The Design, Optimization, and Performance of an Adaptive Middlware Load Balancing Service Insert Name, ******** Say Something Nice :-) ************* Ii

The Design, Optimization, and Performance of an Adaptive Middlware Load Balancing Service by Ossama Othman Master of Science in Electrical and Computer Engineering University of California, Irvine Professor Douglas C. Schmidt, Chair Distributed object computing (DOC) middleware is increasingly used as the infrastructure for applications with stringent quality of service (QoS) requirements, including scalability. One way to improve the scalability of distributed applications is to balance system processing load among multiple servers. Load balancing can help improve overall system scalability by ensuring that client application requests are distributed and processed equitably across groups of servers. Earlier generations of load balancing middleware services were simplistic since they only addressed specific use-cases and environments. These limitations made it hard to use the same load balancing service for anything other than a small class of distributed applications. This lack of generality forced continuous redevelopment of application-specific load balancing services. Not only did this redevelopment increase distributed applications deployment costs, but it also increased the potential of producing nonoptimal load balancing implementations since time-proven load balancing service optimizations could not be reused directly without undue effort. This thesis presents the following contributions to research on load balancing techniques for DOC middleware: 1. It describes deficiencies with common load-balancing techniques, such as introducing unnecessary overhead or not adapting dynamically to changing load conditions. 2. It presents a novel adaptive load balancing service called that can be implemented efficiently using the capabilities of CORBA, which is a widely used, standards-based DOC middleware specification. 3. It explains how alleviates existing middleware load balancing services limitations, such as lack of server-side transparency, centralized load balancing, sole support for stateless replication, fixed load monitoring granularities, lack of fault tolerant load balancing, non-extensible load balancing algorithms, and simplistic replica management. 4. It discusses the forthcoming OMG Load Balancing and Monitoring specification, which is based on the research conducted for this thesis. 5. It describes the key design challenges faced when integrating the load balancing service in the The ACE ORB (TAO) and how these challenges were resolved by applying patterns. 6. It presents the results of benchmark experiments that empirically evaluate different load balancing strategies by measuring the overhead of each strategy and showing how well each strategy balances system load. Professor Douglas C. Schmidt Dissertation Committee Chair