Cluster Server Design Using Load Balancing and High Availability to Improve web server Capabilities in Electronic Learning (ELearning) Servers

• Main Author: KRESNA A - NIM: 23515052 , IQSYAHIRO
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/28103
• Institution: Institut Teknologi Bandung

<p align="justify">Web server technology is already a major requirement for web-based service providers, ranging from giant companies like Google and Facebook to local service providers such as Penerimaan Peserta Didik Baru (PPDB) dan Electronic Learning (E-Learning). The computing ability of a web server depends on the number of users accessing the system over a period of time. There are several computing-related options for this web server, including using an expensive and hard-to-reach high end single server or a cluster server that can be built from a number of low-cost that easy-to-run Desktop PCs, including PPDB committee in various regions in Indonesia. <br /> <br /> <br /> This research reports the comparative results between the performances of a high end single server with a cluster server consisting of a number of Desktop PCs with several alternative configurations. In this research, there are seven indicators measured, availability, throughput, transaction rate, lost, successful transaction, response time, and longest transactions. <br /> <br /> <br /> In a test with 100000 requests using eight servers with load balancing and high availability with least connection algorithm on the cluster server obtained an average value of availability of 100%, the average throughput value of 12.32 Mbps, the average transaction rate of 751,861, the value the average request lost 0 request, the average successful transaction value of 100000 requests, the average response time of 1.32 seconds, and the average longest transactions of 3.926 seconds. <br /> <br /> With the cluster server configuration, of the seven indicators measured, the cluster server excels across all performance indicators, except the response time value. In addition, the cluster server has another advantage that is the ability of mirroring that if one of the computing elements die, the service can still run even though its performance decreased. The average downtime time on failover testing on cluster server is 6.787 seconds, while on high end single server the service is totally down. <br /> <br /> <br /> The main conclusion of this case study research, an IBM x3750 high end single server with 16 GB of RAM and 24 internal CPUs (Dual Xeon) equivalent with a cluster server consisting of seven to eight HP Pavilion Desktop PCs with 4 GB of RAM and 4 internal CPU (Intel i3).<p align="justify">