benchmark the least-recently-used caches which are available on npm
An LRU cache is a cache with bounded memory use. The point of a cache is to improve performance, so how performant are the available implementations?
LRUs achive bounded memory use by removing the oldest items when a threashold number of items is reached. We measure 3 cases, adding an item, updating an item, and adding items which push other items out of the LRU.
There is a previous benchmark but it did not describe it's methodology. (and since it measures the memory used, but tests everything in the same process, it does not get clear results)
I run a very simple benchmark. In three phases, set the LRU to fit max N=100,000 items, add N random numbers to the cache, with keys 0-N, then update those keys with new random numbers, then evict those keys, by adding keys N-2N.
Operations per millisecond (higher is better)
(see updated results at the bottom!)
name, set, update, evict
lru-native, 621, 901, 654
modern-lru, 415, 508, 336
lru-cache, 199, 521, 238
lru_cache, 2778, 8333, 23
tiny-lru, 2222, 4000, 23
lru, 1220, 2128, 7
simple-lru-cache, 2128, 6250, 7
mkc, 336, 7, 2
lru-fast, 6250, 2941, 7
faster-lru-cache, 1, 1, 1
secondary-cache, 917, 7, 2
We can group the results in a few categories,
- all rounders (lru-native, modern-lru, lru-cache) where the performance to add update and evict are comparable.
- fast-write, slow-evict (lru_cache, tiny-lur, lru, simple-lru-cache, lru-fast) these have better set/update times, but for some reason are quite slow to evict items!
- slow in at least 2 categories (mkc, faster-lur-cache, secondary-cache)
This is very early results, take any difference smaller than an order of magnitude with a grain of salt. It is necessary to measure the statistical significance of the results to know accurately the relative performance of two closely matched implementations.
I also didn't test the memory usage. This should be done running the benchmarks each in a separate process, so that the memory used by each run is not left over while the next is running.
It appears that all-round performance is the most difficult to achive, and there are no implementations that have very high eviction performance. It is also interesting that of the implementations which can set and update very fast, none can evict quickly. Also, some have faster add than update, and some faster update than add.
modern-lru gets pretty close to lru-native perf.
I think there opportunities for better perf. For example, it should be possible to avoid an alloction when evicting by recycling the least recently used item ito the new value. I would expect this to also have better GC behaviour.
I implemented a new LRU algorithm hashlru that is simpler and faster across the board. It's O(1) like LRU, but does less per operation.
These results also include read times.
name, set, get1, update, get2, evict
hashlru, 6250, 8333, 6250, 7143, 4167
lru-native, 714, 1053, 935, 1042, 510
modern-lru, 239, 581, 498, 578, 370
lru-cache, 300, 1449, 592, 1786, 247
lru_cache, 3226, 14286, 11111, 12500, 22
lru, 1887, 2778, 1724, 2857, 7
simple-lru-cache, 3448, 6667, 6667, 9091, 7
mkc, 410, 7, 3, 2, 1
lru-fast, 1493, 4762, 12500, 14286, 7
faster-lru-cache, 1, 1, 1, 1, 1
secondary-cache, 935, 7, 3, 2, 1
I removed tiny-lru because it was crashing in the get phase.
javascript is generally slow, so one of the best ways to make it fast is to write less of it. LRUs are also quite difficult to implement (linked lists!). In trying to come up with a faster LRU implementation I realized that something far simpler could do the same job. Especially given the strengths and weaknesses of javascript, this is significantly faster than any of the other implementations, including the C implementation. Likely, the overhead of the C<->js boundry is partly to blame here.
MIT