What it is and the problem it solves
Arista Networks is an American provider of client-to-cloud networking services for large data centre/AI, campus and routing environments. It solves the brittleness of legacy network OSes by decoupling state from process execution.
How it works
Arista EOS runs on an unmodified Linux kernel. It uses a multi-process state-sharing architecture. State information is separated from operational processes. Protocol processing, security functions, and hardware drivers run as independent agents in user address space. SysDB, an in-memory database, governs system state via publish/subscribe. Agents publish state changes to SysDB and receive asynchronous notifications without direct communication.
What works
State decoupling works. Agent isolation works. SysDB’s publish/subscribe model works. Running protocol stacks in user space works. The separation enables deterministic restarts, live patching, and consistent telemetry across hardware generations.
What does not
The architecture does not eliminate hardware dependency. It does not guarantee zero-downtime upgrades across all platforms. It does not abstract away the complexity of routing protocol convergence or hardware-specific failure modes.
What it changes
It changes how network operators reason about failure: faults are localised to agents, not cascading through shared kernel memory. It changes upgrade velocity: agents can be updated independently. It changes observability: SysDB provides a single source of truth for real-time state.
Is it worth your time
Yes—if you operate large-scale data centre or AI networks and need predictable, debuggable, upgradeable switching software. No—if your environment relies on monolithic, vendor-locked firmware or lacks Linux operations expertise.
