Skip to content

PLAN.md

Latest commit

 

History

History
133 lines (89 loc) · 7.5 KB

PLAN.md

File metadata and controls

133 lines (89 loc) · 7.5 KB
Raw

RXNM 2.0: The Monolithic Convergent Evolution

Status: Future Planning (Post-v1.0 RC3)

Target: Unified Network Runtime for Extremis (MCU), Handhelds, and Cloud.

Core Philosophy: Convergence of the "Micro" and "Full Fat" stacks into a single, power-aware high-performance C engine with eBPF/XDP at the center.

1. Vision: From "Hybrid" to "Monolithic"

RXNM 1.0 established the Hybrid Path (Bash logic + C acceleration + systemd-networkd). RXNM 2.0 moves to a Converged Engine where the C Agent becomes the primary logic engine. For handhelds and extremis environments, this eliminates the overhead of the systemd-networkd daemon and its associated XML/INI parsing logic in favor of a monolithic state machine.

Evolution Comparison

Metric 1.1 (Hybrid) OpenWrt (netifd) 2.0 (Converged Mono)
Logic Engine Bash / systemd-networkd C / ubus / Shell Scripts Monolithic C Engine
Connectivity iwd (D-Bus) hostapd / wpa_s Internalized ell/iwd Logic
Data Plane Kernel IP Stack Kernel IP Stack / Bridge eBPF / XDP (Primary)
IPC D-Bus (System/Lite) ubus (libubox) Zero-IPC (Internal)
Service Logic Native unshare/setns (Agent) Flat Router Namespace Native setns / BPF Maps
CPU Wakeups (Idle) ~20-40 / sec ~10-20 / sec < 2 / sec
Resident RAM ~7.7 MB ~6.5MB ~2.5MB (Unified)

2. Core Architecture Pillars

A. The "Cannibalized" Engine (Zero-IPC Connectivity)

2.0 integrates core connectivity components directly into the agent memory space.

  • Power Benefit: Eliminates the D-Bus/ubus daemon requirement. By removing the context-switching between connectivity daemons and the manager, the CPU stays in deep sleep (C-state) longer.
  • Unified State: Authentication and L3 addressing happen in the same process memory space, enabling atomic, instant transitions from "Resume" to "Connected."

B. eBPF/XDP: The Power-Aware Data Plane

2.0 uses eBPF maps as the primary source of truth for routing and firewalling.

  • Interrupt Coalescing: XDP allows packets to be processed at the driver level. Inter-service (SOA) traffic never traverses the kernel's heavy IP stack, significantly reducing CPU interrupts.
  • Comparison with netifd: While netifd handles the control plane efficiently, its data plane is standard Linux bridging/routing. RXNM 2.0 uses XDP-Redirect to shunt packets between namespaces with sub-microsecond latency.

C. Reactive Power Management

The 2.0 engine is 100% event-driven, blocking on a single Netlink socket for kernel events.

  • Hardware Filter Offloading: On supported handheld NICs, the engine configures eBPF filters to drop background ARP/MDNS chatter in hardware, ensuring the SoC stays in deep C-states (C10+) longer.

3. Compile-Time Harvesting & Build Complexity

To minimize reinventing the wheel while maintaining an "Extremis" footprint, RXNM 2.0 utilizes a Harvesting Build Pipeline. Instead of linking against heavy external libraries, the build system surgically extracts source files from upstream projects.

A. Upstream Source Harvesting

Component Harvest Target Purpose
iwd src/station.c, src/network.c, src/wsc.c PSK State Machines & Scanning
ell ell/main.c, ell/genl.c, ell/tls.c Event Loop, Netlink, and Crypto Primitives
systemd src/shared/conf-parser.c Standard .network file compatibility

B. Automated Logic Surgery

The build system (Make/Meson) performs automated preprocessing on harvested code:

  • D-Bus Excision: Uses sed and preprocessor macros to strip all dbus_ function calls and object-manager logic from the harvested iwd source.
  • Feature Pruning: Removes Enterprise (EAP) and SIM-card logic from ell/tls, reducing the static binary size by ~40%.
  • Symbol Namespacing: Wraps harvested logic in rxnm_ namespaces to prevent collisions while allowing us to track upstream bug fixes easily.

4. Performance & Efficiency Matrix

Comparison of the Monolithic 2.0 stack against standard systemd-networkd and OpenWrt netifd.

Metric systemd-networkd OpenWrt (netifd) Micro-RXNM 2.0 (Mono)
Resident RAM ~7.7 MB (Total Stack) ~6.5 MB ~2.5 MB
Binary Footprint ~5.2 MB ~1.2 MB ~0.9 MB
Cold Start Latency ~450ms ~250ms ~15ms
USB/TB Hotplug ~180ms ~80ms < 5ms
Idle Wakeups ~25/sec ~12/sec < 2/sec
Throughput (PPS) Kernel Limited Kernel Limited Line Rate (XDP)

5. Use Case Scenarios: The ROI of Zero-IPC

A. Handhelds (Anbernic, Powkiddy, Retroid, Ayaneo)

  • Problem: RAM is contested between network daemons and emulators.
  • RXNM Solution: Recovers 4MB - 45MB of Resident RAM vs. traditional stacks.
  • Power: Extended standby by eliminating ubus/dbus polling interrupts.

B. Embedded Router (MIPS/RISC-V MCU)

  • Comparison with OpenWrt: netifd is the standard here, but RXNM 2.0's SOA approach allows for hardware-isolated namespaces (e.g., WAN vs. LAN) on the same 16MB Flash board.
  • Benefit: XDP offloading allows a 64MB RAM MCU to route gigabit traffic without pegging the CPU.

C. Cloud & Container Edge

  • CNI Replacement: RXNM 2.0 acts as a high-density networking runtime.
  • Efficiency: Supports up to 20x more isolated services on the same hardware compared to standard container networking due to the monolithic management plane.

6. Implementation Roadmap

Phase 1: Harvesting Infrastructure

  • Create scripts/harvest-upstream.sh to pull specific source trees.
  • Establish the "Surgery" patchset to strip D-Bus/Glib from harvested files.
  • Implement the raw bpf() loader in the Agent.

Phase 2: Integrated Logic Convergence

  • Integrate harvested iwd station logic into the Agent's event loop.
  • Wakeup Audit: Optimize the main loop to ensure zero wakeups when idle.
  • Demonstrate Zero-IPC WiFi connection (no external daemon).

Phase 3: XDP-Native SOA & Data Plane

  • Implement the inter-namespace fast-path using XDP-Redirect.
  • Migrate "Nullify Mode" and "Firewall" logic to driver-level XDP programs.

Phase 4: Release & Extremis Validation

  • Update the rxnm dispatcher to detect hardware capabilities and launch the 2.0 runtime.
  • Final validation on 16MB SPI Flash and 64MB RAM targets.

7. Functionality Matrix: The 2.0 Standard

Feature RXNM 1.1 (Hybrid) OpenWrt (netifd) RXNM 2.0 (Mono)
Logic Engine systemd-networkd C + Shell Scripts Internal C Logic
WiFi Auth External iwd External wpa_s Internal Module
Firewall iptables/nft fw4 (nftables) eBPF (Stateless)
IPC Bus D-Bus ubus NONE (Monolithic)
Power Mgmt Passive Passive Proactive

8. Summary

RXNM 2.0 is the evolution from a "Manager" to a "Network Runtime." By cannibalizing the best-in-class logic from iwd and merging it with XDP hardware-acceleration, we create a stack that is invisible to the user but carrier-grade in performance.

While OpenWrt's netifd is significantly leaner than systemd-networkd, it still relies on an IPC bus (ubus) and external shell scripts for L3 configuration. RXNM 2.0 eliminates these remaining overheads, providing a single-binary networking solution that scales from 64MB MCUs to 128-core x86_64 cloud hosts.