Autopilot: adaptive control of distributed applications

Abstract

With increasing development of applications for heterogeneous, distributed c omputing grids, the focus of performance analysis has shifted f r om a posteriori optimization on homogeneous parallel systems to application tuning for heterogeneous resources with time varying availability. This shift has profound implications for performance instrumentation and analysis techniques. Autopilot is a new infrastructure for dynamic performance tuning of heterogeneous computational grids based on closed l o op control. This paper describes the Autopilot model of distributed sensors, actuators, and decision procedures, reports preliminary performance b enchmarks, and presents a case study in which the Autopilot library is utilized in the development of an adaptive parallel input output system.

Online Adaptive Code Generation and Tuning

“…This information is used to model the application behavior and predict the relative benefit of using large pages for different application data structures. Autopilot [19] is an online tuning framework for parallel applications. Based on application request patterns and observed system performance, Autopilot's real-time adaptive control mechanism automatically chooses and configures resource management algorithms.…”

DIPAS: A distributed performance analysis service for grid service-based workflows

“…In contrast to K-WfGrid, it does not support cyclic and Web servicebased workflows. The Autopilot [37] is a distributed performance measurement and resource control system that supports the monitoring and adaptation of distributed applications. G-PM/OCM-G supports the online analysis of MPIbased applications in the Grid [38].…”

Continuous performance monitoring for large-scale parallel applications

“…Tools that monitor application performance at runtime include Autopilot [10] and TAU, through a series of development projects leading to TAUg and TAUoverSupermon [11], [12], [13], [14]. Autopilot attempted to tune the performance of a long-running application through the use of "sensors" and "actuators" to allow a remote client to adjust the performance properties of the application.…”