I have just discovered a bug in DIST2 due to the asymetrical nature of the
BITNET distances. You see, I had designed my algorithm in a mathematical way,
ie with equations and everything, and, as far as topology is concerned, d(a,b)
= d(b,a) for any (a,b) you might dream of. Now this doesn't quite work with
LSVBITFD distances, because some links are not weighted the same way in the
two directions. Thus, d(LEPICS,OHSTVMA) = 92 but d(OHSTVMA,LEPICS) = 95,
d(OHSTVMA,CEARN) = 94 and here you are, the algorithm decides to do a direct
send to OHSTVMA since CEARN is obviously "in the other direction" (and it also
does a direct send to CEARN, just as if the topology was: CEARN <--> LEPICS
<--> XXX (...) <--> OHSTVMA).
Anyway, it is not a very serious bug since the files do get sent and do reach
their destination without problem. So, there is no reason I should rush to
send the fix to everybody. The people who asked me for a beta-test of the
improved DIST2 algorithm will get a copy of the new fixed LSVDIST2 (I didn't
keep a list of those persons so I might well forget 1 or 2 - just drop me a
note if you don't get the file within a few days). If your site is intensively
using DIST2, you might wish to consider asking for a beta-test copy, too.
Otherwise just wait for the next release.
Eric
|