Re: [Caml-list] Question about garbage collection and impact on performance

[Jeff Schultz <jeffschultz@runbox.com>]

On 22/12/2013 14:04, David Sheets wrote:
> On Sun, Dec 22, 2013 at 2:25 AM, Jon Harrop <jon@ffconsultancy.com> wrote:
>> Richard Jones wrote:
>>>> My personal impression is that the question is not that well-posed:
>>>> - if you assume infinite memory, you don't actually need a GC (and for
>>>> any input you can tweak the GC setting to make sure no collection
>>>> happens)
>>>
>>> How could "infinite" memory be implemented without affecting the runtime
>> of programs on such a machine?
>>
>> I guess O(1) lookup would actually be O(n^(1/3)) due to that speed of light
>> thing. ;-)
>
> I think there are some fundamental limits to information density in
> the local universe. I reply here, though, to propose that the lookup
> would be O(n^(1/2)) because, locally, space is Euclidean. I'm not sure
> how you got the 1/3 exponent.

Eight times as much memory, uniformly packed in three dimensions, is at 
most twice as long in any one dimension.

However, IIRC, Gene Amdahl observed that the limiting characteristic of 
memory size versus speed is actually the surface area of the sphere 
containing the memory, because that controls heat dissipation.  This 
gives an asymptotic square root law.  Various assumptions could 
challenge that, but not as long as memory consumes power in proportion 
to its size.


     Jeff Schultz