Session format: compare performance of msgpack vs. pickle

[Scooter Morris]

Once we have a msgpack session implementation, we need to compare the performance of it vs. our current approach.

[Greg Couch]

Here are some preliminary results using a session file that only has 4v8r in it (128780 atoms and 130160 bonds -- no ribbons):

v1 sessions use pickle, v2 session use msgpack

File size:

• compressed v2 session is 0.2% smaller than v1, uncompressed v2 session is 3.5% smaller than v1
• compressed sessions are 6 times smaller than uncompressed ones
• uncompressed v2 session is 31% larger than mmCIF file!

Speed:

opening the mmCIF file is still faster than opening the session

• compressed v1 session is 30% slower (current default)
• uncompressed v1 session is 23% slower
• compressed v2 session is 22% slower
• uncompressed v2 session is 16% slower

so v2 session restoring is 6% faster than restoring v1 sessions.

[Greg Couch]

For 4v8r, 85% of the restore time is in _RestoreManager.resolve_references(). Need equivalent of state.FinalizedState for restores.

[Greg Couch]

After putting FinalizedState object output, only 1.2% of time was spent in _RestoreManager.resolve_references(). In this case, there was a 6X speed up.