home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!mcsun!sun4nl!nikhefh!t68
- From: t68@nikhefh.nikhef.nl (Jos Vermaseren)
- Newsgroups: sci.math.symbolic
- Subject: Re: long symbolic runs
- Message-ID: <1850@nikhefh.nikhef.nl>
- Date: 1 Sep 92 07:46:33 GMT
- References: <JAFFER.92Aug29231912@camelot.ai.mit.edu>
- Reply-To: t68@nikhefh.nikhef.nl (Jos Vermaseren)
- Organization: Nikhef-H, Amsterdam (the Netherlands).
- Lines: 20
-
- I have run rather long programs with FORM (it was designed for that).
- For instance: we have run here some integrals in QCD that took together
- 154 hours on an Apollo DN10000. The final answers fit on four lines.
- In the middle there was an integral that took about 200 Mbytes of diskspace.
- The program is very much optimized so with the exception of (very maybe)
- schoonschip no other program can handle this (to preempt requests for people
- who would like to try this with their favorite system). The strategy was
- to parcel the program a little bit in handier chunks and let UNIX make
- run the show.
- In another big run that took 305 hours on a NeXT station (with a thunder
- storm above my appartment after about 10 days. that makes me nervous) I
- have been trying something with designing Groebner bases.
- And YES, my conclusion with things like this is that there should be
- a mechanism to save intermediate results so that the program may be
- restarted. (As far as this is possible. Some problems won't allow this.
- Backing up 200 Mbytes is not always possible cq desirable).
- Most designers however never run monsterjobs like this, so they never think
- of this.
-
- Jos Vermaseren
-