Power and performance evaluation of Memcached on the TILEPro64 architecture

Power consumption of data centers had become an important factor in the economy and sustainability of large-scale Web services. Researchers and practitioners are spending considerable effort to characterize Web-scale workloads and evaluating their applicability to alternative, more power-efficient architectures. One such workload in particular is the caching layer, which stores expensive-to-regenerate data in fast storage to reduce service times. In this paper we look at one such application, the Memcached key-value store, which is widely deployed at Facebook and other Web services, and one alternative architecture, the TILEPro64 many-core system-on-chip. We explore the performance and power characteristics of Memcached under a variety of workload variations, implementation choices, and communication protocol, and compare them to a traditional implementation on x86-based servers. Our results suggest that the TILEPro64 architecture can significantly outperform x86-based architectures in terms of throughput per Watt for the evaluated version of Memcached.

► The TILEPro64 architecture contains many low-power cores with an internal fabric. ► Porting Memcached to this architecture exhibits improved scalability over x86. ► This scalability translates to better absolute performance and lower power than x86. ► Furthermore, scaling to large-scale deployments shows even better performance per Watt.