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.TryGet("answer", out int value); //value is "42"
//factory pattern! calls the expensive factory only if not cached yet
cache.GetOrAdd(
key: "answer",
valueFactory: k => 42,
ttl: TimeSpan.FromMilliseconds(100));
//handy overload to prevent captures/closures allocation
cache.GetOrAdd(
key: "answer",
valueFactory: (k, arg) => 42 + arg.Length,
ttl: TimeSpan.FromMilliseconds(100),
factoryArgument: "some state data");
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The above is no longer valid, we have switched to .NET 6 targeting and now use TickCount64 which is free of this problem.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.
