Intelligent Management of Distributed Rack-Based Blade Operations

This presentation will cover research addressing management and operations problems in distributed rack-based blade systems. Specifically, the focus will be on handling distributed configuration management of system-level parameters and optimizing job placement in a distributed system using thermal exhaust characteristics. I will cover not only why these particular issues are important to the community, but how statistical inference models can be used to determine correctness of configurations and provide information relevant to resolving conflicting parameters, and how unique approaches to thermal analysis can help reduce the overall heat footprint in rack-based clusters. Preliminary results from both focus areas will be discussed along with the issues tackled to date.