On Best-Effort and Dependability, Service-Orientation and Panacea

In this short position paper we argue that dependability technologies must be based on best-effort engineering principles, if they are to be useful for general-purpose enterprise and consumer IT. We will explain why ‘best-effort based dependability’ is not an oxymoron, but, instead, a necessity. We will also argue that service-orientation fits the best-effort engineering philosophy, and in that sense is part of the panacea for high-availability. 1 Best-Effort and Dependability We make the following observations: – reliability and availability is increasingly important in enterprise, home and other networked IT systems – these systems must rely on best-effort designs, which focus on scalability and simplicity over reliability and availability – improving reliability or availability will always be at the cost of some other system property (such as scalability, ease-of-use, extensibility) We conclude from these observations that traditional dependability solutions are typically ill-suited to respond to the demand for increased dependability, simply because they would sacrifice the system’s scalability or other properties to too large an extend. Instead, to improve dependability properties of enterprise and consumer IT systems we require technologies based on advanced best-effort methods. Only then can dependability be improved without sacrificing other important system properties. Advanced best-effort has the following implications on dependability design and research. First, we have to relax the stringency of the properties we are after. Instead of strict reliability guarantees, we need a probabilistic statement, or even a subjective statement about the objective. Subjective statements are possible, since they can be verified after the fact through user questionairres and 1 This article is based on the author’s position statement at ISAS 2005 during the panel discussion on ‘Are Service-Oriented Architectures the Panacea for the HighAvailability Challenge?’ M. Malek, E. Nette, and N. Suri (Eds.): ISAS 2005, LNCS 3694, pp. 99–101, 2005. c © Springer-Verlag Berlin Heidelberg 2005