class Person {
String givenName, surName;
int salary;
}README
Table of Contents
Chronicle Software
About
This project provides a Java API for hashing any sequence of bytes in Java, including all kinds of primitive arrays, buffers, CharSequence and more.
Written for Java 8+ under Apache 2.0 licence.
The key difference compared to other similar projects, e.g.
Guava hashing, is that this has no object allocation during the hash computation and does not use ThreadLocal.
The implementation utilises native access where possible, but is also platform-endianness-agnostic. This provides consistent results whatever the byte order, while only moderately affecting performance.
Currently long-valued hash function interface is defined for 64-bit hash, and long[]-valued hash function interface for more than 64-bit hash, with the following implementations (in alphabetical order):
-
CityHash, version 1.1 (latest; 1.1.1 is a C++ language-specific maintenance release).
-
Two algorithms from FarmHash:
farmhashna(introduced in FarmHash 1.0) andfarmhashuo(introduced in FarmHash 1.1). -
MetroHash (using the metrohash64_2 initialization vector).
-
MurmurHash3 128-bit and low 64-bit.
-
wyHash, version 3.
-
xxh3, xxh128, 128-bit and 64 bit.
These are thoroughly tested with LTS JDKs 8, 11, 17, and 21, plus the latest non-LTS JDKs on both little- and big-endian platforms. Other non-LTS JDKs from 9 should also work, but they will not be tested from half a year after EOL.
Performance
Tested on Intel Core i7-4870HQ CPU @ 2.50GHz
| Algorithm | Speed, GB/s | Bootstrap, ns |
|---|---|---|
xxHash | 9.5 | 6 |
FarmHash | 9.0 | 6 |
FarmHash | 7.2 | 7 |
CityHash | 7.0 | 7 |
MurmurHash | 5.3 | 12 |
MetroHash | ||
WyHash |
To sum up,
When to use Zero-Allocation Hashing
-
You need to hash plain byte sequences, memory blocks or "flat" objects.
-
You want zero-allocation and good performance (at Java scale).
-
You need hashing to be agile with regards to byte ordering.
When not to use Zero-Allocation Hashing
-
You need to hash POJOs whose actual data is scattered in memory between managed objects. There is no simple way to hash these using this project, for example, classes such as:
-
You need to hash byte sequences of unknown length, for the simplest example,
Iterator<Byte>. -
You need to transform the byte sequence (e.g. encode or decode it with a specific coding), and hash the resulting byte sequence on the way without dumping it to memory.
Quick start
Gradle:
dependencies {
implementation 'net.openhft:zero-allocation-hashing:0.27ea1'
}Or Maven:
<dependency>
<groupId>net.openhft</groupId>
<artifactId>zero-allocation-hashing</artifactId>
<version>0.27ea1</version>
</dependency>In Java:
long hash = LongHashFunction.wy_3().hashChars("hello");See JavaDocs for more information.
Contributions are most welcome!
See the list of open issues.
Contributors
Similar repos
Languages
Java
99.9%
README
Table of Contents
Chronicle Software
About
This project provides a Java API for hashing any sequence of bytes in Java, including all kinds of primitive arrays, buffers, CharSequence and more.
Written for Java 8+ under Apache 2.0 licence.
The key difference compared to other similar projects, e.g.
Guava hashing, is that this has no object allocation during the hash computation and does not use ThreadLocal.
The implementation utilises native access where possible, but is also platform-endianness-agnostic. This provides consistent results whatever the byte order, while only moderately affecting performance.
Currently long-valued hash function interface is defined for 64-bit hash, and long[]-valued hash function interface for more than 64-bit hash, with the following implementations (in alphabetical order):
-
CityHash, version 1.1 (latest; 1.1.1 is a C++ language-specific maintenance release).
-
Two algorithms from FarmHash:
farmhashna(introduced in FarmHash 1.0) andfarmhashuo(introduced in FarmHash 1.1). -
MetroHash (using the metrohash64_2 initialization vector).
-
MurmurHash3 128-bit and low 64-bit.
-
wyHash, version 3.
-
xxh3, xxh128, 128-bit and 64 bit.
These are thoroughly tested with LTS JDKs 8, 11, 17, and 21, plus the latest non-LTS JDKs on both little- and big-endian platforms. Other non-LTS JDKs from 9 should also work, but they will not be tested from half a year after EOL.
Performance
Tested on Intel Core i7-4870HQ CPU @ 2.50GHz
| Algorithm | Speed, GB/s | Bootstrap, ns |
|---|---|---|
xxHash | 9.5 | 6 |
FarmHash | 9.0 | 6 |
FarmHash | 7.2 | 7 |
CityHash | 7.0 | 7 |
MurmurHash | 5.3 | 12 |
MetroHash | ||
WyHash |
To sum up,
When to use Zero-Allocation Hashing
-
You need to hash plain byte sequences, memory blocks or "flat" objects.
-
You want zero-allocation and good performance (at Java scale).
-
You need hashing to be agile with regards to byte ordering.
When not to use Zero-Allocation Hashing
-
You need to hash POJOs whose actual data is scattered in memory between managed objects. There is no simple way to hash these using this project, for example, classes such as:
class Person { String givenName, surName; int salary; } -
You need to hash byte sequences of unknown length, for the simplest example,
Iterator<Byte>. -
You need to transform the byte sequence (e.g. encode or decode it with a specific coding), and hash the resulting byte sequence on the way without dumping it to memory.
Quick start
Gradle:
dependencies {
implementation 'net.openhft:zero-allocation-hashing:0.27ea1'
}Or Maven:
<dependency>
<groupId>net.openhft</groupId>
<artifactId>zero-allocation-hashing</artifactId>
<version>0.27ea1</version>
</dependency>In Java:
long hash = LongHashFunction.wy_3().hashChars("hello");See JavaDocs for more information.
Contributions are most welcome!
See the list of open issues.
Similar repos
Contributors
Languages
Java
99.9%