Skip to content

PHASE_2_RESULTS.md

Latest commit

 

History

History
58 lines (41 loc) · 2.83 KB

PHASE_2_RESULTS.md

File metadata and controls

58 lines (41 loc) · 2.83 KB
Raw

Phase 2 Baseline Results

Validation run: 2026-07-03

The functional validation of the Phase 2 TCP over TLS relay was executed locally on the loopback interface, confirming that the ingress and egress agents can successfully proxy and transparently relay data without truncation. Ingress validates the egress certificate; mutual authentication is deferred to a later security milestone.

The operator orchestrated the agents and the benchmark tool through a local shell script (scripts/test-phase2.sh):

./scripts/make-dev-certs.sh
./build/accelerator-bench server --listen 127.0.0.1 --port 5201 --runs 1 &
./build/accelerator-egress --config config/egress.example.toml &
./build/accelerator-ingress --config config/ingress.example.toml &
./build/accelerator-bench client --host 127.0.0.1 --port 1080 --runs 1 --bytes 1048576

This validation transferred a deterministic 1 MiB payload over the unlabelled profile across the loopback network. The receiver correctly matched the FNV-1a fingerprint hash on the payload and the 24-byte response.

Profile Runs Payload Size Goodput Duration Verified
unlabelled (Loopback) 1 1048576 Bytes (1 MiB) ~139.3 Mbit/s 0.060 s true

The results show standard local-network performance characteristics bounded by the added TLS encryption overhead, OpenSSL record fragmentation, and context switching across the three independent user-space processes (Client -> Ingress -> Egress -> Server) communicating via the loopback interface.

Environment

  • Kernel: Linux 7.0.0-22-generic x86_64
  • Compiler: GCC 15.2.0
  • TLS Stack: OpenSSL 3.5.5 in the recorded environment
  • Local Proxy Ports: Ingress (1080), Egress (4433), Target (5201)

Interpretation limits

  • These numbers reflect local loopback validation runs ensuring cryptographic verification and proper proxy teardown without packet truncation.
  • Full namespace impairment tests utilizing scripts/wan-lab.sh with terrestrial-wan and other lab profiles require privileged execution (sudo) and were deferred.
  • The TCP plain-relay architecture (Phase 2) is designed for architectural correctness rather than performance scaling. The primary goal was to ensure the tunnel framing and proxy lifecycle were flawless before introducing QUIC multiplexing in Phase 3.

Validation addendum: 2026-08-18

After correcting the mixed Phase 2/Phase 3 working tree, the Phase 2 loopback smoke test was rerun with freshly generated local development certificates:

./scripts/test-phase2.sh

Observed benchmark result:

Profile Runs Payload Size Goodput Verified
unlabelled (Loopback) 1 1048576 Bytes (1 MiB) 162.323251 Mbit/s true

This validates that the active build once again represents the Phase 2 framed TCP/TLS relay rather than the parked Phase 3 QUIC experiment.