Server-Division Planning Algorithm for Fault Tolerance in QoS-Aware Cluster System

C. Choi and S. Kim (Korea)


Cluster System, QoS-Aware System, SLA, DependabilityEvaluation, Server Division Planning


for critical information exchange, it is necessary to provide service level agreement mechanism (SLA) at the web server. So, cluster systems may be a viable solution for web server. Fig. 1 depicts the two-tier architecture of web server cluster. First, domain name server (DNS), routing switch and firewall serve as access tier. Routing switch parses the request's URL and routes it to a web server typically according to a load balancing policy of DNS. Also, DNS equipped with admission control component checks whether this request should be processed. Firewall provides basic functions such as intrusion detection. Next, virtual local area network (vLAN), a logically defined network, is used for dividing LAN into sub-networks [4]. vLAN can be implemented with inter-switch link (ISL) protocol for external tagging or 802.1Q protocol for internal tagging. Web servers in vlan1 serve as constant components processing static requests (S1, S2, ... , Sn) and those in vlan2 do as self motivated elements processing dynamic requests (D1, D2, ... , Dn). This paper presents server-division planning algorithm for providing QoS support in view of not only performance but also availability in web server cluster, which uses software rejuvenation dealing with inherent software faults. Additionally, we give a demonstration through the experiments under a realistic workload model. Complex software-intensive applications can be built with commercially available systems for cluster systems. For providing predictable quality of service (QoS), it is necessary to develop a solution about native properties of web; best-effort service and bursty nature. In this paper, we propose server division planning algorithm to control overloads in QoS-aware cluster system for web server. We first exploit the dependence among the requests by analyzing and predicting the reference patterns during the periods. Next, we analyze the availability in terms of performance using Markov modeling. Then, we propose server-division planning algorithm in web server cluster for QoS and demonstrate through numerical experiments under a realistic workload model.

Important Links:

Go Back