Re: A problem about fresh and stale KCL

[bean@cs.utexas.edu (Richard Stuart Bean)]

I don't have an answer; but I do think it's a general problem affecting
(potentially) all users.  I've experienced similar performance problems
w/ the public version of AKCL at UT in Austin.  I was working on a very
simple resolution-based theorem prover.  After 10-12 hrs of real time,
performance would slow to a crawl.  A "ps" would show high CPU util, and
a very large (50-70% of physical memory) memory residency.

I started saving state along the way so that I could periodically stop
computation (state was saved in a dribble file), reload and manually
munge on the state data w/ a FRESH copy of AKCL, and restart the
computation.  Typically I saw a 10x speed increase until it started
bogging down again.  And the memory usage went WAY down as well.

Richard Bean
bean@cs.utexas.edu