'optimum' configuration of primary and secondary for a given application. Given an available supply power at a given voltage, what is the best combination of primary and secondary turns, gauge, length, radius, height, toroid, etc? That's quite a big 'configuration space' to explore, but given enough cpu effort, the results could be tabulated and used as a database from which semi-empirical formulae could be extracted. It may take a day or so to model a system, but thats a lot quicker than building and measuring it! > I'm running Win2k but i don't mind going over to a Posix OS on > a spare drive or two. > A dual PII-450 and a dual PII-300 is what I've got. More than adequate. You could perhaps commit one of the machines to Linux and see how you get on with it. Treat yourself to a big new hard disk and RH7. Fit the new drive as the master IDE drive and install RH7. Move the existing drive to the slave IDE slot and tell linux its a windows filesystem and it will mount it accordingly so that you can get at all your existing files. Tell linux to be a fileserver and you can then hook your remaining PCs to it with a bit of ethernet. And if you don't like the results, just move your slave drive back to the master position and reboot. Terry wrote: > I am getting a 733MHz PIII computer for web server use on Friday. > and I have this Redhat Linux 6.0 software box... > If I make the new machine dual boot (Win2K / Linux), will I be on > may way to being able to run your program?? RH6 contains everything you need to edit, compile, and run this software, and virtually all other unix software too. It contains compilers for all the mainstream languages (C, fortran, etc), document processing and typesetting (Tex, etc), as well as all the usual file & web serving, email routing, etc. > Forgive if this question is stupid, but I know very little about > UNIX... I know it has some version of Xwindows which I think > your program needs. Neither tlap nor tsim need X, but if X is available, tlap puts up a window to show the E-field as it evolves. The programs are driven entirely from the command line so that they can be scripted into a search pattern (when looking for resonances) and the output file format is geared towards postprocessing with standard unix tools. > Let me know if I am generally on the right track and if I need some > (hopefully free) C compiler. The gnu C compiler gcc which comes with linux is pretty much the industry standard that other compilers try to copy. Most unix systems used for software development ditch the native compiler and use gcc instead - it'll cross compile for any machine. It'll be the last C compiler you'll ever need to not buy. The only problem with the dual boot idea is that you probably won't want to take your win2k web server down every time you want to play real computing. Unless you have very high net traffic I'd recommend using the oldest, slowest machine you have as a webserving, masquerading, firewall PC running RH. Setup your nice new machine as a workstation and main filesystem. Export your web files from the new machine by NFS, mounting it readonly onto the firewall machine. I would recommend getting RH7 instead though, RH6 has been around a while now, and the RH7 package contains a lot of upgrades and some extra new stuff, eg it contains all the main Beowulf clustering systems. Chances are if you're buying a new computer you can get it pre-installed, but its more informative setting it up yourself. While we're talking about computing, and in view of the interest on this list in examining the secondary when driven through a primary, it could now be an appropriate time to consider adding in the necessary code to the tsim program to enable it to model the primary and its coupling. I've held off doing this until the secondary model has settled down and had the benefit of some experimental verification. I'm fairly satisfied with this now and the program seems to be able to represent the secondary with reasonable accuracy, and least for coils not too far off the ground. All the necessary pieces of code to handle coupling from arbitrary concentric primary structures are in front of me. Simply by introducing another mutual inductance matrix the coupling can be accounted for on a turn by turn basis, just as the secondary mutual inductance is dealt with now. At the same time the reaction of the secondary on the primary would also be taken care of. The result would be an input impedance spectrum for the primary, and the secondary V/I profiles would be those of the inductively driven coil. Once this is done, and if we can come to some sort of a satisfactory position on Q factors, and given an estimate of bulk impedance of corona/arc loading, then we could put your computer systems to good use in a distributed effort to tabulate tesla coil performances in the way mentioned above. With such a goal in mind I think any effort spent now in getting linux installed and getting to know how to use it is going to be well spent. Regards, -- Paul Nicholson, Manchester, UK. --
Maintainer Paul Nicholson, paul@abelian.demon.co.uk.