[plt-scheme] 3m defaults to --enable-account

At Sat, 17 Jan 2004 13:18:05 -0500, Shriram Krishnamurthi wrote:
> I appreciate your lengthy response, and in general I have a lot of
> respect for your insights on this topic.  In this message, however,
> you're being a bit too platitudinous (ahem).

Heh.

> What you've mostly
> provided is an account of the high-level differences, but (the SirMail
> instance aside) nothing concretely numeric.  I guess I (and quite
> possibly Jim and others) would be interested in actual stats.

I'm not sure how to respond to this outside of "so would I." Designing
and running memory benchmarks is not entirely trivial if you want good
data. It requires a little instrumentation and a lot of uninterrupted 
computer time, and that's only if you've got a number of interesting 
benchmark cases sitting premade in front of you (which I don't). I've 
not had a moment where my interest in these numbers outweighed other 
demands on my sadly finite time. My understanding is that lag/drag
results would be worse in all respects (instrumentation required, time
to run, direct usefulness of results to my current research), but I've
not looked into the matter deeply, and I'll defer to your experience if
you think otherwise.

When I have concrete numbers I will pass them on, and I do plan to
run these benchmarks at some point. At the moment all the insight I can
give Jim, yourself and the rest of the list is my experience. Given 
that this is comparing almost exactly the same system hooked up to the
two variants, I figured this would be useful preliminary data.

> expected.  The conservative GC seemed to be doing pretty well in the
> face of hard data collection.

Yes, and this is actually my argument. "Pretty well" is pretty bad 
after you let a program run for some length of time. As mentioned
previously, I can't give you benchmark-reinforced values for this, 
only my experience using these systems.

(If you'd like some of it: I've found running SirMail, given my email load,
can become problematic after a few days using the conservative system, and
stays happily well under 64M under 3m for weeks on end. On some programs, 
I've found DrScheme itself can become a serious problem after more than a few
hours of use, whereas I haven't had much of a problem with 3m. My personal
experience with the web server and handin server is extremely limited, and
you'd have to ask those people who use these over long time periods.)

> (As you know, leakage is not necessarily a GC strategy issue.  It

Yes. Conservative collection simply adds another source of potential leaks.
Which, unlike the leaks you describe, are very difficult to fix.


-Adam