Modeling and adaptive feedback control for multi-class service quality assurance in stress-resilient Internet servers

互联网服务具有流量突发性的特点。为了保护高级客户免受流量激增的影响，同时又不过度影响基本客户的需求，具有压力弹性的服务器必须能够以协调的方式控制多类请求的质量，以便在压力条件下公平而优雅地降低性能。该项目的目标是在网络边缘开发一个自主资源管理系统，以支持此类服务质量保证模型。资源管理系统有三个关键创新。首先是用于多类质量保证的应用级qos感知资源管理框架。它的新颖之处在于能够在多层网站中提供客户端感知到的端到端页面浏览响应时间的保证。第二个是2D服务差异化模型，该模型在会话间和会话内两个维度上制定了基于会话的工作负载需求，并将其与电子商务应用中的收入最大化目标联系起来。其三是无模型自整定反馈控制器，动态调节资源分配过程，在长、短时间尺度上都具有高度的控制鲁棒性。边缘服务器是互联网的重要组成部分，对我们的经济和社会产生了深远的影响。本研究将促进对服务器在压力条件下服务质量保证问题的发现和理解。通过将自主资源管理技术融合到当前服务器中，本研究将最终提高服务的可用性和抗压力和DoS攻击的生存能力。有关互联网服务的研究材料也将被灌输到本科和研究生分布式计算课程中。

Internet services have a characteristic of traffic burstiness. To protect premium customers from traffic surge without over-compromising the needs of basic clients, a stress-resilient server must be able to control the quality of multi-class requests in a coordinated way for fair and graceful performance degradation in stress conditions. The goal of this project is to develop an autonomic resource management system in the network edge in support of such service quality assurance models. The resource management system features three key innovations. First is an application-level QoS-aware resource management framework for multi-class quality assurance. Its novelty lies in the ability to providing guarantees of client-perceived end-to-end page-view response time in multi-tier web sites. Second is a 2D service differentiation model that formulates the requirements of session-based workload in both inter-session and intra-session dimensions and relates it with a revenue maximization objective in e-commerce applications. Third novelty is a model-free self-tuning feedback controller that regulates the process of resource allocation dynamically and achieves a high degree of control robustness in both long and short time scales.Edge servers are a critical building block of the Internet with profound impact on our economy and society. This research will advance discovery and understanding of the service quality assurance problems in servers under a stressed condition. By fusing the autonomic resource management technologies into current servers, this research will ultimately enhance the service availability and survivability to stress and DoS attacks. Research-based materials about Internet services will also be instilled into the undergraduate and graduate distributed computing curriculum.