A Signature Based Concurrency Scheme for Scalable Distributed Data Structures Thomas Schwarz, S.J. Department of Computer Engineering Santa Clara University tjschwarz@scu.edu JoAnne Holliday Department of Computer Engineering Santa Clara University jholliday@scu.edu Abstract We evaluate theoretically a new validation scheme based on signatures with algebraic properties. In this proposal, we read records without regard to transactional isolation, but we validate these reads before we commit any writes. We evaluate the concurrency level of this basic scheme and propose a way to integrate it with a typical hash based SDDS that avoids phantoms. Keywords: Scalable Distributed Data Structures, Transaction Processing, Algebraic Signa- tures, Concurrency Control. 1 Introduction Transaction processing is a cornerstone of modern information technology. Techniques to allow concurrent transactions have been studied for at least the last three decades. In this paper, we investigate one recently proposed concurrency algorithm [11, 12] for distributed databases, especially Scalable Distributed Data Structures (SDDS), that uses signatures to validate reads. In particular, we investigate the degree of isolation and also how to achieve “the third degree”, that is, isolated, serializable, and repeatable reads. As it turns out, the original proposal only implements the ANSI SQL Read Committed level, but an enhancement yields full isolation levels with excellent concurrency. A transaction in our scheme reads records without regards for isolation control. It can determine its read set dynamically, that is, it does not have to predeclare reads. Before the transaction commits its writes and before it commits to the user, it verifies all its reads with a multicast. If all its reads have been successfully verified, the transaction performs all its writes with another multicast and returns to the client application. 1