Skip to main content

Scalability and performance

For SENEX, scalability is not only request volume. A system must preserve owner control, predictable behavior, recoverability, and policy enforcement as participation grows. Improving throughput while weakening those properties would not meet the product objective.

No public production benchmark

SENEX does not currently claim production-scale capacity, global availability, linear scaling, a specific uptime level, or guaranteed response time. The working foundation is local-first; network workloads are a V1-testnet target, and benchmark and capacity claims need external validation. Performance claims require reproducible evidence. Status labels are defined in Current status.

Performance dimensions​

Public evaluation should report multiple dimensions rather than one headline number:

  • Local responsiveness: time and resource cost for tasks that remain on the owner's device.
  • Interaction latency: time added when an authorized task crosses a network boundary.
  • Sustained throughput: useful completed work over time, including verification and policy checks.
  • Resource efficiency: computation, memory, storage, energy, and network use for a defined workload.
  • Reliability: successful completion and correct recovery during expected operating conditions.
  • Degraded behavior: what users experience during overload, disconnection, slow peers, or partial failure.
  • State growth: how operational records grow and how retention policy affects capacity.
  • Coordination overhead: work required to authorize, verify, review, and reconcile shared outcomes.
  • Privacy and safety cost: the measurable overhead of controls that must not be bypassed for speed.

Workload classes​

Benchmarks should distinguish workloads with materially different constraints:

Local AIA tasks​

Evaluation includes startup behavior, interactive response, local retrieval, storage growth, and operation with limited connectivity. Results must identify the tested hardware and software configuration.

Governed peer interactions​

Evaluation includes authorization, connection setup, policy checks, message exchange, revocation, retries, and evidence needed for review. A fast happy path does not demonstrate safe operation under failure.

Shared-service workflows​

V1-testnet work may evaluate task dispatch, provider response, verification, and test-only ASHA accounting as one end-to-end flow. Provider settlement and real rewards are not live.

Administrative and recovery workflows​

Capacity planning must include upgrades, incident containment, policy changes, audit queries, backup, restoration, and participant exit—not only normal requests.

V1-testnet validation plan​

A credible testnet report should:

  1. define the exact release, configuration, workload, and measurement method;
  2. separate local, network, and external-service time;
  3. include normal load, burst load, long-running load, and failure scenarios;
  4. report errors, retries, tail behavior, and resource use—not only averages;
  5. test revocation, recovery, and policy enforcement under load;
  6. disclose material limitations and unresolved bottlenecks; and
  7. make enough of the harness and results available for independent review without publishing security-sensitive internals.

Scaling principles​

  • Prefer local execution when it meets the user's need and policy.
  • Add shared coordination only when the outcome requires it.
  • Bound resource use so one participant cannot silently degrade others.
  • Apply backpressure and explicit failure rather than concealing overload.
  • Preserve compatibility and recovery during staged upgrades.
  • Treat heterogeneous devices and connectivity as normal conditions.
  • Keep observability useful for operators without turning it into user surveillance.
  • Optimize after measurement, not from unverified projections.

Release gates​

SENEX should not publish production-scale claims until the relevant results are reproducible, independently reviewed, and tied to a named release. V1-testnet results describe that test environment only; they do not establish V1 capacity.