gistfile1.txt

import mmap
import ctypes


class Record(ctypes.Structure):
    _fields_ = [('index', ctypes.c_uint64),
                ('key', ctypes.c_char * 256)]

RECORD_COUNT = 256

storage_mem = mmap.mmap(-1, RECORD_COUNT * ctypes.sizeof(Record))
array = (Record * RECORD_COUNT).from_buffer(storage_mem)

for i in xrange(RECORD_COUNT):
    array[i].index = i
    array[i].key = 'item ' + str(i)

for i in xrange(RECORD_COUNT):
    assert array[i].index == i
    assert array[i].key == 'item ' + str(i)

Check out https://gist.github.com/53405d856137175aa59d

Using c_wchar_p in the Array causes the child processes to use about a quarter the amount of memory that the parent process uses (if they read all of the data or a large chunk).

Using Record in the Array causes the child processes to use FAR less memory.....WTF.