Application-level Performance Prediction Across Multi-Core Processor Configurations

Multiple cores, simultaneous multi-threading, and onchip caches are important processor configurations for servers that support Internet applications. However, certain settings of the processor configurations may yield lower performance while consuming more power. Since processors account for a significant portion of the energy budget, up to 50-60% in low-end servers [4], inefficient processor configurations can be costly. For example, a data center reduced its power consumption (without affecting performance) by 10% and saved $70,000 per-year in unnecessary energy costs [5]. Recurring energy costs are expected to become the primary cost for data centers by the end of the decade [4].