Exploiting Client Disks for High Performance Client- Server Architectures. Submitted for Publication

7 Conclusions EOS is a storage manager that has been prototyped at AT&T Bell Laboratories as a vehicle for research into distributed storage architectures for database systems and specially those that integrate programming languages and databases. EOS is the storage manager of ODE 1], an object oriented database management system also being developed at Bell Laboratories. In this paper we have described the client-server architecture of EOS and the concurrency control and recovery mechanisms it provides. The MG-2V-2PL concurrency control protocol was chosen with the goal of increasing the concurrency level of the system in the client-server environment. The no-undo/redo recovery method was designed with the goal of minimizing the impact of the recovery related activities during normal transaction processing, while providing fast transaction abort and system restart times. We also presented measurements of the recovery implementation in EOS. From the results computed, and from the limited number of published performance results for logging and recovery systems, we concluded that the recovery overhead in EOS is minimal, despite the write-intensive nature of the tests we ran. We are currently working on issues related to inter-transaction caching within the same application and across applications that share the same cache, providing support for multiple servers, distributed transactions , and client-side logging 6, 13]. References 1] R. Agrawal and N. Gehani. ODE (Object Database and Environment): The language and the data model. A study of three alternative workstation-server archi-tectures for object-oriented database systems. In Pro-ture of the ORION next-generation database system. A transaction recovery method supporting ne-granularity locking and partial roll-backs using write-ahead logging. DB Execution Abort Restart name time (sec) time time FewObj 17.2 1.