Language and Compiler Support for Adaptive Distributed Applications

Abstract

Many distributed applications have to meet their performance or quality-of-service goals in environments where available resources change contantly. Important classes of distributed applications (including distributed multimedia codes, applications for mobile devices, and computational grid codes) use runtime adaptation in order to achieve their goals. The adaptation behavior in these applications is usually programmed in ad hoc code that is directly incorporated into the base application, resulting in systems that are complex to develop, maintain, modify, and debug. Furthermore, it is virtually impossible to extract high-level information about adaptive behaviour using program analysis, even if there were compiler and runtime sytems that could exploit such information. The goal of our research is to develop compiler and programming language support to simplify the development and improve the performance of adaptive distribtued applications. We describe a simple set of language extensions for adaptive distributed applications, and discuss potential compiler techniqes to support such appliations. We also propose a task-graph-based framework that can be used to formalize the description of a wide range of adaptation operations.

Empirical Optimization for a Sparse Linear Solver: A Case Study

“…An alternative approach extends existing languages to provide hints or directives to the compiler about the dynamic nature of the application. Adve et al (28) describe an extension to the class hierarchy of an objected-oriented model of computation that defines three basic concepts for dynamic adaptationadaptors, metrics and events.…”

An approach to implementing dynamic adaptation in C++

“…Transparent to the client, the invocation is actually made on the filter object, as depicted in step 1. 1 The filter object responds by reifying the call into a message that is then sent to an AMH object (steps 2-3). A message comprises a method object paired with an array that contains the actual values that were passed as arguments to the method call.…”

“…More recently, TRAP/J introduced the notion of filter classes that package the input-filter functionality behind an interface that is plug-compatible with the application class whose objects we wish to adorn with filters [22]. The "attachment" of input filters to selected objects is then implemented by systematically replacing client-side requests to instantiate selected classes with requests to instantiate the corresponding filter 1 This client-unaware redirection is achieved using techniques that are explained in Section 2.2. 2 The definition of method signature that we use in this paper differs slightly from the one that some authorities use in that we do not consider the class name associated with a method to be part of the method's signature.…”

“…Numerous approaches have been used to address this shortcoming, particularly in the case of Java. Examples include extending the language itself [1,14,25] or extending the Java Virtual Machine [21,12,19,20,8]. An alternative approach that has gained recent attention is to "automatically" augment the application code with support for behavioral reflection, in particular, code to intercept and redirect interactions among objects [7,23,24,26].…”

Performance vs. accuracy trade-offs for large-scale image analysis applications

“…Prior efforts by different groups include providing language and compiler support to meet the QoS demands [8], [9] for such applications. EPIQ [10], [11], ERDoS [12], [13], ActiveHarmony [14], [15], Agilos [16] and, [17], [18] are all examples of middleware efforts that have been developed over the years to address the problem of dynamic QoS-driven resource management for a wide range of adaptive applications including visualization, image processing, e-commerce, object-tracking and multimedia processing.…”