使用JMH实际测试EhCache、GuavaCache和Caffeine之间的性能
测试代码
EhCache
@BenchmarkMode({Mode.AverageTime}) @OutputTimeUnit(TimeUnit.MICROSECONDS) @Warmup(iterations=3, time = 5, timeUnit = TimeUnit.MILLISECONDS) @Measurement(iterations=5,time = 1,timeUnit = TimeUnit.SECONDS) @Threads(8) @Fork(3) @State(Scope.Thread) public class EhCacheTest { private static CacheManager cacheManager = CacheManagerBuilder.newCacheManagerBuilder() .build(true); private static Cache<String, String> cache = cacheManager.createCache("myCache", CacheConfigurationBuilder.newCacheConfigurationBuilder(String.class, String.class, ResourcePoolsBuilder.newResourcePoolsBuilder().heap(100, MemoryUnit.MB)) .withExpiry(ExpiryPolicyBuilder.timeToLiveExpiration(Duration.ofSeconds(1L))).build()); static { cache.put("test","test"); } @Benchmark public void test(){ cache.get("test"); } }
Guava Cache
@BenchmarkMode({Mode.AverageTime}) @OutputTimeUnit(TimeUnit.MICROSECONDS) @Warmup(iterations=3, time = 5, timeUnit = TimeUnit.MILLISECONDS) @Measurement(iterations=5,time = 1,timeUnit = TimeUnit.SECONDS) @Threads(8) @Fork(3) @State(Scope.Thread) public class GuavaTest { private static Cache<String,String> cache = CacheBuilder.newBuilder() .maximumSize(100) .expireAfterWrite(1,TimeUnit.SECONDS) .build(); static { cache.put("test","test"); } @Benchmark public void test(){ cache.getIfPresent("test"); } }
Caffeine
@BenchmarkMode({Mode.AverageTime}) @OutputTimeUnit(TimeUnit.MICROSECONDS) @Warmup(iterations=3, time = 5, timeUnit = TimeUnit.MILLISECONDS) @Measurement(iterations=5,time = 1,timeUnit = TimeUnit.SECONDS) @Threads(8) @Fork(3) @State(Scope.Thread) public class CaffeineTest { private static Cache<String,String> cache = Caffeine.newBuilder() .maximumSize(100) .expireAfterWrite(1,TimeUnit.SECONDS) .build(); static { cache.put("test","test"); } @Benchmark public void test(){ cache.getIfPresent("test"); } }
说明,实际测试代码共有六种情况(其实还有别的情况,不加任何参数使Caffeine使用UnboundedLocalCache,性能应该还会改变,大家如果有兴趣可以自己尝试)
直接放结果
- 奇特的地方
- 在加过期时间的情况下三个缓存方案的性能均有所提升
- Guava不加过期时间的情况下高并发会OOM
- EhCache在加过期时间的情况下竟然比Guava的性能要好
- Caffeine读比写的性能要高很多
总结
- Guava使用jdk的Queue记录缓存的写读情况,导致OOM;而Caffeine使用Disruptor的RingBuffer数据结构记录
- Caffeine对写的所有线程共用一个RingBuffer;而对读的每个线程维护一个RingBuffer
- 使用Caffeine是你不会后悔的选择
- 可参考Caffeine