An Integrated Approach for Quality of Service in Cluster Networks

多处理器互连的设计主要是为了提供高吞吐量和低延迟的最佳努力流量。处理节点架构和通信软件也是如此。然而，随着这些系统（特别是集群）越来越多地用于交互式应用，如多媒体、web服务器和电子商务，可预测的通信性能，称为服务质量（QoS），已成为一个主要问题。除了低延迟和高吞吐量之外，许多这些应用程序还需要保证通信延迟和带宽。本研究的主要重点是为集群系统设计端到端通信基础设施，以支持具有不同QoS要求的不同应用。研究工作包括：(1)路由器架构设计与评估，以提供可预测的高性能；(2)网络接口和软件消息层支持按QoS要求向网络注入/弹出流量；(3)将这些能力传播到应用程序级别的CPU调度机制。研究建议将这三个组件无缝集成，以提供具有QoS保证的高性能通信体系结构。

Multiprocessor interconnects have been designed primarily to provide highthroughput and low latency for best-effort traffic. This is true forprocessing node architectures and communication software as well. However,with the increasing use of these systems (clusters in particular) forinteractive applications such as multimedia, web servers and electroniccommerce, predictable communication performance, called Quality of service(QoS), has become a major concern. Many of these applications requireguaranteed communication delay and bandwidth, in addition to low latencyand high throughput. The main focus of this research is to design anend-to-end communication infrastructure for a cluster system to supportdifferent applications with varying QoS requirements. The researchactivities include (1) router architecture design and evaluation to providepredictable and high performance; (2) network interface and softwaremessaging layer support to inject/eject traffic into/from the network asper QoS requirements; and (3) CPU scheduling mechanisms that propagatethese capabilities up to the application level. The research proposes toseamlessly integrate these three components to provide a high performancecommunication architecture with QoS guarantees.