Distributed databases principles and systems
Distributed databases principles and systems
Ethernet: distributed packet switching for local computer networks
Communications of the ACM
| Hi-index | 0.00 |
In this paper, we describe an experiment to evaluate a distributed architecture via functional database workload decomposition. A workload in a circuit pack assembly environment was decomposed and mapped onto a frontend/backend distributed computer architecture. To evaluate this distributed architecture, an operational model for capacity planning was devised, and performance and cost-effectiveness measures were chosen. Model parameters were estimated through benchmark experiments in a distributed system consisting of various super-microcomputers connected by a CSMA/CD local area network with INGRES as the database management system.The frontend/backend architecture consists of a backend data repository and analysis computer system and a few frontend computer systems dedicated for data collection and manufacturing process verification. Because of the significant software overhead in communication protocol and database processing, information exchange was batched between the backend and frontend systems to amortize such cost to improve overall system performance.Results of the experiments were analyzed to gain quantitative insight on the feasibility of such decomposition and its mapping onto the proposed architecture. With sufficient batching, the proposed distributed architecture not only has more overall system capacity, but also is more cost-effective than the typical centralized architecture. The approach described is applicable in more general contexts. Advantages of such distributed systems include the relative robustness of the distributed architecture under single point failure mode and the ease of capacity growth by upgrading the computer systems and/or by increasing the number of frontend systems.