Jitbit.FastCache 1.0.4

FastCache

7x-10x faster alternative to MemoryCache. A high-performance, lighweight (8KB dll) and thread-safe memory cache for .NET.

TL;DR

Basically it's just a ConcurrentDictionary with expiration.

Benchmarks under Windows

Benchmarks under Linux (Ubuntu, Docker)

How is FastCache better

Compared to System.Runtime.Caching.MemoryCache and Microsoft.Extensions.Caching.MemoryCache FastCache is

• 7X faster reads (11X under Linux!)
• 10x faster writes
• Thread safe and atomic
• Generic (strongly typed keys and values) to avoid boxing/unboxing primitive types
• MemoryCache uses string keys only, so it allocates strings for keying
• MemoryCache comes with performance counters that can't be turned off
• MemoryCache uses heuristic and black magic to evict keys under memory pressure
• MemoryCache uses more memory, can crash during a key scan

Usage

Install via nuget

Install-Package Jitbit.FastCache

Then use

var cache = new FastCache<string, int>();

cache.AddOrUpdate(
	key: "answer",
	value: 42,
	ttl: TimeSpan.FromMinutes(1));

cache.TeyGet("answer", out int value); //value is "42"

cache.GetOrAdd(
	key: "answer",
	valueFactory: k => 42,
	ttl: TimeSpan.FromMilliseconds(100));

Tradeoffs

FastCache uses Environment.TickCount to monitor items' TTL. Environment.TickCount is 104x times faster than using DateTime.Now and 26x times faster than DateTime.UtcNow.

But Environment.TickCount is limited to Int32. Which means it resets to int.MinValue once overflowed. This is not a problem, we do have a workaround for that. However this means you cannot cache stuff for more than 25 days (2.4 billion milliseconds).

The above is no longer valid, we have switched to .NET 6 targeting and now use TickCount64 which is free of this problem.

Another tradeoff: MemoryCache watches memory usage, and evicts items once it senses memory pressure. FastCache does not do any of that it is up to you to keep your caches reasonably sized. After all, it's just a dictionary.