The leapfrog crate contains two hash map implementations, HashMap, which is a single-threaded HashMap, and LeapMap, which is fast, concurrent version of the HashMap, where all operations can be performed concurrently from any number of threads. If the key and value types support atomic operations then the map is lock-free, and when there is no atomic support for either the key or value, then operations within the map on that non-atomic type use an efficient spinlock and the locking is very fine-grained.
The performance of both maps is good on the benchmarks it's been tested on. The HashMap is between 1.2 and 1.5x the std::collections::HashMap. Compared to RwLock<HashMap>, on 16 cores the LeapMap is around 13.5x faster. It also scales very well up to high core counts, for multiple workloads. Benchmark results can be found at rust hashmap benchmarks. These bechmarks, however, are limited in the use cases that they cover, and should be extended to cover a much wider scope. Nevertheless, for those bechmarks, the LeapMap is the fastest map.
For more details on the API and the differences compared to std::collections::HashMap, please see the crate documentation.
The implementations of these maps have recently been updated to store keys, so the API for both is much closer to std::collections::HashMap. There is now also support for serde with the optional "serde" feature. The maps also now have iterator support. The default hasher has been changed to be the DOS resistant hasher from the standard library.
The "stable_alloc" feature flag has been added to allow the crate to be used with the stable toolchain.
These is not yet support for rayon, and there are not yet Set versions of the Maps.
A snapshot of the results for the LeapMap on a read heavy workload using a 16 core AMD 3950x CPU are the following (with throughput in Mops/second, cores in brackets, and performance realtive to std::collections::HashMap with RwLock):
| Map | Throughput (1) | Relative (1) | Throughput (16) | Relative (16) |
|---|---|---|---|---|
| RwLock + HashMap | 19.4 | 1.0 | 11.7 | 0.60 |
| Flurry | 11.2 | 0.58 | 80.8 | 4.16 |
| DashMap | 14.1 | 0.72 | 87.5 | 4.51 |
| LeapMap | 17.8 | 0.92 | 148.0 | 7.62 |
On a 12-core M2 Max the results are the following (again values are all relative to std::collections::HashMap):
| Map | Throughput (1) | Relative (1) | Throughput (12) | Relative (12) |
|---|---|---|---|---|
| RwLock + HashMap | 21.2 | 1.0 | 3.4 | 0.16 |
| Flurry | 8.0 | 0.37 | 64.5 | 3.04 |
| DashMap | 10.4 | 0.49 | 42.2 | 2.00 |
| Evmap | 13.5 | 0.64 | 18.7 | 0.88 |
| LeapMap | 13.9 | 0.66 | 106.1 | 5.00 |
On benchmarks of other use cases, leapfrog's LeapMap is faster by even higher factors.
For the single threaded leapfrog HashMap, a benchmark of random inserts and deletes (a port of this benchmark of C++ hashmaps) has the rust std HashMap at around 40Mops/s and the leapfrog HashMap at around 50Mops/s, with default hasher from the standard library, which are competitive with the C++ hash maps from the linked benchmark.
Both maps use the same leapfrog probing strategy, for which you can find more information on Jeff Preshing's leapfrog probing blog post. A c++ implementation of the map can be found at his Junction library, which was the starting point for this implementation. Those maps don't have support for keys and therefore collisions, which these maps do. The probing strategy requires 8 bytes per key-value pair and and additional 8 bytes are stored for the hashed key so that it doesn't have to be rehashed for comparisons.
There is optional support for serde, via the "serde" feature.
This crate requires the allocator_api feature, which is only available on nightly. To enable use of the crate with the stable toolchain, the "stable_alloc" feature has been added.
If/when the allocator_api feature is no longer experimental, this feature flag will be removed.
This crate is licensed under either the Apache License, Version 2.0, see here or here or the MIT license see here or here, at your chosing. It can be used freely for both commercial and non-commercial applications, so long as you publish the contents of your chosen license file somewhere in your documentation.