Hermes Wiki

ScatterGather

A request is scattered to multiple parallel workers/services, then their results are gathered back into one response — fan-out + fan-in combined for a single request/response cycle.

Why we need this / what value this brings

Parallelizes otherwise-sequential lookups against multiple sources, cutting total latency to roughly the slowest single source instead of the sum of all of them.

When to use this

When a single response needs data gathered from multiple independent, parallelizable sources within one request.

How to use or implement this

Set a per-source timeout, decide whether a partial result (some sources responded, others timed out) is acceptable to return, and cap total wait time to the slowest acceptable source.

Research questions

  • Searching multiple provider systems or data sources in parallel and merging results before returning to the client.

Empty folder — drop notes, links, and findings here as you research.

Hermes Wiki