Serialization Benchmarks

Nalix features a custom binary serialization engine, LiteSerializer, designed for maximum throughput and minimal allocation in high-performance network transport.

LiteSerializer Performance#

Detailed micro-benchmarks of standard serialization operations, including unmanaged structs, custom formatters, inline span fills, and formatter resolving.

Method	Mean	Error	StdDev	P95	Gen0	Gen1	Gen2	Allocated
Serialize_Unmanaged_Small	8.4605 ns	0.3905 ns	0.5214 ns	9.1123 ns	0.0015	-	-	56 B
Serialize_Unmanaged_Large	49.7401 ns	1.3714 ns	1.7831 ns	52.4651 ns	0.0144	-	-	536 B
Deserialize_Unmanaged_Small	3.3755 ns	0.0330 ns	0.0441 ns	3.4550 ns	-	-	-	0 B
Deserialize_Unmanaged_Large	22.0984 ns	0.2677 ns	0.3083 ns	22.6873 ns	-	-	-	0 B
Serialize_Formatter	58.3652 ns	2.3481 ns	2.6099 ns	60.5600 ns	0.0041	-	-	152 B
Deserialize_Formatter	60.4312 ns	0.7158 ns	0.9556 ns	62.0313 ns	0.0079	-	-	296 B
Fill_IntoSpan_Small	1.0629 ns	0.0188 ns	0.0245 ns	1.1034 ns	-	-	-	0 B
Fill_IntoSpan_Large	5.1200 ns	0.0450 ns	0.0585 ns	5.2316 ns	-	-	-	0 B
Resolve_Formatter	0.9009 ns	0.0128 ns	0.0162 ns	0.9219 ns	-	-	-	0 B

Why Nalix Serialization?#

• Zero-Allocation Deserialization: Deserialization of unmanaged structures (Deserialize_Unmanaged_Small and Deserialize_Unmanaged_Large) is fully allocation-free (0 B) and runs in single-digit to low double-digit nanoseconds.
• Fast Path Span Fills: Copying primitive data directly into target spans (Fill_IntoSpan_Small) takes approximately 1 nanosecond (~1.06 ns) with zero overhead.
• Direct Memory Blitting: For unmanaged structs, LiteSerializer uses low-level memory block copying via the Unsafe class, producing CPU instructions that operate at raw hardware limits.
• Aggressive Inlining: Key methods in the serialization path are decorated with aggressive compilation attributes, allowing RyuJIT to inline and optimize out method call overhead.