Skip to main content
ParalleliX Docs

// Chapter 06 · Protocol

Scale Model

Scalability is a structural property of this design, not a deferred add-on.

4 min3 sectionsProtocol

The policy does not get rewritten

// 6.0 · scalability as a structural property, not a deferred addition

// Invariant

One scheduling policy. Ten to ten thousand nodes. No rewrite.

Scalability is structural here, not a deferred addition. The network grows by absorbing nodes, and the scheduler is not rewritten as the active set grows. Three mechanisms make that possible.

Three mechanisms

// 6.1 · 6.2 · 6.3 · each addresses a different growth surface

// 6.1 · dynamic node integration3 stages · cold-start to dispatch
  1. // 01

    Manifest submitted

    Operator signs capability declaration.

  2. // 02

    Signature verified

    Coordinator validates key binding.

  3. // 03

    Pool integrated

    Node enters available capacity immediately.

Node onboarding is permissionless from launch. Anyone who stakes the 50,000 $PRLX minimum can register; no allowlist, no off-site outreach gating.

// 6.2 · adaptive schedulingpolicy unchanged · pool varies
  • 10 nodespermissionless · low loadevery node sees most requests · priority queue mostly empty
  • 100 nodespermissionless · class-stratifiedbest-fit dominates · idle nodes routinely available
  • 1,000 nodesopen marketplace · plannedreputation breaks ties · queue depth tracks workload elasticity
  • 10,000 nodesglobal · multi-regionpolicy unchanged · class-specific tuning planned

The priority queue and best-fit allocation track the current shape of available capacity without per-tier reconfiguration.

// 6.3 · heterogeneous workload supportone scheduler · all classes
ONE SCHEDULERAISCIDATABLOCKCHAINRENDER

Capability declarations on nodes are the routing key. One scheduler, all workload classes.

Note

Workload-class-specific allocation policies are planned on top of the same primitives. The current scheduler is workload-agnostic.

Fixed policy, variable pool

// 6.4 · what the network preserves as it grows

// Scale invariants · scheduler layer

  • Policy structurepriority queue + best-fit allocationcandidate pool size
  • Signal ordercapability → capacity → priority → reputationnumber of nodes per capability class
  • Registration flowstake on-chain → signed manifest → verify → poolcandidate pool size
  • Routing keycapability declarations on nodesnumber of declared workload classes
  • Workload classesAI · SCI · DATA · BLOCKCHAIN · RENDERper-class allocation tuning is planned

Access boundaries for node onboarding

// 6.1.1 · node onboarding is permissionless from launch

// Node access register · live now vs plannedpermissionless now

  1. // 01

    Permissionless staking

    · Permissionless· Available now

    Node onboarding is permissionless from launch. Anyone who stakes the 50,000 $PRLX minimum calls registerNode on NodeRegistryLocker; no allowlist, no off-site outreach gating. No slashing; principal returned in full on unstake (7-day cooldown). Reputation is constant 1.0.

  2. // 02

    Open marketplace + reputation

    · Permissionless· Planned

    Demand opens beyond ParalleliX AI to third-party submitters, with dynamic pricing and SLA enforcement. A reputation score (uptime, completion, validation) activates: it lowers dispatch priority and can drop a node from the active set, but never seizes stake.

// Where to go next · reading path