tarekziade/bench-db

Disk-based id lookup

Store 1 Million string ids on disk and provide a fast id lookup, without loading everything in memory.

Implementations:

• dbm - ndbm (stdlib)
• sqlite3 (stdlib)
• pysos -- https://github.com/dagnelies/pysos
• tkrzw -- https://dbmx.net/tkrzw/api-python
• ldbm -- https://github.com/Dobatymo/lmdb-python-dbm

TLDR;

• dbm has faster lookups!
• sqlite3 is smaller on disk
• tkrzw looks promising -- needs tweaking

dbm (ndbm)

Maximum RSS: 11MiB

Adding 1 million ids and timestamps
took 3.8714396953582764 seconds
Final count 1000000
max id len is 10
Looking for one id
0.00015072201495058835 seconds
Size of DB on disk is 95.359375 MiB

sqlite3

Maximum RSS: 13MiB

Adding 1 million ids and timestamps
took 2.6603636741638184 seconds
Final count 1000000
max id len is 10
Looking for one id
0.012375471997074783 seconds
Size of DB on disk is 34.44921875 MiB

tkrzw

Maximum RSS: 19MiB

Adding 1 million ids and timestamps
took 10.578145027160645 seconds
Final count 1000000
max id len is 10
Looking for one id
0.00043975599692203104 seconds
Size of DB on disk is 55.308563232421875 MiB

Pysos

Maximum RSS: 300MiB

Adding 1 million ids and timestamps
took 15.620530843734741 seconds
Final count 1000000
max id len is 10
Looking for one id
0.0018717409984674305 seconds

LDBM

Maximum RSS: 19MiB

Adding 1 million ids and timestamps
took 1.4407689571380615 seconds
Final count 1000000
max id len is 10
Looking for one id
0.000999119016341865 seconds
Size of DB on disk is 81MiB