Disk ANN

Disk ANN

• Intermediate aggregator (# of shards < # of machines)

• Index always better than the other?

• Access pattern different?

• Already have graph embeddings, why build embeddings at all?