[erlang-questions] 64-bit erlang

[Ulf Wiger (TN/EAB)]

Alex Arnon skrev:
> 
> 
> On Nov 28, 2007 11:02 AM, Ulf Wiger (TN/EAB) <ulf.wiger@REDACTED 
> <mailto:ulf.wiger@REDACTED>> wrote:
> ...
> 
> 
>     How about using 32-bit words inside the process heap, and
>     "far pointers" to any off-heap data (including other
>     processes)?
> 
> ...
> 
> * runs away to the hills *
> 
> How insensitive - not everyone has recovered from 16-bit x86 assembly 
> terminology :)

Apologies, but at least it got the discussion started. (:


> But seriously - I this would make sense.
> Will this allow Mnesia/{d}ets tables to be unlimited then?

Rewriting dets to support larger files is of course a
separate issue, but ets tables can become very large
in 64-bit Erlang.

I ran some tests two years ago on a 16 GB machine:

Summary:
http://www.erlang.org/pipermail/erlang-questions/2005-November/017740.html

More detail:
http://www.erlang.org/ml-archive/erlang-questions/200511/msg00118.html

One of the notable observations from that test was perhaps that
a disc_copy table using up 15 GB of RAM used only 3.2 GB on disk.

When I switched to term_to_binary(Value,[compressed]), I read the
table as using 686 MB of RAM, compared to 373 MB on disk, but
that probably overlooked the fact that much of the binary data
was not reported as 'ets' usage, but rather as 'binary'...

BR,
Ulf W