Get started

MARS-S · Sharding research

A larger registry.A more local answer.

MARS-S explores how a registry could grow across regional shards while keeping records close to the software that uses them. A proposed architecture for the next scale of discovery.

  • Draft specification
  • Feature-gated prototype
  • Current persistence: PostgreSQL
A closer lookA map for a much bigger catalog.

The MARS-S specification describes hierarchical routing across shards. This is an architectural illustration of the proposed design.

  1. 01Locate the namespace
  2. 02Follow the routing hierarchy
  3. 03Resolve the intended shard

01 / architecture

Separate the record from its root.

The draft proposes a regional layer for full entity data and a global layer for compact root commitments. Each carries a different part of the workload.

Regional shard layer

Full records would live in a sharded Weave topology. Affinity groups entities by the organizations and capabilities most likely to interact, with replication and reassignment described in the draft.

Proposed distributed persistence.

Global anchor layer

Sigil would hold compact commitments for human roots, group roots, revocations, and lineage roots. The goal is shared verification without placing every full record on the global path.

Proposed MARS-S authority model.

02 / resolution

Look nearby. Widen when needed.

The proposed resolution cascade begins with the local shard. A miss broadens the search to a regional cluster, then to global root information.

  1. 01

    Local shard

    Serve records close to their publishers and frequent consumers. Affinity is designed around relationships, not only geography.

  2. 02

    Regional cluster

    Consult neighboring shards when the local view cannot satisfy a request. Regional replication is intended to absorb common cross-organization reads.

  3. 03

    Global anchor

    Use compact global root information as the proposed verification foundation, outside the ordinary full-record hot path.

This is a design sequence. The former 12 ms, 48 ms, and billion-entry figures are architecture targets, not published performance measurements.

03 / topology

A map of relationships.

A conceptual topology places compact roots at the center and full-record shards around them. Lines describe proposed relationships, not a live network.

Conceptual MARS-S topology: twelve regional shards surrounding a global root layer.ROOTSPROPOSEDS01S02S03S04S05S06S07S08S09S10S11S12

Primary and shadow assignments. The draft separates a record’s main shard from its replication targets.

Change as demand changes. Splits, merges, migrations, and failure recovery are part of the proposed shard lifecycle.

Keep the contract recognizable. MARS-S is designed as an extension to the registry model, with a single-shard starting point and progressively broader conformance levels.

04 / status

An open direction. A clear starting point.

  • Reference service

    The current production profile uses PostgreSQL, with the canonical v1 record and HTTP contracts.

  • Research implementation

    mars-shard and feature-gated storage code provide a place to develop and evaluate the proposed topology.

  • Work ahead

    Distributed persistence, routing, migration, and performance need independent runtime evidence before their design targets become service claims.

Read the MARS-S draft