This section will recommend designs and starting values for vector-indexed tables and searches: what to choose for a given workload, and how to size for it. Recommendations are a starting point, based on benchmarks and deployments. Always measure with your own data before you rely on them.
What this section will contain
- Search settings by workload: starting values for the probe settings for RAG, recommendation, near-duplicate detection and filtered multi-tenant search.
- Choosing quantization: which quantization pair to start with for a given recall target and table size.
- Sizing and throughput: how queries per second and latency scale with CNodes and concurrency.
- Multi-tenant designs: one table with a tenant filter or one table per tenant, and when each fits.
Each page will say which VAST Cluster version and test setup its recommendations come from.
Where to find how it works
The rest of the VAST Vector Search docs explain how each feature works, what each setting does and how to measure the result: search session settings, quantization, design a vector table, filters, measure recall and latency, performance and sizing. Best-practices pages link to those pages rather than repeat them.
Status
This section is being written. Until its pages are published, start with the pages above, and ask your VAST account team for guidance on your workload.