Schema Design

•

Sharding strategy: The ideal sharding strategy requires updates only to the latest shard, e.g., the “today’s data” shard. Data can be updated in any shard, but an updated shard must be merged for updates to take effect. If you update lots of shards continually, substantial processing is consumed by merging. OLAP works best when data is a forward-moving stream of data added to the latest shard with only occasional updates to older shards.

•

Use xlinks sparingly: Ideally, each shard is a self-contained “mini database”, where objects are linked only to objects in the same shard. Traversing link paths within a shard is very efficient. Xlink fields can be used to create cross-shard relationships, but processing xlinks is much slower and requires substantially more memory.