SGER: Context-Aware Multi-Resource Management for High Service Availability on Internet Servers

背景互联网服务具有突发流量的特征。为了应对突发事件，保护有价值的用户免受流量激增的影响，具有压力弹性的服务器必须能够协调控制不同类别请求的质量，以保证在压力条件下公平而优雅地降低性能。这种质量保证技术还可以通过降低可疑请求的服务质量来保护服务器免受分布式拒绝服务攻击。该项目的目标是开发一种应用级多资源管理机制和一种协调的空间/时间请求调度策略，以实现流媒体服务器的压力弹性和高服务可用性。首先是一种谐波网络--I/O带宽分配方案。该方案不仅最大化了整体带宽利用率，而且提供了自管理和性能隔离功能，以保证不同请求类别之间的流比特率的质量间隔，并提高服务可用性。其次，提出了一种新的具有自校正反馈控制的请求调度算法，保证了请求相对于其启动延迟的公平和优雅的性能下降。自校正反馈控制处理互联网流量突发的影响，并在长时间和短时间范围内提供一致的质量保证。该项目的第三个创新之处在于一个协调的多资源管理策略，该策略决定何时以及如何调度流请求，以满足流码率和启动延迟的质量要求，同时最大化流服务的可用性。广泛影响当今的互联网服务经常容易受到流量激增的影响。这项研究将促进对压力条件下质量保证问题的发现和理解。通过将提出的压力恢复技术融合到流媒体服务中，本研究将最终提高服务的可用性和对压力和DDoS攻击的生存力。此外，该项目还包括一个完整的教育部分。有关互联网服务的研究材料将被灌输到本科生和研究生的分布式计算课程中。代表不足的学生也将被创新，以参与这一综合研究和教育项目。

BackgroundInternet services have a characteristic of bursty traffic. To deal with the burstiness and protect valuablecustomers from traffic surge, a stress-resilient server must be able to control the quality of different classes of requests in a coordinated way so as to guarantee a fair and graceful performance degradation in stress conditions.Such quality assurance technologies can also protect the server from distributed denial-of-serviceattacks by downgrading the service quality of the requests with suspicious intentions. The goal of this projectis to develop an application-level multi-resource management mechanism and a coordinated space/time request scheduling strategy for stress-resilience and high service availability on streaming servers.Intellectual MeritThe proposed resource management system features three key innovations. First is a harmonic network-I/O bandwidth allocation scheme. The proposed scheme not only maximizes the overall bandwidth utilization, but also provides self-management and performance isolation features to guarantee quality spacings in terms of stream bitrate between different request classes and to enhance the service availability. Second is a novel request scheduling algorithm with self-tuning feedback control that ensures fair and graceful performance degradation of requests with respect to their startup latency. Self-tuning feedback control deals with the impact of Internet traffic burstiness and provides consistent quality assurance in both long and short timescales. The third novelty of this project lies in a coordinated multi-resource management strategy that determines when and how to schedule stream requests so as to meet the quality requirements in terms of both stream bitrate and startup latency and meanwhile maximizing the streaming service availability.Broader ImpactTodays Internet services are often vulnerable to traffic surge. This research will advance discovery and understanding of the quality assurance problem in stress conditions. By fusing the proposed stress-resilient technology into streaming services, this research will ultimately enhance the service availability and survivability to stress and DDoS attacks. In addition, this project has an integral education component. Research-based materials about Internet services will be instilled into the undergraduate and graduate distributed computing curriculum. Under-represented students will also be innovated to participatein this integrated research and education project.