lilcache

Interprocess and thread safe light weight cache

Note: This is under development and has bad performance (concurrency, response time, error prone, etc. you name it!). It is not ready for production use. If you like taking risks, I won't stop you from using it.

How to

$ pip install lilcache

$ python
>>> import lilcache
>>> lilcache.init()
>>> lilcache.set('foo', 'bar')
True
>>> lilcache.get('foo')
'bar'

Benchmarks

lilcache vs redis

Note: lilcache was running without connection pool

For obvious reasons, redis is performing better than lilcache and the performance gap increases a lot with increasing concurrency.

Benchmarking tool run on Intel(R) Core(TM) i7-10700 CPU @ 2.90GHz

$ python3 tests/benchmarks/performance.py
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| Cache type   |   Concurrency |   Number of operations (get/set/delete) |   Mean response time (ms) |   Lowest time (ms) |   Highest time (ms) |
+==============+===============+=========================================+===========================+====================+=====================+
| lilcache     |             1 |                                    1000 |                 0.0368855 |          0.0350475 |            0.200033 |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |             1 |                                    1000 |                 0.0323646 |          0.0274181 |            1.01137  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| lilcache     |             5 |                                    1000 |                 0.0754766 |          0.0412464 |            0.260592 |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |             5 |                                    1000 |                 0.0399675 |          0.0283718 |            1.13201  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| lilcache     |            10 |                                    1000 |                 0.185351  |          0.142813  |            0.473261 |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |            10 |                                    1000 |                 0.06568   |          0.0295639 |            1.37901  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| lilcache     |            20 |                                    1000 |                 0.376953  |          0.166655  |            1.16491  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |            20 |                                    1000 |                 0.102541  |          0.0293255 |            1.88661  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+

lilcache vs redis

Note: lilcache was running with connection pool High number of threads is performing worse than no threads. Seems like Python GIL is biting us here.

$ python3 tests/benchmarks/performance.py
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| Cache type   |   Concurrency |   Number of operations (get/set/delete) |   Mean response time (ms) |   Lowest time (ms) |   Highest time (ms) |
+==============+===============+=========================================+===========================+====================+=====================+
| lilcache     |             1 |                                    1000 |                 0.0102515 |         0.00786781 |            0.223398 |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |             1 |                                    1000 |                 0.0333166 |         0.0283718  |            1.07622  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| lilcache     |             5 |                                    1000 |                 0.0693438 |         0.0123978  |            0.691175 |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |             5 |                                    1000 |                 0.0532017 |         0.0312328  |            1.34492  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| lilcache     |            10 |                                    1000 |                 0.233874  |         0.0140667  |            2.32434  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |            10 |                                    1000 |                 0.0799031 |         0.0324249  |            2.46119  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| lilcache     |            20 |                                    1000 |                 0.578124  |         0.0114441  |            8.94666  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+
| redis        |            20 |                                    1000 |                 0.100958  |         0.0321865  |            4.00949  |
+--------------+---------------+-----------------------------------------+---------------------------+--------------------+---------------------+

Timeline

• Write unittests
• Stress testing
• Improve error handling
• Extension to current implementation: Implement connection pool and client reuse connections
• Implementation 2: Twisted bsaed concurrency
• Implementation 3: Incordinated cache (file db + locks for synchronization)
• For each implementation, perform benchmark tests and stress testing
• Release to PyPi (pre-alpha release done)
• Implementation 4: Named PIPE based approach and full duplex communication
• Add more features (snapshot, Cache expiry, etc.)
• Production ready code

Contributing

Contributions are welcome! For more details, please read the CONTRIBUTING.md