Extended Memory Support for High Performance Transaction Systems
To achieve high performance transaction processing vertical as well as horizontal system growth is considered. A prime obstacle for vertical growth is the unfavorable ratio of I/O time vs. CPU time making it increasingly difficult to utilize fast CPUs and multiprocessors. Prerequisites for horizontal growth are a low communication overhead and effective load balancing; both subgoals are the more difficult to meet the more systems to be utilized. We propose the use of a fast and non-volatile extended memory which provides synchronous access for closely coupled systems. We discuss its properties supporting high volume transaction processing. Subsequently, we investigate its performance behavior for centralized and distributed computing environments. Simulation results are presented for synthetic Debit-Credit transactions as well as for real-life workloads represented by database traces. The use of non-volatile extended memory permits significant response time and throughput improvements, in particular for the real-life workloads. Lock contention, communication overhead and load balancing are Improved to a large extent compared to conventional architectures.
KeywordsTransaction Processing Storage Hierarchy Extended Memory Database Sharing
Computing Reviews ClassificationH.2.4 B.3.2 C.4
Unable to display preview. Download preview PDF.
- [An85]Anon et al.: A Measure of Transaction Processing Power. Datamation, April 1985,112–118.Google Scholar
- [BBD82]Balter, R.; Berard, P.; DeCitre, P.: Why Control of the Concurrency Level in Distributed Systems is More Fundamental than Deadlock Management. Proc 1st Symp. on Principles of Distributed Computing, 183–193,1982.Google Scholar
- [CKKS89]Copeland, G.; Keller, T.; Krishnamurthi, R.; Smith, M.: The Case for Safe RAM. Proc. 15th VLDB, 1989.Google Scholar
- [FRT90]Franaszek, P.A.; Robinson, J.T.; Thomasian, A.: Access Invariance and its Use in High Contention Environments. Proc. 6th IEEE Data Engineering Conf., 1990.Google Scholar
- [GP87]Gray, J.; Putzolu, F.: The 5 Minute Rule for Trading Memory for Disk Accesses and the 10 Byte Rule for Trading Memory for CPU Time. Proc. ACM SIGMOD conf., 395–398,1987.Google Scholar
- [Gr85]Gray, J. et al.: One Thousand Transactions per Second. Proc. IEEE Spring CompCon, San Francisco, 96–101,1985.Google Scholar
- [HGLW87]Herman, G.; Gopal, G.; Lee, K.C.; Weinrib, A.: A Datacycle Architecture for Very High Throughput Database Systems, Proc. SIGMOD ’87 Conf., San Francisco, CA, 97–103,1987.Google Scholar
- [Ra88]Rahm, E.: Empirical Performance Evaluation of Concurrency and Coherency Control Protocols for Data Sharing. IBM Research Report RC 14325, IBM TJ. Watson Research Center, 1988.Google Scholar
- [Ra90]Rahm, E.: Utilization of Extended Storage Architectures for High-Volume Transaction Processing. Technical Report (ZRI-Bericht 6/90), Computer Science Dept., Univ. Kaiserslautern, 1990.Google Scholar
- [Ra91]Rahm, E.: Performance Evaluation of Extended Storage Architecturesfor Transaction Processing. Technical Report, Computer Science Dept., Univ. Kaiserslautern, 1991.Google Scholar
- [Sh85]Shoens, K. et al.: The AMOEBA Project, Proc. IEEE Spring CompCon. 102–105,1985.Google Scholar