Uppnå Prestanda för Beräkningsintensiva Tjänster i SaaS

Abstract: Enterprise business applications are following a rapid trend to move their solution model from specic organizational users to global access. This creates opportunities for organization to expand geographically with partners and resellers but it also increases simultaneous user requests. These solution models are commonly based on many data-set states and a web applications that perform multiple tasks in its work-ow including compute intensive requests to separate Compute Intensive Services (CIS). This research is based on these solution models and a special type of CIS that build and reuse in-memory cache to reduce response latency. Performance factors like additional simultaneous requests and cache building requests can increase response latency if not enough CIS are available to handle load peaks. Additional compute services can be added to the infrastructure but such solutions increase cost and these additional services are not required all the time. Main goal of this research is to study and design an architecture to achieve cost-e ective performance for solution model of CIS. First, a study have been performed on dedicated servers approach, to nd impact of these performance factors. Next, a prototype Software as a Service (SaaS) architecture has been presented which detects and reduces load peaks created by performance factors. SaaS architecture has been designed by using cloud computing products of Amazon Web Services (AWS). Few supplementary components have been identied and developed during research to overcome shortcomings of standard cloud products. It aims to reduce load peaks with scalability and elasticity. Experiments have been performed on SaaS architecture to and its advantages and limitations for solution model of CIS. An essential part of this research are two solution proposals, which are based on designed SaaS architecture. First solution proposal has been made for multi-tenant architecture because multi-tenancy can help to enhance cost-effective performance. The second solution proposal has been made to achieve low latency response by optimizing usage of in-memory cache. This optimization can help enterprises to change data-set states more often and achieve predictable low latency. It also adds exibility in SaaS architecture to reduce number of required servers.