> I need ideas.  I've included below some output from (room) and verbose
> GC output from one of our applications.  (In this sample, I had built
> the application to start with one very large "oldspace", believing
> that that might help the performance.
>
> [...]
>
> New #x39fa000(11567104)   735:17607     250:4       10154928:1199672
> New #x2ef2000(11567104)    -----         -----            -----
> Old #x2ab2000(4456448)      0:0           0:0        2947104:1507000
> Old #x98f50(44142768)       0:984354    226:65879     338464:34271664

If you are loading lots of rather permanent data at once, I've found
the TENURING macro to be very helpful.  It can help prevent newspace
from growing so large (which is of course a big problem since it takes up
twice its usable space).

I assume you've already tried to cut down on garbage creation - this
is usually more effective than gc tuning (although after your
experience with lucid I can understand your focus on the
collector). Try the allegro space profiler if you haven't already.  Also,
READ can be very inefficient -  let me know if you want functions to
read numbers from a text stream.


Wheeler
-- 
Wheeler Ruml, Aiken 220, <eecs.harvard.edu, at ruml> (617) 496-1066 (fax)
http://www.das.harvard.edu/users/students/Wheeler_Ruml/Wheeler_Ruml.html