Proportional Delay Differentiation Service in Web Application Servers: A Feedback Control Approach

2009 International Conference on Information Engineering and Computer Science

et al. 2009

Self Cite

The design of Web QoS based on feedback control has an imperfect real-time because of the offline identification. This paper presents a self-tuning control framework to provide proportional delay differentiation guarantees on Web Server. The approach updates the model and controller parameters based on the variations of object model to reduce system error and optimize the performances through an online identification. The Matlab simulation and network tests show that the controller has some advantages compared with the traditional methods and could maintain the relative delay guarantees well even in poor network environment.

Section: Related Workmentioning

confidence: 99%

A Self-Tuning Control for Web QoS

Dai

2009 International Conference on Information Engineering and Computer Science

et al. 2009

Self Cite

“…The pseudo-random sequence is generated as Equation (10) and the relationship of ( ) k ε and ( 1) The experiment lasts 30 minutes with the sampling time 10 s T = and we get 150 sets of effective data. Recursive least square (RLS) estimate algorithm is used to get parameter θ [8]. Suppose N θ is got from the former N m + sampled data, …”

Section: A System Identification and Testingmentioning

confidence: 99%

Proportional Hit Rate in Caching Service: A Feedback Control Approach

2009 International Symposium on Computer Network and Multimedia Technology

et al. 2009

Self Cite

As the heterogeneity of Web clients increasing, the differentiated service (DifferSer) becomes an important issue especially for e-commerce Web site. Web caching as a key accelerator on the internet plays an important role in alleviating the client-perceived delay. Based on feedback control theory, this paper proposes an isolated cache model and designs a storage controller to achieve proportional hit rate. Through system identification, the linear model of isolated cache is identified to design controller. At each sampling time, by dynamically reallocating cache spaces between different web objects, the controller could guarantee the priority Web objects a higher hit rate and the proportions of hit rates between different kinds of Web objects constant. The experimental results demonstrate the proposed approach achieves the proportional hit rat guarantee under GDSF, LRU and LFU cache replacement algorithms.

“…The HTTP requests are classified into different business flows (1...N ). By adjusting the quotas of every class, the performance isolation and differentiation service are achieved (see [8,14]). The initial ideal of PDDS-LB is a unified scheduling of all back-ends' processings/threades and a fair consumption of different servers' resource to the same flows.…”

Section: Pdds-lb Architechurementioning

confidence: 99%

“…The residue delay ℓ i is proportional to the queue length n i,j and inverse proportional to the number of service threads, i.e. τ i,j (see in [8,14]). For the clients supported HTTP 1.1 , a Web session is always a sequence of HTTP requests on a consistent TCP connection as a manner of Pipeline, so there is…”

Section: Maximum Idle Firstmentioning

confidence: 99%

The Research of Differentiated Service and Load Balancing in Web Cluster

2014

INT J COMPUT COMMUN

Differentiated service, as a key solution to meet the heterogenicity of Web clients' QoS requirements, has been widely used to optimize the server utilization without over-providing resources. Based on the relative differentiated service, this paper treats the application of proportional delay as a optimal control problem, and focuses on the cluster-side architecture improvement as well as QoS controller design. A load balancing Web cluster architecture supported differentiated service is proposed and implemented. By system identification and resource optimal control, the frontend dispatcher could adjust the resource quotas assigned to different classes in every single back-end server, and Multi-class based Maximum Idle First load balancing strategy is designed to ensure a fair resource consumption among back-end nodes. As a result, the end-to-end delay is controlled and proportional delay is guaranteed. The experiments demonstrate that no matter using Round-Robin, Least Connection Scheduling or Maximum Idle First load balancing strategy, the proposed resource optimal controller could hold the relationship among different classes. Compared to Round-Robin and Least Connection First Scheduling, Maximum Idle First strategy increases the cluster throughput by 33% and reduces the average delay by 21%.