marianotepper commented on issue #14681: URL: https://github.com/apache/lucene/issues/14681#issuecomment-3005770808
> Also, calling this a “disk-based” solution seems a bit misleading if the graph still has to be built fully in memory. That’s often the core problem people are trying to get around. JVector does not store the graph in memory. It stores the vast majority of graph on disk. It uses a mixed approach between HNSW and DiskANN, where the top layers of the HNSW-style hierarchy are stored in memory and the base layer is a DiskANN-style graph stored on disk. Note that in HNSW, the upper layers are usually a very small fraction of the overall footprint (~3% for a max-degree of 32). As you say, in the current implementation, the graph is built in memory and then written to disk. We are currently working towards removing this constraint. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org For additional commands, e-mail: issues-h...@lucene.apache.org
