Re: Parallel & RT GC (was Re: Real-Time GC (was Re: Widespread C++...?)

[dhanley@matisse.eecs.uic.edu (David Hanley)]

Robert I. Eachus (eachus@spectre.mitre.org) wrote:
:     There is another major reason for putting the custom hardware in
: the memory system.  There is a fairly large class of problems which
: require time proportional to n^k lg2 n on any von Neuman architecture,
: but can be solved in time n^(k-1) lg2 on hardware which supports
: clearing large sections of memory to zero in constant time.  (Note
: that if k is one, you can get orders of magnitude improvement.)

        Hmmm.  While I suppose this is possible, I haven't yet run
across this "large body of algorithms" in which the execution time is
reduced by the order of magnitude when memory blocks can be zeroed out
very quickly.  Even if the algorithm did have to zero out the memory
itself, I don't see how that could take it from, say, n^2 to n^3.
Could you please point some of these out to me?

:      Since many memory systems can support this happy property, all you
: need to do to get the faster algorithms is to bypass the OS, or more
: politely, have an OS call which clears a section of memory.

        Hmm.  Not sure what this has to do with GC, really.  In any
case, it would be pretty easy to design memory hardware that would
zero out pages really quickly( just don't refresh them one cycle ).
And the bzero() function on your C compiler could be set to use it.
But mabye I'm just confused.  Could you please explaain if this is
wrong?

--
------------------------------------------------------------------------------
| David James Hanley  -- dhanley@lac.eecs.uic.edu  -- C++, OOD, martial arts |
| Laboratory for advanced computing      |      My employer barely KNOWS me. |
------------------------------------------------------------------------------
"There are trivial truths & there are great truths. The opposite of a
trivial truth is plainly false."
"The opposite of a great truth is also true."

-Neils Bohr