Computing Systems Overview

Principles

Definition

Computing systems is the academic field dedicated to the study, design, analysis, and implementation of computational artifacts, encompassing the complex interplay of hardware and software components. This field investigates principles governing performance, reliability, security, and efficiency across diverse scales and applications, underpinning the infrastructure of the digital age.

Ontological type

Subfields and Topics

Core Principles

Architectures and Abstractions

Key developments

Key Figures

Real-Time Policy Parallelism era

M. Frans Kaashoek [1] was active across Massachusetts Institute of Technology [3] and Vrije Universiteit Amsterdam [4] during the Real-Time Policy Parallelism era. His key contribution in this era is the Exokernel [5], which rethinks operating system design by exposing hardware resources to application-level software and providing secure, policy-driven resource multiplexing. Dawson Engler [2] was active at Massachusetts Institute of Technology [3] during this era. Along with Kaashoek [1], Engler [2] helped co-design Exokernel [5], a contribution that advanced policy-driven resource management and laid the groundwork for user-level schedulers and scalable OS architectures in real-time parallelism.

Predictable Heterogeneous Scaling era

David A. Patterson [1] stands as a pivotal figure in this era of predictable heterogeneous scaling, with affiliations at Washington University in St. Louis [3] and University of California, San Diego [4]. In the 2010 paper A view of cloud computing [7], Patterson articulated a blueprint for cloud abstraction, resource pooling, and service models that advanced scalable and predictable infrastructures across heterogeneous platforms. Ion Stoica [2] is similarly influential in this era, with affiliations at the Massachusetts Institute of Technology [5] and Johns Hopkins University [6]. In the same A view of cloud computing [7], Stoica emphasized deterministic performance, scalable runtimes, and coordinated network and storage behavior to enable QoS-aware provisioning in cloud and distributed systems, a core driver of the era's predictable heterogeneity.

Memory-Centric AI Compute era

Dong Yu [1] is a prominent figure shaping memory-centric AI compute during this era, with affiliations at Johns Hopkins University [3] and Nanyang Technological University [4]. Dong Yu [1] contributed to this era through the 1-bit stochastic gradient descent and its application to data-parallel distributed training of speech DNNs [6], a method that reduces communication overhead and enables scalable, energy-aware learning in memory-centric systems. Jasha Droppo [2] is associated with Seattle University [5] in this era. The 1-bit stochastic gradient descent paper [6], co-authored by Droppo [2], demonstrates gradient quantization that reduces data movement and communication bottlenecks, critical for edge deployments and memory-aware runtimes in memory-centric AI compute.