Thu Mar 3 22:21:38 CET 2011
On Mar 3, 2011, at 2:56 PM, Bob Cowdery wrote:
> One of the things I need to do is move large binaries of up to 4096
> floats around between processes. As there are real-time constraints on
> processing this data I just wanted to check a few things.
> The docs say that large binaries are zero copy (presumable only in the
> same node). How exactly does this work in a lock free environment?
Large binaries are stored in a separate heap and are garbage collected separately.
A reference is passed around between processes (on the same node)
> 1> T=[1.0].
> 2> B=term_to_binary(T).
> 3> size(B).
> The docs also say that binary data is efficient. Why does 1 float
> convert to 39 bytes? This ratio only reduces slightly with larger lists
> . If erlang holds floats internally in 64 bit format I don't understand
> why it needs 39 bytes.
First of all, in your example, T is a list containing one float element.
Secondly, if you want to store a float in a binary, a better way would be
1> F = 1.0.
2> B = <<F/float>>.
More information about the erlang-questions