Performance Benchmarks

Reference performance measurements for HDDS components.

Test Environment

Component	Specification
CPU	AMD EPYC 7763 (64 cores)
Memory	256 GB DDR4-3200
Network	25 Gbps Ethernet
OS	Linux 6.1 (Debian 12)
Rust	1.75+ (release build)

Latency Benchmarks

End-to-End Latency (Same Host)

Payload Size	p50	p95	p99
64 bytes	8 us	15 us	25 us
256 bytes	10 us	18 us	30 us
1 KB	15 us	25 us	45 us
4 KB	25 us	45 us	80 us
64 KB	120 us	200 us	350 us

Shared Memory Transport

Payload Size	p50	p95	p99
64 bytes	1.2 us	2.5 us	4 us
256 bytes	1.5 us	3 us	5 us
1 KB	2 us	4 us	7 us
4 KB	4 us	8 us	15 us

UDP Unicast (Same LAN)

Payload Size	p50	p95	p99
64 bytes	45 us	80 us	150 us
256 bytes	50 us	90 us	180 us
1 KB	60 us	110 us	220 us
4 KB	90 us	160 us	300 us

Throughput Benchmarks

Message Rate (BestEffort)

Payload Size	Throughput (msg/s)	Bandwidth
64 bytes	2.5 M	160 MB/s
256 bytes	1.8 M	460 MB/s
1 KB	850 K	850 MB/s
4 KB	280 K	1.1 GB/s
64 KB	35 K	2.2 GB/s

Message Rate (Reliable)

Payload Size	Throughput (msg/s)	Bandwidth
64 bytes	1.2 M	77 MB/s
256 bytes	900 K	230 MB/s
1 KB	450 K	450 MB/s
4 KB	150 K	600 MB/s

Sustained Throughput

RingBuffer ingestion: 10.4 M msg/s (256 bytes)
Capture file write:   481 MB/s (sequential)
Capture file read:    5.3 M frames/s

Component Benchmarks

Serialization (CDR2)

Operation	64 B	256 B	1 KB	4 KB
Serialize	50 ns	120 ns	400 ns	1.5 us
Deserialize	40 ns	100 ns	350 ns	1.3 us

History Cache

Operation	Time
Insert sample	150 ns
Lookup by key	80 ns
Remove sample	120 ns
Match ACK	200 ns

Discovery

Operation	Time
SPDP parse	5 us
SEDP parse	8 us
Endpoint match	2 us
Cold start (2 participants)	150-300 ms

Memory Usage

Per-Entity Overhead

Entity	Memory
DomainParticipant	~2 MB
Publisher	~64 KB
Subscriber	~64 KB
DataWriter	~128 KB + history
DataReader	~128 KB + history
Topic	~16 KB

History Cache Memory

Memory = instances x history_depth x sample_size + overhead

Example (100 sensors, 10-deep history, 1KB samples):
= 100 x 10 x 1024 + 100 x 256
= 1.02 MB + 25 KB = ~1.05 MB

Running Benchmarks

Built-in Benchmarks

# Run all benchmarks
cargo bench --package hdds

# Latency benchmark
cargo bench --package hdds -- latency

# Throughput benchmark
cargo bench --package hdds -- throughput

# Serialization benchmark
cargo bench --package hdds -- serialize

Performance Testing Tool

# Publisher side
hdds-perf pub --topic Benchmark --rate 10000 --size 256

# Subscriber side
hdds-perf sub --topic Benchmark --stats

Output:

Received 100000 samples in 10.02s
  Throughput: 9980 msg/s (2.55 MB/s)
  Latency p50: 45 us, p95: 82 us, p99: 156 us
  Jitter: 12 us (std dev)

Profiling

CPU Profiling

# Using perf
perf record -g cargo run --release --example benchmark
perf report

# Using flamegraph
cargo flamegraph --example benchmark

Memory Profiling

# Using heaptrack
heaptrack cargo run --release --example benchmark
heaptrack_gui heaptrack.benchmark.*.gz

Performance Targets

Real-Time Targets

Metric	Target	Notes
Latency p99	< 100 us	Same host, < 1 KB
Jitter	< 20 us	Standard deviation
Message rate	> 100 K/s	Per writer

High-Throughput Targets

Metric	Target	Notes
Bandwidth	> 1 GB/s	Large payloads
Message rate	> 1 M/s	Small payloads
CPU efficiency	< 5% per 100K msg/s

Comparison with Other Implementations

Metric	HDDS	FastDDS	CycloneDDS
Latency (64B)	8 us	12 us	10 us
Throughput	2.5 M/s	1.8 M/s	2.2 M/s
Memory (idle)	2 MB	4 MB	3 MB

Measurements on equivalent hardware and configuration.

Optimization Impact

QoS Impact on Performance

Configuration	Latency	Throughput
BestEffort + KeepLast(1)	Baseline	Baseline
Reliable + KeepLast(10)	+20%	-30%
Reliable + KeepAll	+50%	-50%
+ TransientLocal	+10%	-10%

Transport Impact

Transport	Latency	Throughput
Shared Memory	1x	1x
UDP Loopback	4x	0.7x
UDP Network	10x	0.5x

Next Steps

• Latency Tuning - Minimize latency
• Throughput Tuning - Maximize bandwidth