AmeCache - Simple High-Performance Local Cache
Description
Implementation of various local Cache and test it benchmark performance
following Tour.
Based on them, reimplement a high-performance Cache according to BigCache, Considering sharding Lock and GC optimization
I call it Ame Cache, Ame is the ID of Carry from LGD Dota2 Esport Team.
Intro
SCache
Simple Cache. It implement three kind of cache strategies, LRU ; LFU ; FIFO. However, it use only one Lock to ensure the thread-safe. Geting Lock and releasing Lock will cost many performance overhead
FCache
Fast Cache. It optimize the performance through sharding records. Thus, two records from different shards won't share one lock. It improves the performance in Parallel Operation.
AmeCache
Copy the design pattern of bigCache.
Shard lock can help to relieve the load of concurrent lock.
Considering the GC optimization, only use basic data type in map. When Go version > 1.5, GC will regard map as only one object if the map only contains basic data type like different kinds of Integar type. So create a special data structure: bytequeue. It is an FiFO-like byte circle queue. For any k-v pair, AmeCache will serialize it into a record, and insert it into the ByteQueue. The index only maps the hash value of the key into the record's position in the byteQueue. Thus the index map only contains int type. During the GC process, the map will be treated as an object.
As for cache strategy, it implements FIFO. However, when allocating extra memory, it will change the tail and head pointer, which will break the FIFO rules. You can check the whole process in the below figure.
Benchmark Test
$ cd test/benchmark
$ go test -bench=. -benchmem ./... -timeout 30mgoos: darwin
goarch: amd64
pkg: github.com/lingze/localCache/test/benchmark
cpu: Intel(R) Core(TM) i5-7267U CPU @ 3.10GHz
BenchmarkMapSet-4 2574151 470.1 ns/op 208 B/op 2 allocs/op
BenchmarkSimpleCacheSet-4 1274461 860.3 ns/op 337 B/op 6 allocs/op
BenchmarkFastCacheSet-4 2236088 575.1 ns/op 240 B/op 6 allocs/op
BenchmarkConcurrentMapSet-4 1000000 1471 ns/op 339 B/op 8 allocs/op
BenchmarkBigCacheSet-4 2131330 645.4 ns/op 340 B/op 2 allocs/op
BenchmarkAmeCacheSet-4 1864418 635.0 ns/op 353 B/op 2 allocs/op
BenchmarkMapGet-4 3877886 346.6 ns/op 23 B/op 1 allocs/op
BenchmarkSimpleCacheGet-4 3217207 376.6 ns/op 23 B/op 1 allocs/op
BenchmarkFastCacheGet-4 5540294 214.8 ns/op 24 B/op 1 allocs/op
BenchmarkConcurrentMapGet-4 2957997 411.7 ns/op 23 B/op 1 allocs/op
BenchmarkBigCacheGet-4 2541733 494.0 ns/op 151 B/op 3 allocs/op
BenchmarkAmeCacheGet-4 2986725 401.2 ns/op 23 B/op 1 allocs/op
BenchmarkSimpleCacheSetParallel-4 1000000 1178 ns/op 369 B/op 6 allocs/op
BenchmarkFastCacheSetParallel-4 3458870 345.4 ns/op 244 B/op 6 allocs/op
BenchmarkConcurrentMapSetParallel-4 3468811 353.1 ns/op 190 B/op 5 allocs/op
BenchmarkBigCacheSetParallel-4 4113654 311.6 ns/op 349 B/op 3 allocs/op
BenchmarkAmeCacheSetParallel-4 3922638 398.0 ns/op 357 B/op 3 allocs/op
BenchmarkSimpleCacheGetParallel-4 8150322 165.3 ns/op 23 B/op 1 allocs/op
BenchmarkFastCacheGetParallel-4 10355632 131.4 ns/op 24 B/op 1 allocs/op
BenchmarkConcurrentMapGetParallel-4 3260037 549.2 ns/op 23 B/op 1 allocs/op
BenchmarkBigCacheGetParallel-4 5215998 197.8 ns/op 151 B/op 3 allocs/op
BenchmarkAmeCacheGetParallel-4 8396762 187.4 ns/op 23 B/op 1 allocs/op
BenchmarkSimpleCacheSetGetParallel-4 1000000 1627 ns/op 399 B/op 9 allocs/op
BenchmarkFastCacheSetGetParallel-4 2220926 536.1 ns/op 276 B/op 9 allocs/op
BenchmarkConcurrentSetGetParallel-4 867662 1998 ns/op 394 B/op 11 allocs/op
BenchmarkBigCacheSetGetParallel-4 2135581 747.5 ns/op 500 B/op 7 allocs/op
BenchmarkAmeCacheSetGetParallel-4 2352194 501.1 ns/op 362 B/op 4 allocs/op