Benchmarks
The benchmarks use Java microbenchmark harness to provide an accurate analysis. The caches are configured as,
- Caffeine and ConcurrentLinkedHashMap size their internal structures based on the number of CPUs
- Guava was configured with a concurrency level of
64(defaults to4to reduces memory usage). Note that Guava #2063 would resolve the poor performance, but has been backlogged for years (up to 25x faster!). - Ehcache v2 is internally hardcoded to 100 segments, whereas v3 is not segmented
- Infinispan "old" is a Guava-like cache configured with a concurrency level of
64 - Infinispan "new" is a rewrite using a lock-free deque (default in v7.2+)
Run on a MacBook Pro i7-4870HQ CPU @ 2.50GHz (4 core) 16 GB Yosemite.
In this benchmark the caches are unbounded, fully populated, and the computation returns a constant value. This benchmark demonstrates the overhead due to locking when the entry is present. Caffeine performs a lock-free prescreening before falling back to an atomic compute if absent call. The scenarios graphed are all threads retrieving a single entry ("sameKey") and threads retrieving different keys based on a Zipf distribution ("spread").
In this benchmark 8 threads concurrently read from a cache configured with a maximum size.
In this benchmark 6 threads concurrently read from and 2 threads write to a cache configured with a maximum size.
In this benchmark 8 threads concurrently write to a cache configured with a maximum size.
The benchmarks were run on an Azure G4 instance, the largest available during a free trial from the major cloud providers. The machine was reported as a single socket Xeon E5-2698B v3 @ 2.00GHz (16 core, hyperthreading disabled), 224 GB, Ubuntu 15.04.
| Cache | same key | spread |
|---|---|---|
| ConcurrentHashMap | 29,679,839 | 65,726,864 |
| Caffeine | 1,581,524,763 | 530,182,873 |
| Guava | 25,132,366 | 114,608,951 |
| Unbounded | ops/s (8 threads) | ops/s (16 threads) |
|---|---|---|
| ConcurrentHashMap (v8) | 560,367,163 | 1,171,389,095 |
| ConcurrentHashMap (v7) | 301,331,240 | 542,304,172 |
| Bounded | ||
| Caffeine | 181,703,298 | 382,355,194 |
| ConcurrentLinkedHashMap | 154,771,582 | 313,892,223 |
| LinkedHashMap_Lru | 9,209,065 | 13,598,576 |
| Guava (default) | 12,434,655 | 10,647,238 |
| Guava (64) | 24,533,922 | 43,101,468 |
| Ehcache2_Lru | 11,252,172 | 20,750,543 |
| Ehcache3_Lru | 11,415,248 | 17,611,169 |
| Infinispan_Old_Lru | 29,073,439 | 49,719,833 |
| Infinispan_New_Lru | 4,888,027 | 4,749,506 |
| Unbounded | ops/s (8 threads) | ops/s (16 threads) |
|---|---|---|
| ConcurrentHashMap (v8) | 441,965,711 | 790,602,730 |
| ConcurrentHashMap (v7) | 196,215,481 | 346,479,582 |
| Bounded | ||
| Caffeine | 144,193,725 | 279,440,749 |
| ConcurrentLinkedHashMap | 63,968,369 | 122,342,605 |
| LinkedHashMap_Lru | 8,668,785 | 12,779,625 |
| Guava (default) | 11,782,063 | 11,886,673 |
| Guava (64) | 22,782,431 | 37,332,090 |
| Ehcache2_Lru | 9,472,810 | 8,471,016 |
| Ehcache3_Lru | 10,958,697 | 17,302,523 |
| Infinispan_Old_Lru | 22,663,359 | 37,270,102 |
| Infinispan_New_Lru | 4,753,313 | 4,885,061 |
| Unbounded | ops/s (8 threads) | ops/s (16 threads) |
|---|---|---|
| ConcurrentHashMap (v8) | 60,477,550 | 50,591,346 |
| ConcurrentHashMap (v7) | 46,204,091 | 36,659,485 |
| Bounded | ||
| Caffeine | 55,281,751 | 48,295,360 |
| ConcurrentLinkedHashMap | 23,819,597 | 39,797,969 |
| LinkedHashMap_Lru | 10,179,891 | 10,859,549 |
| Guava (default) | 4,764,056 | 5,446,282 |
| Guava (64) | 8,128,024 | 7,483,986 |
| Ehcache2_Lru | 4,205,936 | 4,697,745 |
| Ehcache3_Lru | 10,051,020 | 13,939,317 |
| Infinispan_Old_Lru | 7,538,859 | 7,332,973 |
| Infinispan_New_Lru | 4,797,502 | 5,086,305 |
