A compiler-enabled model- and measurement-driven adaptation environment for dependability and performance

• Main Authors: Adve, V.S., Agbaria, A., Hiltunen, M.A., Iyer, R.K., Joshi, K.R., Kalbarczyk, Z., Lefever, R.M., Plante, R., Sanders, W.H., Schlichting, R.D.
• Format: Conference Proceeding
• Language: English
• Subjects: Adaptation model; Area measurement; Buildings; Computer applications; Costs; Distributed computing; Grid computing; Military computing; Predictive models; Runtime environment
• ISSN: 1530-2075
• DOI: 10.1109/IPDPS.2005.9

Traditional techniques for building dependable, high-performance distributed systems are too expensive for most non-critical systems, often causing dependability to be sidelined as a design goal. Nevertheless, systems are expected to be dependable, and if dependability could be provided at a lower cost, many applications would stand to benefit. We believe that compiler techniques can be used to create novel and enhance existing dependability mechanisms to create a wider range of cost/dependability tradeoffs than is currently available. Similarly, compilers can assist in the area of error detection by expanding the range of errors that can be detected. New compiler techniques, combined with model-driven adaptation and control mechanisms, can be used to dynamically guide a system as it makes choices, with cost, dependability, and performance tradeoffs, in response to the occurrence of faults and changes in the environment. This paper reports on a new project that is exploring the approach. The broad goal of the work is to create a powerful yet flexible runtime environment for dependable and high-performance systems that operate within much lower cost constraints than is currently possible.