High-throughput web services processing streams or JSON payloads often allocate thousands of temporary byte[] instances per second. Under high concurrency, these transient allocations hammer Gen0 and Gen1, triggering frequent garbage collection pauses.
Before: Constant Heap Allocations
byte[] buffer = new byte[4096]; int read = await stream.ReadAsync(buffer); ProcessData(buffer.AsSpan(0, read)); After: Rented Pooled Buffers
byte[] buffer = ArrayPool<byte>.Shared.Rent(4096); try { int read = await stream.ReadAsync(buffer); ProcessData(buffer.AsSpan(0, read)); } finally { ArrayPool<byte>.Shared.Return(buffer); } Why It Matters
- Zero Allocations:
ArrayPool<T>.Sharedchecks out existing buckets in memory, eliminating per-request allocations entirely. - LOH Avoidance: Renting buffers greater than 85,000 bytes prevents large objects from fragmenting the Large Object Heap.
- Actual Size vs Requested Size: Rented arrays round up to powers of two. Always pass the actual bytes read using
AsSpan(0, bytesRead)rather than readingbuffer.Length.
Watch out: Return(buffer) does not zero out contents by default. If your buffer handles tenant data or tokens, pass clearArray: true to avoid data leakage.
Do you use ArrayPool<T>.Shared or pipe streams directly via PipeReader?


