Eliminating energy waste in memory systems by improving the software

计算机系统消耗大量能源。2005年，全球数据中心消耗了1520亿千瓦时，相当于整个墨西哥的消耗量！2010年，消费量增长了56%，预计将继续增长，以满足智能手机用户的数据需求，到2013年，智能手机用户的数量将达到13亿。数据中心的大部分能源都花在服务器上，而在服务器中，超过50%的能源用于CPU和内存。 这个项目将通过学习如何编写更好的软件来解决CPU和内存系统的能源效率问题。在现代多核系统上会发生能源浪费，因为程序花费太多时间等待数据从主存到达CPU。我们将通过消除内存层次结构中的冗余数据移动以及预测和“预先排序”未来的数据请求来减少能源浪费。 我们的项目将创建编写更快，更节能的应用程序的方法。这些技术将在现有的硬件上工作，并为构建未来的硬件提供见解。通过抢先体验我们的成果并将其整合到产品中，我们的加拿大行业合作伙伴将从竞争优势中受益。最后，我们的项目将培养大量的学生，使他们具备与未来经济相关的技能。

Computer systems use lots of energy. In 2005 data centers worldwide consumed 152 billion kWh - the same amount as consumed by the entire country of Mexico! The consumption rose by 56% in 2010 and is expected to continue growing to meet data demands of smartphone users, whose number will reach 1.3 billion by 2013. Most energy in data centers is spent on servers, and in servers, over 50% is used on CPU and memory. This project will address energy efficiency of CPU and memory systems by learning how to write better software. Energy waste on modern multicore systems occurs, because programs spend too much time waiting for data to arrive from main memory to the CPU. We will reduce energy waste by eliminating redundant movement of data across the memory hierarchy and by anticipating and "pre-ordering" future data requests. Our project will create methods for writing faster and more energy-efficient applications. These techniques will work on existing hardware and provide insight into building future hardware. By gaining early access to our results and integrating them in products, our Canadian industry collaborators will benefit from a competitive advantage. Finally, our project will train a large number of students, equipping them skills relevant for the future economy.