Scheduling in a Database-Based Distributed Operating System

Abstract

Current operating systems date from over 40 years ago and were designed for very different computing requirements, making them ill-equipped to handle serverless workloads as well as modern challenges in scalability, heterogeneity, availability, and security. Hence, we propose a radically new data-centric OS design for serverless computing. This database OS (DBOS) centralizes all cluster state in a uniform data model: database tables stored in a high-performance, distributed, main-memory database management system. Operations on this state will be performed via serverless, stateless tasks.

This thesis presents work done to build a preliminary scheduler and to implement and evaluate various global scheduling algorithms. We also demonstrate the performance of a modern DBMS in executing various scheduling operations.