Towards autonomic distributed and parallel server systems

点击翻译按钮获取中文摘要

The long term goal of this research project is to simplify the programming and management of distributed and parallel computer systems. As computer systems become more powerful, due to their larger scale and escalating hardware complexity, it is becoming harder for humans to maintain them and to effectively exploit their plentiful resources in software. The future seems to hold increasing dependence on expensive teams of expert programmers and administrators. This is a critical problem for industry and academia alike. Highly qualified programmers and system administrators simply take longer to produce than more complex and more plentiful hardware. This proposal has two thrusts that address these challenges by facilitating: i) automated computer system management through a set of adaptive system mechanisms that address performance bottlenecks and repair faults automatically and transparently to the user and ii) simpler application programming in distributed and parallel environments through the use of a new programming paradigm called "Transactional Memory". Although both project directions have wide applicability, we will explore the application of the proposed techniques in a specific complex class of distributed systems, that of large-scale Internet systems that generate content on-the-fly, called dynamic content servers. Dynamic content servers providing modern information services, e.g., Amazon.com and EBay.com, are built by composing a variety of software systems that interact within a networked environment to provide the service. For example, electronic commerce systems typically consist of multiple tiers, such as, database servers, application servers, and web servers. Organizations currently spend much time and effort on managing information services, with no guarantee of success. This project will provide automatic capabilities for reconfiguration, tuning, automatic code parallelization and seamless failure recovery in networked information services, which will make it easier for organizations that rely on these services to deploy and manage them.