ARM64 Linux kernel implementation using the shared Rust core, platform-specific carrier adapter and Gigabit Ethernet.
Loading benchmark description…
Higher is better
STCP/TCP exceeded TLS throughput while using substantially less client CPU.
| Mode | Combined MiB/s | Ops/s | RTT p50 ms | RTT p95 ms | RTT p99 ms | Connect ms | CPU | Errors |
|---|
Each matching test shape is compared automatically across TCP, TLS, STCP/TCP and STCP/UDP. Positive throughput and operations values mean STCP is faster; positive reduction values mean STCP uses less CPU or latency.
The benchmark captures Raspberry Pi server counters from /proc/interrupts, /proc/softirqs and /proc/stat before and after every test. Results are normalized per 1,000 successful operations.
The relative index uses measured network IRQ + NET_RX + NET_TX events per 1,000 operations. TLS is normalized to 100 whenever a matching TLS result exists.
The site distinguishes computational efficiency from actual electrical consumption. Real power comparisons appear only when measured watt or joule data is present.
Combined throughput and operations per second by payload, client count and pipeline depth.
Median, p95 and p99 RTT. Tail latency matters for control systems and interactive device traffic.
Mean connection-establishment time, especially important for reconnecting IoT devices.
Throughput and operations normalized by client CPU percentage.
Error counts, timeout rates and successful-operation ratios.
Performance change from one client to high-concurrency tests and deeper pipelines.
STCP/TCP versus STCP/UDP under identical test shapes.
Measured watts, joules per MiB and joules per 1,000 operations when instrumentation data is available.
Suitable for compact device APIs and constrained-resource access through STCP carriers.
CoAP section →Benchmark and application traffic use the standard socket-oriented STCP interface.