Faster.Collections.Pooled

A high-performance collection library for .NET 10+ that provides pooled alternatives to standard collections, reducing GC pressure and improving throughput.

Features

• Array Pooling - All collections leverage shared array pools to minimize allocations
• High-Performance Hashing - Includes optimized hash algorithms (FastHash, WyHash, xxHash3, CRC32) powered by Faster.Map.
• Drop-in Replacements - Compatible APIs for Dictionary, List, Queue, Stack, and PriorityQueue.
• Comprehensive Testing - Includes comparison tests against BCL collections.
• Benchmarking - Includes a dedicated benchmarking project for performance evaluation.

API Compatibility

| Collection | API Source | |-----------|------------| | PooledList | List | | PooledQueue | Queue | | PooledStack | Stack | | PooledPriorityQueue | PriorityQueue | | PooledDictionary | FasterMap.BlitzMap + Interfaces* of Dictionary |

[!IMPORTANT]
* PooledDictionary implements standard interfaces of Dictionary except ISerializable and IDeserializationCallback

Special API

ToPooledCollection Series

using PooledList<int> list = [1, 2, 3, 4, 5];
list.ToPooledQueue();          // Convert to PooledQueue
list.ToPooledStack();          // Convert to PooledStack

Deque

using PooledDeque<int> deque = [];
deque.EnququeHead(1);
deque.EnququeTail(2);

Construction based on Collection Expression

using PooledDeque<int> deque = [0];
using PooledDictionary<int, int> dict = [ 
    (1, 1), 
    (2, 2),
    (3, 3), 
];

Installation

NuGet Package

dotnet add package Faster.Collections.Pooled

Build from Source

dotnet fsi build.fsx build

Performance

[!NOTE]
Sometimes, benchmark-dotnet or GC will allocate small amount of memory.
For exmaple, the benchmark below shows System.Collections.Generic.PriorityQueue.Dequeue allocates 0 B, but it may allocate a few bytes in some runs.

| Method | N | Mean | Error | StdDev | Median | Ratio | RatioSD | Allocated | Alloc Ratio | |--------------------------- |------- |---------------:|-------------:|-------------:|---------------:|------:|--------:|----------:|------------:| | OriginPriorityQueueDequeue | 100 | 2,741.8 ns | 168.45 ns | 472.36 ns | 2,700.0 ns | 1.03 | 0.24 | 336 B | 1.00 | | PooledPriorityQueueDequeue | 100 | 2,847.4 ns | 158.32 ns | 454.26 ns | 2,800.0 ns | 1.07 | 0.24 | - | 0.00 | | | | | | | | | | | | | OriginPriorityQueueDequeue | 1000 | 41,531.0 ns | 826.78 ns | 1,211.88 ns | 41,400.0 ns | 1.00 | 0.04 | - | NA | | PooledPriorityQueueDequeue | 1000 | 42,242.9 ns | 783.69 ns | 694.72 ns | 42,250.0 ns | 1.02 | 0.03 | - | NA |

PooledList vs List

Add

Speed up 85%, no allocations.

Remove

Almost same speed, no allocations.

PooledDictionary vs Dictionary

Speed up 298%, no allocations.

Speed up 58%, no allocations.

PooledQueue vs Queue

Enqueue

Speed up 65%, no allocations.

Dequeue

Almost same speed, no allocations.

PooledStack vs Stack

Push

Speed up 86%, no allocations.

Pop

Almost same speed, no allocations.

PooledPriorityQueue vs PriorityQueue

Enqueue

Speed up 71%, no allocations.

Dequeue

Almost same speed, no allocations.

Roadmap

• [ ] PooledHashset (based on Faster.Map.BlitzSet)
• [ ] ISerializable and IDeserializationCallback support for PooledDictionary
• [ ] Try using IMemoryOwner<T> ownership model compatible with existing types

Acknowledgments

[!IMPORTANT]
Very special thanks to the following projects:

• Faster.Map - High-performance hashing algorithms
• Collections.Pooled - Pooled collection implementations

License

MIT License