P*TIME: Highly Scalable OLTP DBMS for Managing Update-Intensive Stream Workload Sang K. Cha and Changbin Song   View Paper (PDF)   Return to INDUSTRIAL SESSION 2: NEW DBMS ARCHITECTURES AND PERFORMANCE Abstract Over the past thirty years since the system R and Ingres projects started to lay the foundation for today's RDBMS implementations, the underlying hardware and software platforms have changed dramatically. However, the fundamental RDBMS architecture, especially, the storage engine architecture, largely remains unchanged. While this conventional architecture may suffices for satisfying most of today's applications, its deliverable performance range is far from meeting the so-called growing "real-time enterprise" demand of acquiring and querying high-volume update data streams cost-effectively. P*TIME is a new, memory-centric light-weight OLTP RDBMS designed and built from scratch to deliver orders of magnitude higher scalability on commodity SMP hardware than existing RDBMS implementations, not only in search but also in update performance. Its storage engine layer incorporates our previous innovations for exploiting engine-level micro-parallelism such as differential logging and optimistic latch-free index traversal concurrency control protocol. This paper presents the architecture and performance of P*TIME and reports our experience of deploying P*TIME as the stock market database server at one of the largest on-line brokerage firms. ©2005 Association for Computing Machinery