[issue14381] Intern certain integral floats for memory savings and performance

[Kristján Valur Jónsson]

Kristján Valur Jónsson <kristjan at ccpgames.com> added the comment:

Yes, there is a measurable performance decrease in pybench arithmetic tests.

Integers don't fall out of arithmetic that often, true.  But integral floats are incredibly common in tabular data.  In a game such as Eve Online,  configuration data contains a lot of 0.0, 1.0, -1.0 and so on.  This patch saved us many megabytes on the server.  I can check again...

>>> [sys.getrefcount(float(i)) for i in range(-10, 11)]
[777, 4, 38, 9, 215, 691, 627, 185, 98, 603, 73180, 62111, 8326, 6225, 6357, 11737, 2906, 1393, 3142, 1145, 5601]
>>> sum([sys.getrefcount(float(i)) for i in range(-10, 11)])
185340
>>> 
This is on an idle server.  A server with lots of stuff going on will have this:
[16715, 184, 1477, 34, 1505, 27102, 3878, 1344, 6248, 974, 595889, 313062, 124072, 120054, 65585, 138667, 13265, 2499, 15677, 3175, 24821]
>>> sum([sys.getrefcount(float(i)) for i in range(-10, 11)])
1465155

About half of the interned floats are 0.0
On a 64 bit machine with each float taking 24 bytes, this is 35mb net.

An alternative could be to add a function for manual intering of floating point data, which one can use e.g. when reading tabular data.

----------

_______________________________________
Python tracker <report at bugs.python.org>
<http://bugs.python.org/issue14381>
_______________________________________