Managing opportunistic and dedicated resources in a bi-modal service deployment architecture

• Authors: Shah Asaduzzaman
• Venue: Managing opportunistic and dedicated resources in a bi-modal service deployment architecture
• Year: 2007

With the emergence of service oriented computing, hosting platforms are becoming key elements of the Internet. One popular example of service hosting platform is Internet Data Center that relies on statically dimensioned centralized pools of server resources. Although direct control over the resources is a key advantage here, such platforms are often unable to handle highly varying workloads of many applications. On the other hand, popular peer-to-peer systems that opportunistically aggregate idle resources from widely distributed end-user computers demonstrate the huge potential of these public resources to build highly scalable, low cost and easily deployable platforms. However, due to the uncertain availability of these resources, it is hard to guarantee any performance for the deployed services on such peer-to-peer platforms. In this thesis, we introduced a new bi-modal architecture for a geographically distributed and cost-effective service hosting platform. The proposed architecture utilizes a combination of statically provisioned dedicated resource pools and widely available opportunistic public resources to provide quality assured services. The core idea is that through dynamic management of a combination of these two classes of resources, one can gain from the scalability of the public resources and achieve assured quality services by masking their unreliable behavior with the controlled performance of the dedicated resources. We have explored the combination in two different dimensions—bi-modal computing resources and bi-modal communication links. In the first case, a combination of a dedicated cluster of computers and idle capacities of user computers have been exploited to build a platform to serve compute intensive applications with response time guarantees. In the second case, a collection of geographically distributed dedicated servers inter-connected with a combination of dedicated links and variable capacity public network paths have been utilized to serve distributed stream processing applications that requires simultaneous allocation of computing and communication resources. In both cases, we have observed that by designing appropriate resource management policies, the combined sets of resources can be utilized to increase the overall resource utilization and throughput of the system as well as to increase the client satisfaction in terms of fulfillment of the service agreements.