体系结构内可编程数据平面方法

发布时间：2018-07-31 20:36

【摘要】：随着互联网与云计算的发展,越来越多的应用被从本地迁移到云端,这些应用最终被运行在共享的数据中心.受到数据中心应用复杂并且需求多变特征的影响,传统体系结构中的部分硬件部件(如共享末级缓存、内存控制器、I/O控制器等)固定功能的设计不能很好地满足这些混合多应用的场景需求.为满足这类应用场景的需求,计算机体系结构需要提供一种可编程硬件机制,使得硬件功能能够根据应用需求的变化进行调整.提出了一种可编程数据平面方法:通过在现有硬件部件中增加可编程处理器,使用执行固件代码的方式对硬件的请求进行处理,并通过更新数据平面处理器固件的方式实现硬件功能的扩展.该方法在FPGA原型系统中进行验证,其结果表明,该方法并没有给系统性能带来严重的影响,只使用有限的资源即可为硬件增加更为灵活的可编程能力,使其能够适应应用需求复杂多变的场景.
[Abstract]:With the development of the Internet and cloud computing, more and more applications are migrated from local to cloud, and these applications are eventually run in shared data centers. Affected by the complexity of data center applications and the changing characteristics of requirements, some of the hardware components in traditional architectures, such as shared last level caching, The design of the fixed function of the memory controller / I / O controller can not well meet the needs of these mixed multi-application scenarios. In order to meet the requirements of this kind of application scenarios, the computer architecture needs to provide a programmable hardware mechanism, so that the hardware functions can be adjusted according to the changes of application requirements. In this paper, a programmable data plane method is proposed. By adding a programmable processor to the existing hardware components, the hardware request is processed by executing firmware code. The hardware function is extended by updating the firmware of the data plane processor. The method is validated in the FPGA prototype system. The results show that the method has no serious impact on the system performance. Using only limited resources can increase the programmable ability of the hardware. So that it can adapt to the application needs of complex and changeable scenarios.
【作者单位】：中国科学院计算技术研究所;中国科学院大学;
【基金】：国家“九七三”重点基础研究发展计划基金项目(2011CB302500) 国家自然科学基金项目(61420106013,61221062,61202062)~~
【分类号】：TP303

【参考文献】