Profile-driven Component Placement for Cluster-based Online Services Christopher Stewart, Kai Shen, Sandhya Dwarkadas, Michael L. Scott Department of Computer Science, University of Rochester Jian Yin IBM T. J. Watson Research Center The growth of the Internet and of various intranets has spawned a wealth of online services, most of which are implemented on local-area clusters using remote invocation (e.g., RPC/RMI) among manually placed application components. The placement problem can be a significant challenge for large scale services, particularly when application resource needs are workload dependent. This paper describes our initial work on automatic component placement, with the goal of maximizing overall system throughput. The key idea behind our approach is to construct (off-line) a mapping between input workload and individual component resource consumption. Such mappings, called component profiles, are then employed to support high-performance placement. We describe the basic design framework and present preliminary results on a J2EE-based online auction benchmark. Our results suggest that profile-driven tools can indeed identify placements that achieve near-optimal overall throughput.