Benchmark for hashing "words" using the algorithms available with the JDK plus some that are part of Guava:
- MD5 (JDK 1.7)
- SHA-1 (JDK 1.7)
- SHA-256 (JDK 1.7)
- SHA-512 (JDK 1.7)
- Murmur3 128 (Guava 18.0)
- Murmur3 32 (Guava 18.0)
- CRC-32 (Guava 18.0)
- CRC-32C (Guava 18.0)
- SipHash (Guava 18.0)
The "words" are part of a corpus that is dynamically generated during the benchmark, the command line parameters set the size of the corpus and the length of the "words" it contains. The code uses JMH, and is a Maven project.
> mvn clean install
> java -jar target/benchmarks.jar HashingBenchmark -wi 10 -i 10 -f 1 -p corpusSize=500000 -p stringLength=12
> java -jar target/benchmarks.jar HashingJavaStringBenchmark -wi 10 -i 10 -f 1 -p corpusSize=500000 -p stringLength=12
The results here refer to corpus created with -p corpusSize=500000 -p stringLength=12i.e 500000 words of length 12, in the following JMH output, each result (Score in jmh-speak) is the average time in milliseconds it took to process all of the words in the corpus.
Benchmark (corpusSize) (hasherName) (stringLength) Mode Cnt Score Error Units
HashingBenchmark.hash 500000 MD5 12 avgt 10 108.614 ± 0.851 ms/op
HashingBenchmark.hash 500000 SHA-1 12 avgt 10 147.499 ± 3.498 ms/op
HashingBenchmark.hash 500000 SHA-256 12 avgt 10 186.499 ± 0.709 ms/op
HashingBenchmark.hash 500000 SHA-512 12 avgt 10 287.876 ± 3.372 ms/op
HashingBenchmark.hash 500000 MURMUR3_128 12 avgt 10 70.198 ± 0.988 ms/op
HashingBenchmark.hash 500000 MURMUR3_32 12 avgt 10 71.443 ± 1.115 ms/op
HashingBenchmark.hash 500000 SIP_HASH24 12 avgt 10 78.706 ± 0.609 ms/op
HashingBenchmark.hash 500000 CRC32 12 avgt 10 63.512 ± 1.642 ms/op
HashingBenchmark.hash 500000 CRC32C 12 avgt 10 43.908 ± 0.743 ms/op
HashingJavaStringBenchmark.javaStringHash 500000 12 avgt 10 20.907 ± 6.922 ms/op
When normilized by MD5 performance we have:
| Hasher | Score rel. MD5 |
|---|---|
| MD5 | 1.0 |
| SHA-1 | 1.35 |
| SHA-256 | 1.71 |
| SHA-512 | 2.65 |
| MURMUR3_128 | 0.65 |
| MURMUR3_32 | 0.66 |
| SIP_HASH24 | 0.72 |
| CRC32 | 0.58 |
| CRC32C | 0.40 |
| String.hashCode | 0.19 |
Collisions do occur in some cases, running:
mvn exec:java -Dexec.mainClass="org.benchmark.CollisionDetector" -Dexec.args="500000 12" -Dexec.classpathScope=runtime
results in:
Generated corpus of size 500000 with strings of length 12
# Collisions for SHA-512 : 0
# Collisions for SHA-256 : 0
# Collisions for CRC32C : 43
# Collisions for MURMUR3_128 : 0
# Collisions for MD5 : 0
# Collisions for CRC32 : 31
# Collisions for MURMUR3_32 : 21
# Collisions for SIP_HASH24 : 0
# Collisions for SHA-1 : 0
# Collisions for String.hashCode : 19