pathalias slowdown

[Mark E. Mallett]

In article <RON.90May30172314 at mlfarm.uucp> ron at mlfarm.uucp (Ronald Florence) writes:
>We use pathalias (9.1 87/10/04) with Xenix 2.3.2, on a ps2/80 with six
>megabytes of ram.  Building our map normally takes ~2 minutes.
>
>The recent batch of maps apparently has taken our pathalias over some
>threshold.  Without the map u.usa.ca.1, the build takes the same ~2
>minutes.  With all of the maps included on the command line, pathalias
>churns and swaps, slowly growing in size.  An hour later, when I
>finally killed the process, it was at 5704K (according to ps);

etc.

I spent quite a bit of yesterday tracking this down.  The problem
seems to be that bcsinc (in u.usa.ca.1) and rt1 (in u.usa.nm.1)
declare each other to be terminal links with DEAD cost.  I was
able to replicate the problem with a 3-line pathalias input file:

bcsinc	<rt1>(DEAD)
ditka	<rt1>(DEMAND+FAST),	zinn(HOURLY+FAST)
rt1	<bcsinc>(DEAD)

Running pathalias with trace options shows that rt1->bcsinc and
bcsinc->rt1 (that is, the terminal defs of them) are trying to
be resolved coninuously, with ever-increasing costs and links.

I got around it by adding this to my local-defs file (which is
a file that I feed into the pathalias run to supply local definitions
and overrides):

bcsinc	rt1(DEAD)
rt1	bcsinc(DEAD)

In fact, I originally had "bcsinc" spelled wrong in the addition, and
it still worked.  Apparently all it needs is some other route between
the two sites.

I checked this three-line problem with a neighbor, who ran it with
the -l zinn option (to pretend it was me), and they had the same
results -- pathalias went into an infinite loop.

Incidentally, I also checked this with another neighbor who does NOT
have the problem.  He says he is running an "unreleased" version of
pathalias 9.1, and in fact the trace output is quite different from
mine -- both in the formatting (terminals have <> instead of trailing
') and in the costs for DEAD and back links.  I can hardly wait to see
the update...

-mm-
-- 
Mark E. Mallett  Zinn Computer Co/ PO Box 4188/ Manchester NH/ 03103 
Bus. Phone: 603 645 5069    Home: 603 424 8129     BIX: mmallett
uucp: mem at zinn.MV.COM  (  ...{decvax|elrond|harvard}!zinn!mem   )
Northern MA and Southern NH consultants:  Ask (in mail!) about MV.COM