Date: Wed, 21 Aug 1996 11:42:32 -0700
    From: <Franz.COM at dm> (David Margolies)

      We have some large but interactive applications which we are in the
      process of porting from Lucid Lisp to Allegro CL (4.3).  Sadly, we
      have run into a show-stopper for delivering our applications: global
      GCs which take 8 or 11 minutes.  These applications are in decision
      support roles presenting massive amounts of data intelligibly to our
      users, and our users need to act on that information quickly -- within
      minutes.  Waiting for 10 or so minutes for a GC to complete is not
      acceptable.  This is exactly the sort of behavior that has dirtied the
      name 'Lisp' in the minds of the public; I had thought we had put this
      behind us years ago.

    I sort of wish that you had given us a chance to address this issue
    before attacking on our product and Lisp in general on a public
    forum.  It may very well be that this problem cannot be solved, but if
    it is, it is your attack that will be remembered by readers of the
    public forum rather than our solution.  As far as I can tell, neither
    you nor anyone else ar Ascent has ever asked us about tuning garbage
    collection or gc performance at all.

I'm sorry I failed to communicate clearly, but you have misunderstood.
It was not my intent to attack Lisp or your product.  I _am_ asking
about tuning garbage collection.  Perhaps you picked up the intensity
of my plea due to the seriousness of this problem when our optimism
for a imminent delivery was high and thus the resultant frustration
may have appeared to be an attack.  My comment about "dirtying the
name 'Lisp'" was not an attack, but was meant to say that if we
delivered our application with this GC behavior, it would reinforce
the negative stereotypes that much of the 'public' already has about
Lisp and make it that much harder to propose further work in Lisp.

Believe me, I appreciate the support Franz has supplied Ascent so far,
and if I thought the situation was hopeless, I would not have asked
for help.  Nevertheless, it must be understood that this is a very
serious situation; if the performance cannot be improved, delivery in
Allegro will be impossible.  Our work in other Lisp implementations
did not prepare us for the GC behavior we just found.  I cast my net
widely in the hopes of gathering the fruits of others experiences in
order to gain the most chance of improving the GC performance.

I will answer the rest of your message soon, with the further
details you asked about.  I first wanted to set the record straight
about my intentions in the same public forum in which my original
message was taken as an attack.

		- Pat
-------------------------------------------------------------------------------
  Patrick A. O'Donnell			  Ascent Technology, Inc.
 Development Scientist			 64 Sidney St, Cambridge, MA 02139-4136
<ascent.com at pao>				617-225-0850 x305