DiskANN

DiskANN is a disk-resident graph-based ANN algorithm introduced in a 2019 Microsoft Research paper that enables billion-scale vector search on a single machine with modest RAM by storing most of the index on SSD and keeping only a small graph cache in memory. The algorithm extends ideas from HNSW with disk-aware traversal optimizations including beam search, bidirectional edges, and product quantization compression of the in-memory cache. DiskANN can index a billion 100-dimensional vectors on a single 64GB-RAM workstation, a workload that would otherwise require a distributed cluster of dozens of HNSW-loaded machines. Microsoft uses DiskANN internally in Bing and Azure Cognitive Search, and the open-source release on GitHub has been integrated into Milvus, Vespa, and several other vector databases. DiskANN's cost-efficiency makes it attractive for large enterprise RAG deployments and for embedded AI applications running on edge hardware. AI governance teams evaluate DiskANN against HNSW and IVF-PQ when storage cost dominates the deployment economics.

DiskANN economics through Centralpoint: Centralpoint supports DiskANN-backed vector retrieval as a cost-efficient option for billion-scale RAG, paired with any generative LLM in a model-agnostic stack. Tokens are metered per skill and audience, prompts stay local, and DiskANN-powered chatbots deploy across portals with one line of JavaScript.


Related Keywords:
DiskANN,,