Memory & Storage Benchmarks

Nalix uses a highly optimized memory management subsystem designed to eliminate Garbage Collection (GC) pauses in high-throughput workloads.

Buffer Pooling & Leases#

Comparison of memory acquisition strategies across various sizes (64 B, 1 KB, and 16 KB).

Buffer Allocation Metrics (Size = 64)#

Method	Mean	Error	StdDev	P95	Ratio	Allocated	Alloc Ratio
RawAllocation	16.974 ns	1.3931 ns	1.6043 ns	19.503 ns	1.01	88 B	1.00
ArrayPool_Shared	10.824 ns	0.3976 ns	0.4579 ns	11.244 ns	0.64	0 B	0.00
BufferPoolManager_RentReturn	23.905 ns	0.6627 ns	0.7632 ns	24.545 ns	1.42	0 B	0.00
BufferLease_RentDispose	54.052 ns	0.5018 ns	0.5778 ns	54.617 ns	3.21	0 B	0.00

Buffer Allocation Metrics (Size = 1024)#

Method	Mean	Error	StdDev	P95	Ratio	Allocated	Alloc Ratio
RawAllocation	159.487 ns	29.9061 ns	34.4399 ns	194.306 ns	1.06	1048 B	1.00
ArrayPool_Shared	6.406 ns	0.0902 ns	0.1039 ns	6.556 ns	0.04	0 B	0.00
BufferPoolManager_RentReturn	24.173 ns	0.4341 ns	0.4999 ns	24.471 ns	0.16	0 B	0.00
BufferLease_RentDispose	62.729 ns	1.1419 ns	1.3150 ns	63.688 ns	0.42	0 B	0.00

Buffer Allocation Metrics (Size = 16384)#

Method	Mean	Error	StdDev	P95	Ratio	Allocated	Alloc Ratio
RawAllocation	2,608.051 ns	452.9858 ns	521.6591 ns	3,035.733 ns	1.062	16408 B	1.00
ArrayPool_Shared	10.496 ns	0.8494 ns	0.9782 ns	11.159 ns	0.004	0 B	0.00
BufferPoolManager_RentReturn	23.769 ns	0.6516 ns	0.7504 ns	24.451 ns	0.010	0 B	0.00
BufferLease_RentDispose	53.938 ns	1.3191 ns	1.5190 ns	54.873 ns	0.022	0 B	0.00

Why Nalix Memory?#

• Pooled Array Rental (BufferPoolManager): BufferPoolManager wraps ArrayPool<byte>.Shared with atomic rent/return counters and optional leak-detection diagnostics (BufferOptions.EnableBufferLeakDetection), delivering 0 B allocation across all payload sizes.
• Reference-Counted Lease Safety (BufferLease): BufferLease provides a pooled, reference-counted (Retain() / Dispose()) rental wrapper (IBufferLease) that keeps rent/dispose cycles allocation-free (0 B) while preventing premature buffer returns across async boundaries.

---

Object Pooling#

Memory metrics for reusing class instances via object pools.

Method	Mean	Error	StdDev	P95	Ratio	Allocated	Alloc Ratio
RawAllocation	8.016 ns	0.4384 ns	0.5049 ns	8.474 ns	1.00	32 B	1.00
RentAndReturn_ObjectPool	157.624 ns	5.2398 ns	6.0342 ns	161.363 ns	19.74	0 B	0.00

Behind the design#

• Object Reusability: Reusing instances of complex structures (like session states, packet envelopes, and processing contexts) prevents Gen 0 GC thrashing.
• Hybrid Fast-Path Architecture:
  • Thread-Local Cache: First-level lock-free lookup caches object instances in thread-local storage (ThreadLocalCache<T>) for ultra-low latency reuse on the same thread.
  • Type-Indexed Buckets: When thread-local slots are empty or full, retrieval falls back to an array-backed lookup using compiled type IDs (PoolType<T>.Id) to bypass slow hashing or generic dictionary lookups.
• Reset Logic: To prevent state pollution across rentals, pooled objects implement a reset interface (IPoolable) that automatically wipes data and prepares the instance for reuse upon its return to the pool.