Simplifying heterogeneous migration between x86 and ARM machines

TL;DR

This paper discusses simplifying process migration between x86 and ARM architectures in heterogeneous computing by creating a shared address space, aiming to reduce migration overhead and improve efficiency.

Contribution

It proposes a method to create a uniform address space between x86 and ARM processes, simplifying migration and enabling dynamic scheduling in heterogeneous systems.

Findings

Initial analysis of common stack layouts between x86 and ARM

Development of techniques for shared address space creation

Benchmarking of performance and energy efficiency improvements

Abstract

Heterogeneous computing is the strategy of deploying multiple types of processing elements within a single workflow, and allowing each to perform the tasks to which is best suited. To fully harness the power of heterogeneity, we want to be able to dynamically schedule portions of the code and migrate processes at runtime between the architectures. Also, migration includes transforming the execution state of the process, which induces a significant overhead that offsets the benefits of migrating in the first place. The goal of my PhD is to work on techniques that allow applications to run on heterogeneous cores under a shared programming model, and to tackle the generic problem of creating a uniform address space between processes running on these highly diverse systems. This would greatly simplify the migration process. We will start by examining a common stack layout between x86 and ARM binaries, focusing on these two widely spread architectures, and later we will try to generalize to other more diverse execution environments. On top of that, the performance and energy efficiency of the above effort compared to current approaches will be benchmarked.