Flexible and non-invasive qos for scalable internet services

As the commercial importance and sophistication of web-based technologies continues to grow, Internet services play an increasingly important role in mission-critical tasks. In order for future services to succeed in such critical endeavors it is imperative that they are delivered reliably. However, delivering reliable service quality in large-volume Internet services is a difficult challenge. While both hardware and software-based approaches have been proposed to meet this challenge, none of them address the problem adequately. Hardware-based solutions suffer from poor resource efficiency due to the wide and unpredictable traffic patterns of Internet services. Software solutions are invasive, and they incur high development and maintenance costs due to the magnitude, complexity and fast-pace in which the services must be offered. We postulate that it is possible to solve the scalable QoS problem by employing non-invasive techniques that allow Internet services to be delivered with reliable QoS guarantees without compromising the efficiency or the flexibility requirements of service providers. This dissertation presents Quorum, a novel, non-invasive approach that treats Internet sites as black-box systems and uses traffic shaping, admission control and feedback monitoring at its borders to ensure QoS guarantees more efficiently and with greater flexibility than is currently possible. We present a formal model for Internet services as well as a model for describing the computational behavior of Internet sites. Based on these models we design and implement a complete working prototype of Quorum and present results for realistic and complex Internet services running in large-scale clustered settings. We show how Quorum can enforce throughput and response time guarantees for unmodified Internet services independently of site configuration internals, changes in the existing infrastructure, reprogramming of the software, or instrumentation of the hardware resources. In addition, we demonstrate how this increase in flexibility will enable new mission-critical services to emerge.