Hyperion Benchmarks

We benchmarked Hyperion XDP against a baseline Linux network stack using iperf3 for throughput and icmp for latency.

Results: Throughput (Gbps)

Hyperion maintains wire-speed performance even with Full DPI enabled.

Configuration	Throughput (Gbps)	Degradation
Baseline (No XDP)	64.32 Gbps	0%
Header Filtering	63.42 Gbps	-1.4%
Full DPI (Signatures)	65.28 Gbps*	+1.5%

> Note: The slight increase in DPI mode is likely due to measurement variance or XDP’s early-drop efficiency clearing RX queues faster for valid traffic.

Results: Latency (Microseconds)

The cost of Deep Packet Inspection (DPI) is measured in microseconds.

Metric	Baseline	Header Filter	Full DPI
Mean Latency	99 µs	100 µs	100 µs
P99 Latency	178 µs	177 µs	143 µs

Latency Distribution

xychart-beta
    title "P99 Latency Comparison (Lower is Better)"
    x-axis ["Baseline", "Header Filter", "Full DPI"]
    y-axis "Latency (µs)" 0 --> 200
    bar [178, 177, 143]

CPU Utilization

XDP shifts processing from “User CPU” to “System CPU” (SoftIRQ context).

Configuration	User CPU	System CPU	Total Load
Baseline	1.49%	30.65%	32.14%
Full DPI	1.25%	22.07%	23.32%

Efficiency Verdict

Hyperion XDP is more CPU efficient (~23% load) than the standard Linux stack (~32% load) because it drops malicious packets earlier in the pipeline, saving the kernel from the expensive TCP/IP stack traversing.