A cloud collaboration system with active application control scheme and its experimental performance analysis

Abstract

The scientific collaboration conducted in cyberspace needs a suitable platform that can enable users to rapidly build customized solutions around their content items. Also, different QoSs should be guaranteed to users in according to their specifications. In this thesis, on the base of an active-content based collaboration platform, we propose the MapChem Cloud Collaboration System (MCCS) for cloud collaboration in the field of chemistry/pharmacy. The core module in MCCS is the MapChem-Broker which bridges the application policy on the user’s side and the resource policy on the service provider’s side. By deploying a policy-based workflow scheduling mechanism and a virtual machine (VM) management scheme, the MapChem-Broker can provide QoS guarantee by management of users’ SLAs. The policy-based workflow scheduling mechanism can translate users’ QoS requirements into corresponding expected resource capability parameters and then the VM provisioning scheme will use these resource parameters to allocate a proper resource to job processing. Additionally, when the number of job requests changes, the Adaptive Resource Collaboration Framework (ARCF) module within the MapChem-Broker can adjust the system’s resource provision capability through an auto-scaling scheme. For the experiment evaluation, we show the integrated services in Chemical application for users to compose and submit to MCCS their own workflow applications and we measure the system performance. Through the experimental evaluation, we show that the VM resource provisioning scheme outperforms the physical machine provisioning scheme in the view of cost-performance. We also experiment on a cloud-based testbed to show that the proposed scheme can realize auto-scaling in the VM management. Moreover, we compare the proposed scheme and an existing scheme - Dynamic Scaling Scheme (DSS) for VM resource provisioning in our cloud-based testbed, in terms of their performance, cost, and cost-performance. Results s...