Nanosecond Indexing of Graph Data With Hash Maps and VLists
Andrew Carter, Andrew Rodriguez, Yiming Yang, Scott M. Meyer · 2019
We introduce a wait-free, multi-reader, single-writer, "kill -9" durable, indexing structure for in-memory social graph databases. This structure requires no communication from the readers back to the writer, allowing for trivial read scalability and isolation. We support online updates without compromising availability or read performance. Our structure supports looking up small subgraphs in 80 nanoseconds and a materialization rate of 12 nanoseconds per edge. Storage takes 7 bytes per edge per index and supports almost 1 million online writes per second.