Detecting Interference Between Applications and Improving the Scheduling Using Malleable Application Proxies

Abstract

LIMITLESS is a lightweight and scalable framework that provides a holistic view of the system employing the combination of both platform and application monitoring. This paper presents a novel feature for improving the scheduling process based on the performance prediction and the detection of interference between real applications. This feature consists of using malleable synthetic benchmark clones (proxies) for the applications executed in the system with two objectives: (1) build large and representative datasets that can be used to train the machine learning algorithms for predicting, and (2) evaluate if two applications can share the same compute node in order to leverage the unused node resources. Other related works use detailed micro-architecture independent metrics obtained from functional simulators, which are hard to generate in many new applications. The results are proxies that preserve many of the original features of the applications (control flow, memory access pattern, etc.), and their code needs obfuscation to make impossible the use of reverse engineering. LIMITLESS generates application proxies based on generic-purpose performance information collected from monitoring. It means that other methods may obtain more accurate execution behaviours. However, LIMITLESS’ proxies generate similar performance without extracting data from the binaries, without the necessity of managing code or data from the applications, and they can be shared securely because they have not been generated using any piece of the original code. LIMITLESS leverages the generated proxies to execute them offline. Each execution increases the datasets of the machine learning algorithms to improve the application scheduling. Besides, the executions between proxies are combined to detect performance degradation (interference) without the necessity of waiting for the execution of the real applications, which depends on the users. In this work, we evaluate the proposed proxy generation approach on a set of benchmarks and applications. We compare the performance obtained during the execution of the proxies and the applications to show their similarity. Finally, we include an evaluation of the interference detection using this approach. As far as we know, this is the first work that uses malleable proxies.