A data structure for optimistic concurrency control on ranges of bitwise-lexicographically-ordered keys.

Intended as an alternative to FoundationDB's skip list.

Hardware for all benchmarks is an AMD Ryzen 9 7900 with (2x32GB) 5600MT/s CL28-34-34-89 1.35V RAM.

Compiler is Ubuntu clang version 20.0.0 (++20241029082144+7544d3af0e28-1~exp1~20241029082307.506).

Microbenchmark

Skip list

ns/op	op/s	err%	ins/op	cyc/op	IPC	bra/op	miss%	total	benchmark
159.65	6,263,576.52	1.6%	2,972.36	820.37	3.623	504.59	0.0%	0.01	point reads
156.32	6,397,320.65	0.7%	2,913.62	806.87	3.611	490.19	0.0%	0.01	prefix reads
229.18	4,363,293.65	1.2%	3,541.05	1,219.75	2.903	629.33	0.0%	0.01	range reads
363.37	2,752,026.30	0.3%	5,273.63	1,951.54	2.702	851.66	1.7%	0.01	point writes
364.99	2,739,787.02	0.3%	5,250.92	1,958.54	2.681	839.24	1.7%	0.01	prefix writes
242.26	4,127,796.58	2.9%	3,117.33	1,304.41	2.390	541.07	2.8%	0.02	range writes
562.48	1,777,855.27	0.8%	7,305.21	3,034.34	2.408	1,329.30	1.3%	0.01	monotonic increasing point writes
122,688.57	8,150.72	0.7%	798,766.00	666,842.00	1.198	144,584.50	0.1%	0.01	worst case for radix tree
41.71	23,976,459.34	1.7%	885.00	219.17	4.038	132.00	0.0%	0.01	create and destroy

Radix tree (this implementation)

ns/op	op/s	err%	ins/op	cyc/op	IPC	bra/op	miss%	total	benchmark
12.63	79,186,868.18	1.4%	241.61	64.76	3.731	31.64	0.8%	0.01	point reads
14.48	69,078,073.40	0.3%	292.42	74.69	3.915	41.49	0.5%	0.01	prefix reads
34.37	29,094,694.11	0.2%	759.53	179.77	4.225	100.38	0.2%	0.01	range reads
19.34	51,713,896.36	0.7%	369.70	101.81	3.631	47.88	0.6%	0.01	point writes
39.16	25,538,968.61	0.2%	653.16	206.77	3.159	89.62	0.8%	0.01	prefix writes
40.58	24,642,681.12	4.7%	718.44	216.44	3.319	99.28	0.6%	0.01	range writes
78.77	12,694,520.69	3.8%	1,395.55	421.73	3.309	249.81	0.1%	0.01	monotonic increasing point writes
287,760.50	3,475.11	0.5%	3,929,266.50	1,550,225.50	2.535	639,064.00	0.0%	0.01	worst case for radix tree
104.76	9,545,250.65	3.1%	2,000.00	552.82	3.618	342.00	0.0%	0.01	create and destroy

"Real data" test

Point queries only, best of three runs. Gc ratio is the ratio of time spent doing garbage collection to time spent adding writes or doing garbage collection. Lower is better.

skip list

Check: 4.39702 seconds, 370.83 MB/s, Add: 4.50025 seconds, 124.583 MB/s, Gc ratio: 29.1333%, Peak idle memory: 5.51852e+06

radix tree

Check: 0.987757 seconds, 1650.76 MB/s, Add: 1.24815 seconds, 449.186 MB/s, Gc ratio: 41.4675%, Peak idle memory: 2.02872e+06

hash table

(The hash table implementation doesn't work on range queries, and its purpose is to provide an idea of how fast point queries can be)

Check: 0.84256 seconds, 1935.23 MB/s, Add: 0.697204 seconds, 804.146 MB/s, Gc ratio: 35.4091%