CSR: Medium: Energy-Efficient Architectures for Emerging Big-Data Workloads

点击翻译按钮获取中文摘要

In modern server architectures, the processor socket and the memory systemare implemented as separate modules. Data exchange between these modulesis expensive -- it is slow, it consumes a large amount of energy, and thereare long wait times for narrow data links. Emerging big-data workloads willrequire especially large amounts of data movement between the processorand memory. To reduce the cost of data movement for big-data workloads,the project attempts to design new server architectures that can leverage3D stacking technology. The proposed approach, referred to as Near DataComputing (NDC), reduces the distance between a subset of computationalunits and a subset of memory, and can yield high efficiency for workloadsthat exhibit locality. The project will also develop new big-data algorithms and runtime systems that can exploit the properties of thenew architectures.The project will lead to technologies that can boost performance andreduce the energy demands of big-data workloads. Several reports havecited the importance of these workloads to national, industrial, andscientific computing infrastructures. The project outcomes will beintegrated into University of Utah curricula and will play a significantrole in a new degree program on datacenter design and operation. ThePIs will broaden their impact by publicly distributing parts of theirsoftware infrastructure and by engaging in outreach programs thatinvolve minorities and K-12 students.