Data Processing & Framing Benchmarks

Detailed performance metrics for the Nalix data processing and transformation pipelines, including LZ4 compression and framing transforms.

Frame Processing Pipeline#

The frame pipeline handles end-to-end outbound serialization (compression + encryption) and inbound deserialization (decryption + decompression) processes.

Pipeline Performance (Payload Size = 64)#

Method	PayloadSize	Mean	Error	StdDev	P95	Gen0	Allocated
ProcessOutbound_CompressOnly	64	172.0 ns	1.21 ns	1.40 ns	173.7 ns	-	0 B
ProcessOutbound_EncryptOnly	64	571.1 ns	3.56 ns	3.96 ns	575.9 ns	0.0048	208 B
ProcessOutbound_Full	64	698.7 ns	4.84 ns	5.58 ns	707.0 ns	0.0048	208 B
ProcessInbound_DecompressOnly	64	172.0 ns	1.84 ns	2.12 ns	174.9 ns	-	0 B
ProcessInbound_DecryptOnly	64	1,160.3 ns	8.00 ns	9.21 ns	1,173.9 ns	0.0095	416 B
ProcessInbound_Full	64	1,287.7 ns	8.90 ns	9.53 ns	1,300.5 ns	0.0134	544 B

Pipeline Performance (Payload Size = 512)#

Method	PayloadSize	Mean	Error	StdDev	P95	Gen0	Allocated
ProcessOutbound_CompressOnly	512	441.0 ns	3.56 ns	4.09 ns	447.8 ns	-	0 B
ProcessOutbound_EncryptOnly	512	1,177.2 ns	7.53 ns	8.67 ns	1,192.2 ns	0.0286	1104 B
ProcessOutbound_Full	512	1,530.4 ns	9.24 ns	10.64 ns	1,544.5 ns	0.0286	1104 B
ProcessInbound_DecompressOnly	512	443.6 ns	4.05 ns	4.50 ns	450.5 ns	-	0 B
ProcessInbound_DecryptOnly	512	2,324.8 ns	17.66 ns	19.63 ns	2,352.5 ns	0.0572	2208 B
ProcessInbound_Full	512	2,858.6 ns	24.84 ns	28.60 ns	2,904.0 ns	0.0839	3232 B

Pipeline Performance (Payload Size = 4096)#

Method	PayloadSize	Mean	Error	StdDev	P95	Gen0	Gen1	Allocated
ProcessOutbound_CompressOnly	4096	1,182.0 ns	8.85 ns	10.19 ns	1,197.1 ns	-	-	0 B
ProcessOutbound_EncryptOnly	4096	6,172.2 ns	40.13 ns	44.60 ns	6,228.2 ns	0.2365	0.0076	8272 B
ProcessOutbound_Full	4096	7,304.3 ns	51.40 ns	57.13 ns	7,383.6 ns	0.2365	-	8272 B
ProcessInbound_DecompressOnly	4096	1,205.0 ns	7.17 ns	7.97 ns	1,215.2 ns	-	-	0 B
ProcessInbound_DecryptOnly	4096	12,472.2 ns	98.50 ns	113.43 ns	12,648.5 ns	0.4730	0.0153	16544 B
ProcessInbound_Full	4096	14,019.7 ns	72.64 ns	83.65 ns	14,148.0 ns	0.6866	0.0305	24736 B

Behind the design#

• Linear Scaling: Latency scales predictably relative to the payload size.
• Pipelined Execution: Outbound processing streams data directly, enabling sub-microsecond to low-microsecond framing for common payload sizes (e.g. 1.53 μs for a 512B payload).

---

Frame Transformations#

Individual pipeline transformers handle compression/decompression and encryption/decryption routines.

Transformer Performance (Payload Size = 64)#

Method	PayloadSize	Mean	Error	StdDev	P95	Allocated
Encrypt_AEAD_ChaCha20Poly1305	64	610.8 ns	3.37 ns	3.74 ns	616.3 ns	0 B
Decrypt_AEAD_ChaCha20Poly1305	64	1,164.5 ns	6.01 ns	6.92 ns	1,177.1 ns	0 B
Encrypt_AEAD_Salsa20Poly1305	64	560.2 ns	5.67 ns	6.30 ns	571.0 ns	0 B
Decrypt_AEAD_Salsa20Poly1305	64	1,062.4 ns	5.68 ns	6.55 ns	1,070.1 ns	0 B
Encrypt_Symmetric_ChaCha20	64	338.2 ns	1.51 ns	1.61 ns	340.9 ns	0 B
Decrypt_Symmetric_ChaCha20	64	578.4 ns	3.50 ns	3.74 ns	584.3 ns	0 B
Encrypt_Symmetric_Salsa20	64	317.7 ns	1.98 ns	2.20 ns	320.9 ns	0 B
Decrypt_Symmetric_Salsa20	64	537.4 ns	4.06 ns	4.51 ns	545.0 ns	0 B
Compress_LZ4	64	175.7 ns	0.80 ns	0.92 ns	176.9 ns	0 B
Decompress_LZ4	64	242.5 ns	2.03 ns	2.34 ns	245.3 ns	0 B

Transformer Performance (Payload Size = 1024)#

Method	PayloadSize	Mean	Error	StdDev	P95	Allocated
Encrypt_AEAD_ChaCha20Poly1305	1024	1,652.3 ns	9.13 ns	10.52 ns	1,665.8 ns	0 B
Decrypt_AEAD_ChaCha20Poly1305	1024	3,265.3 ns	17.39 ns	19.33 ns	3,295.6 ns	0 B
Encrypt_AEAD_Salsa20Poly1305	1024	2,760.1 ns	37.46 ns	38.47 ns	2,830.6 ns	0 B
Decrypt_AEAD_Salsa20Poly1305	1024	5,463.3 ns	51.18 ns	58.94 ns	5,553.4 ns	0 B
Encrypt_Symmetric_ChaCha20	1024	907.0 ns	1.72 ns	1.77 ns	909.5 ns	0 B
Decrypt_Symmetric_ChaCha20	1024	1,726.8 ns	4.48 ns	4.97 ns	1,735.4 ns	0 B
Encrypt_Symmetric_Salsa20	1024	2,020.7 ns	8.57 ns	9.52 ns	2,033.1 ns	0 B
Decrypt_Symmetric_Salsa20	1024	3,933.9 ns	22.79 ns	26.24 ns	3,980.6 ns	0 B
Compress_LZ4	1024	608.1 ns	4.36 ns	5.03 ns	614.5 ns	0 B
Decompress_LZ4	1024	674.4 ns	4.64 ns	5.15 ns	681.2 ns	0 B

Why Nalix Data Processing?#

• Zero-Allocation Transforms: Rather than creating intermediate garbage arrays, encryption and compression operate directly on rentals from BufferPoolManager using Span<byte>, consuming 0 B of heap allocation for transformer operations.
• Hardware-Friendly Speeds: Symmetric encryption of 1 KB of data executes in ~900 ns for ChaCha20 and ~2.0 μs for Salsa20.
• LZ4 Inline Compression: Built-in LZ4 compression integrates seamlessly with the buffer pipeline, performing a 1 KB compression in ~608 ns and 64 B compression in under 176 ns.