GC Method (was Questions for Discussion)

Bill House bhouse@dazsi.com
Wed, 28 May 1997 13:34:37 -0700

IDK about Hennessey, but the citation for the Wilson & Johnstone piece is:

 author      ="Wilson, Paul R. and Johnstone, Mark S.",
 title       ="Real-Time Non-Copying Garbage Collection",
 address     ="Washington D.C.",
 booktitle   ="ACM OOPSLA Workshop on Memory Management and 
               Garbage Collection",
 month       ="September",
 year        ="1993",
 scope       ="gc",
 abstractURL ="http://www.cs.utexas.edu/oops/papers.html#real-time-gc",
 documentURL ="file://ftp.cs.utexas.edu/pub/garbage/GC93/wilson.ps"

Unfortunately, I don't seem to be able to access the FTP site.

Jones & Lins mention Kaleida as well, but don't say anything nice about its
performance, which they call "disappointing."  The authors apparently intended
that all of these implementations be considered lacking for truly RT GC, but
did not find fault with the algorithms themselves. 

Seems to me that being good for RT is a better choice than non-RT GC schemes
(I'm defining RT as GC with a guaranteed worst-case that is still fast enough
to use). I say that because the lack of RT GC would seem to preclude using
LispOS for RT work, at least without turning off the GC. Since control systems
are an applications area where AI is hot, it would be a shame to summarily
dismiss the RT part of it. 

I guess the alternative to having truly RT GC is having a way to turn off the
GC, or to declare objects as belonging in non-GCed memory. Maybe such options
are desirable IAC -- doesn't Eiffel provide something like that?

Bill House
The views expressed are mine alone,
unless you agree with me.