Amnesia — A Distributed Real-Time Primary Memory DBMS with a Deductive Query Language

Amnesia is a DBMS for telecommunication applications. It is written in Erlang[1]. Erlang is a distributed functional real-time language with explicit parallelism and light-weight processes which runs on many platforms. Processes communicates by asynchronous message passing. Erlang programs can be run stand alone on a single computer or on a distributed network of computers.

Amnesia aims to solve problems which need: 1) Fast real-time key-value lookup. 2) Complicated non-realtime queries, 3) Distributed data and 4) High fault-tolerance.

Amnesia provides a) primary memory storage. b) a deductive query language. c) a fast program interface for very simple queries. d) transactions. e) crash recovery, f) distributed transactions and query evaluation and g) replication of data on separate nodes.

The query language is Datalog extended with lists and the Erlang data type tuples. The resolution is SLDAL[3]. Non-recursive parts of large queries are optimized with the randomized algorithm 2-Phase Optimization[2]. For now we use standard distributed algorithms for transactions, logging and checkpointing. Logs and checkpoints are written to disks or other nodes at certain intervals.

A high-level symbolic language with inherent distribution enables the implementation of Amnesia by a very small group of people within a relatively short time. A special key-value dictionary in C with simple patternmatching was added to Erlang to get speed (30μs constant time for lookups on a SPARC Station 5).

[1] Joe Armstrong. Robert Virding, and Mike Williams. Concnrren Programrning in ERLANG. Prentice Hall. 1993.

[2] Yannis E. loannidis and Younkyung Cha Kang. Randomized algorithms for optimizing large join queries. In Proceedings of the 1990 ACMSIGMOD Conference. pages 312-321. 1990.

[3] L Vielle. Recursive query processing: The power of logic. Theoretical Computer Science, 69(1). December 1989.