Sven Graupner Hewlett-Packard Laboratories Vadim Kotov Carnegie Mellon University Artur Andrzejak Zuse Institute Berlin Holger Trinks Santa Clara University 36 JULY • AUGUST 2003 Published by the IEEE Computer Society 1089-7801/03/$17.00©2003 IEEE IEEE INTERNET COMPUTING The Grid Grows Up Service-Centric Globally Distributed Computing An automated service demand–supply control system can improve a large-scale grid infrastructure comprising a federation of distributed utility data centers. T he growing complexity of large- scale systems raises concerns about their effectiveness and man- ageability. New approaches to system design, use, and management address these concerns through the aggregation and consolidation of system and applica- tion components into larger building blocks; systematic and standard ways of integrating such systems and communi- cating between them; sharing of distrib- uted resources; and automated system management and operation control. Due to their scale and dynamism, large-scale systems cannot provide global informa- tion about resource availability and ser- vice demand. Decision-making algo- rithms thus need to deal with partial information, yet provide good approxi- mations of localized assignment solu- tions. Moreover, they need to make deci- sions within the duration of the overall control cycle of an automated resource demand–supply control system. Various control cycles occur simultaneously, ranging from short-term reactive response (range of seconds to minutes) to mid-term system rebalancing (hours to days) to longer-term adjustments and system evolvement (months). Usually, decision algorithms implement trade-offs between the time spent identifying a solution and the quality of that solution. Two of the most prominent and prac- tical approaches to solving these prob- lems are utility computing and grid com- puting. Adopting the utility computing approach, Hewlett-Packard developed the Utility Data Center (UDC; see www.hp. com/go/hpudc/) to consolidate comput- ing resources. Virtual data centers 1 con- solidate the resources of a federation of distributed UDCs into virtual resources that users, user applications, and user ser- vices can share using grid-type mecha- nisms, thus significantly reducing deploy- ment and operation costs. We propose an architecture for such an