[gclist] Re: Articles expiring too fast!

[Henry G. Baker]

Fiterman:
> It strikes me that a simple heuristic might help. News groups
> like talk.politics.misc with very large volumes would actually
> benefit from a faster expiration while those with small volumes
> like comp.compiler.misc would be hurt. So a first shot would
> be to tune the expiration to the volume of the group.
> 
> A minor twiddle on this would be not to kill any article within
> three days of its getting a followup.

I'll bet that there's a classical optimization problem here, just
like putting disk drives of multiple speeds on a computer system, or
areas of multiple allocation rates on a GC system (Tony Hoare's paper,
Stefan Arnborg's paper, etc., from the early 1970's).

In many of these Lagrange multiplier optimization problems, the answer
turns out to be proportional to the square root of something.  In the
case of the news problem, the answer may be to make the sizes of the
news files proportional to the square roots of the rates of their
daily volumes.  I haven't done the model or the math, though.

-- 
Henry Baker
www/ftp directory:
ftp.netcom.com:/pub/hb/hbaker/home.html