|Compiling with Continuations/Andrew Appel firstname.lastname@example.org (1992-01-23)|
|Re: Compiling with Continuations/Andrew Appel email@example.com (1992-01-27)|
|Re: Compiling with Continuations/Andrew Appel firstname.lastname@example.org (1992-01-27)|
|Re: Compiling with Continuations/Andrew Appel email@example.com (1992-01-27)|
|Re: Compiling with Continuations/Andrew Appel firstname.lastname@example.org (Vincent Delacour) (1992-01-28)|
|Re: Compiling with Continuations/Andrew Appel email@example.com (1992-01-29)|
|Re: Compiling with Continuations/Andrew Appel firstname.lastname@example.org (1992-01-29)|
|Re: Compiling with Continuations/Andrew Appel email@example.com (1992-01-29)|
|Re: Compiling with Continuations/Andrew Appel firstname.lastname@example.org (1992-01-30)|
|From:||email@example.com (Hans Boehm)|
|Keywords:||storage, optimize, ML, books|
|Date:||Wed, 29 Jan 1992 10:08:32 PST|
firstname.lastname@example.org (Vincent Delacour) writes:
> ... However, unless you work on a cray or a pc, the real cost of
>operations in memory can not be measured in instructions! The huge amount
>of memory used by sml-nj puts an enormous burden on the virtual memory
>system of the machine it runs on. Appel recommands a ratio H/A of about 7
>(15 without the generational version of the GC). Now consider you run a
>program asking for, say 20 Mbytes for a certain amount of time. With
>SML-NJ, you are really asking for 140 Mbytes of virtual memory.
There are also other complications. Even pc's often have caches these
days, and stack allocation tends to be much more cache friendly. This can
make a huge difference in execution time. On some architectures, there
are similar problems related to a shortage of memory mapping resources,
even if the machine has enough physical memory.
It's also worth remembering that just paging in 100 Mbytes once takes on
the order of 15 minutes on a typical workstation. There is essentially no
such thing as a 140 Mbyte application with good VM behavior.
As was pointed out by previous messages, this doesn't mean the techniques
are inapplicable; you just have to be careful how and where you apply
Hans-J. Boehm (email@example.com)
Return to the
Search the comp.compilers archives again.