Optimise web browsing on heterogeneous mobile platforms: A machine learning based approach

Abstract

Web browsing is an activity that billions of mobile users perform on a daily basis. Battery life is a primary concern to many mobile users who often find their phone has died at most inconvenient times. The heterogeneous multi-core architecture is a solution for energy-efficient processing. However, the current mobile web browsers rely on the operating system to exploit the underlying hardware, which has no knowledge of individual web contents and often leads to poor energy efficiency. This paper describes an automatic approach to render mobile web workloads for performance and energy efficiency. It achieves this by developing a machine learning based approach to predict which processor to use to run the web rendering engine and at what frequencies the processors should operate. Our predictor learns offline from a set of training web workloads. The built predictor is then integrated into the browser to predict the optimal processor configuration at runtime, taking into account the web workload characteristics and the optimisation goal: whether it is load time, energy consumption or a trade-off between them. We evaluate our approach on a representative ARM big.LITTLE mobile architecture using the hottest 500 webpages. Our approach achieves 80% of the performance delivered by an ideal predictor. We obtain, on average, 45%, 63.5% and 81% improvement respectively for load time, energy consumption and the energy delay product, when compared to the Linux heterogeneous multi-processing scheduler.

A survey of techniques for improving efficiency of mobile web browsing

“…FIGURE 3 Dependency graph on loading a sample webpage 13 that has a PLT of 380 ms. The dashed line shows the critical path (A) FIGURE 6 An illustration of A, a symmetric multicore processor; B, an asymmetric multicore processor having two cores with the same ISA but different configurations and; C, an asymmetric multicore processor having two cores with different ISAs Techniques that work by predicting … PLT and energy consumption of a webpage 14,23,24 reading time of the user 16 image quality for user satisfaction 25 next website to visit 26,27 next subresource to visit 28 size of object to be fetched 29 Network protocol SPDY 3,[30][31][32][33][34] HTTPS 3 HTTP nearly all others Use of algorithms stochastic gradient boosting, 27 gradient-boosted regression tree, 16 gradient descent algorithm, 35 integer linear programming, 8 support vector machine, 23 random forest learning, 29 greedy algorithm, 9 breadth-first search, 36 regression modeling, 14 statistical modeling, 37 control-theoretic approach 38 TABLE 3 Optimization objective of different works and the data set used by them…”

“…In this section, we discuss works that evaluate browser applications on different processor architectures (Section 6.1); propose DVFS, power gating, and asymmetric multicore scheduling (Section 6.2); perform intelligent scheduling of browser threads to cores (Section 6.3); and propose hardware customization (Section 6.4) and QoS abstractions (Section 6.5) for MWB. Power gating 46,50,52,59 Using extra hardware to operate on different properties of the ''style'' kernel in parallel 44 Thread/task scheduling [37][38][39][46][47][48][50][51][52]59 Basis of Thread/Task Scheduling Smart app (involving user interaction) vs maintenance functions 50,59 Critical (which impact PLT) vs noncritical threads 46,52 Long-lived vs short-lived threads 48 Task latency and/or deadline considerations [37][38][39]47 Power considerations 14,23 Thermal considerations 47 Functionalities of processing units 39 Consolidating threads on few cores to remove false parallelism 48…”

Adaptive optimization for OpenCL programs on embedded heterogeneous systems

“…Predictive Modeling Machine learning based predictive modeling is emerging as a powerful technique for optimizing parallel programs [28,29,33,37,[39][40][41]. Its great advantage is its ability to adapt to changing platforms as it has no a prior assumptions about their behavior.…”

“…Unlike these approaches which all use analytic models or hard-wired heuristics to perform optimization for a specific goal, we develop a portable method that can automatically re-target for any optimization metric. Our approach shares the same spirit as the work presented by Ren et al [32], we both use machine learning to build predictive models for energy optimization. However, we target optimizing OPENCL programs on heterogeneous systems, while [32] focuses on scheduling mobile web browsing processes on CPUs.…”

“…Our approach shares the same spirit as the work presented by Ren et al [32], we both use machine learning to build predictive models for energy optimization. However, we target optimizing OPENCL programs on heterogeneous systems, while [32] focuses on scheduling mobile web browsing processes on CPUs.…”