Understanding fault-tolerant distributed systems
Communications of the ACM
An efficient scheme for providing high availability
SIGMOD '92 Proceedings of the 1992 ACM SIGMOD international conference on Management of data
| Hi-index | 0.00 |
The increased performance capabilities of UNIX server systems have led to their acceptance as the server of choice for medium-sized and large organizations. But performance is just one facet. Another facet is the end user perception of the availability of an information system.Traditional mainframe based IS shops have a long experience in supplying computing services to their commercial end users. The ultimate goal of the end user is to have no downtimes for his work at his PC or workstation terminal. Key issues related to system availability in client/server based information systems remain the same as in the mainframe based world, e.g. system responsiveness, maximum downtime per year and maximum number of system outages per year. But there are also new aspects, which have been introduced into the discussion.In a multi-tiered client/server based information system the OLTP workload is distributed on different servers. Hence one can ask: Why should a failure of one server automatically imply downtime for the whole system ? Can't most of the system continue to operate ? Redistribution of the workload on the remaining active servers can be used to attack this problem. Workload balancing can be applied for replicated system services. Other techniques have to be applied for non-replicated system services.This paper considers client/server based applications running in a local or wide area network of computers in a distributed system.