shared-memory

Python Shared Memory LRU - POC Test. Performance:

local
    0.000235 ms/opp
functools.lru_cache
    0.000269 ms/opp
shared memory_lru v1 - 3 lists: key, prev, next
    0.007810 ms/opp
shared memory_lru v2 - list of (key, prev, next)
    0.005886 ms/opp
shared memory_lru v3 - list of (key, prev, next) - no LRU touch on __get__
    0.002728 ms/opp
shared memory_lru v4 - data in sm - 10% lru touch
    0.004762 ms/opp
current: lock + 13% lru touch
    0.004744 ms/opp
Manager().dict - no LRU
    0.030712 ms/opp
redis
    0.050342 ms/opp

Todo:

• self.root in shared memory
• better conflict handling in _del_index

Random notes

• marshal.dumps((a, b, b)) is 2x faster than a.to_bytes(...) + b.to_bytes() + c.to_bytes() but I still use to_bytes to write one value of the tuple, without reading
• 0.004ms to access a view is acceptable for the views cache of Odoo (~30 calls for a web page)