CSR: Small: Rethinking the OS/Network Boundary

点击翻译按钮获取中文摘要

Operating systems and networks developed largely independently, due tothe nature of their development -- operating systems were developed bygroups within organizations/companies, while networking requiringinteroperability was developed across organizations. As a result ofthis development history, the boundary between the networkingsubsystem and the rest of the operating system has remained largelyopaque, with little information regarding the performance or progressof transfers being communicated to applications. This researchaddresses this limitation by exploring a richer boundary interfacethat allows programs to see the ongoing behavior of network transfersand to control their operation at a fine granularity. These interfaceswould augment, rather than replace, the existing interfaces, so theportions of programs that are satisfied with the current interfacescould continue unmodified, and developers could change only theportions that can benefit from richer information. This research islargely experimental in nature, and will combine the design andimplementation of new interface abstractions along with thedevelopment of new or modified programs to take advantage of theseinterfaces. In addition to the design work associated with theinterfaces, additional criteria for success include examining theperformance and delay behavior of modified applications, and theeffort needed to modify such applications. The expected results fromthis work include applications that can more precisely control theirnetworking behavior, reducing memory and processor usage, improvingenergy efficiency, and improving delay behavior. High-performance ortime-sensitive applications, such as networked video, should benefitfrom this research.