Experimental study of router buffer sizing

直到最近，人们还普遍认为互联网路由器需要大缓冲区。今天的商用路由器具有巨大的数据包缓冲区，通常存储数百万个数据包，假设大的缓冲区会导致良好的统计多路复用，从而有效地利用昂贵的长途链路。最近，我们（在斯坦福大学的时候）展示了一个可以用更小的缓冲区构建网络——在每个路由器上只有几十个数据包缓冲区，代价是100%的链路利用率。如果这是正确的，将会产生有趣的实际结果。它可以消除构建光分组交换路由器的一个主要障碍。最近的技术进步（UCSB）也使光学FCFS数据包缓冲区成为可能，它可以在一个集成的光电芯片中容纳几十个数据包。更大的全光缓冲器仍然不可行，除非使用笨重的光纤线轴（只能实现延迟线，而不是真正的FCFS数据包缓冲器）。我们感兴趣的是探索在每个路由器中只有几十个光包缓冲区的可操作全光网络的可行性，并研究我们的理论结果所支持的边界条件。以前在这个方向上的尝试包括在有限的实验室环境中进行的实验。有限数量的路由器和交换机（通常是一两个），固定的流量模式，琐碎的哑铃状拓扑，……这类实验有一些缺点。在这个项目中，我们提出了一组更广泛的实验来探索我们的缓冲区大小结果的有效性边界。为此，我们将使用斯坦福大学最近设计的4端口千兆以太网路由器，称为可控和可观察缓冲区（COB）路由器。COB路由器建立在NetFPGA平台上，具有可配置的缓冲区大小，并且能够以16ns的时间分辨率（小于一个数据包的到达时间）监视和报告缓冲区占用情况。除了在我们的实验室中构建大型COB路由器拓扑的可能性之外，这将使我们比任何商用路由器具有更多的控制和可观察性。

Until recently, Internet routers were widely believed to need large buffers. Commercial routers today have huge packet buffers, often storing millions of packets, under the assumption that large buffers lead to good statistical multiplexing and hence efficient use of expensive long-haul links. Recently, we (while at Stanford) showed that one can build a network with much smaller buffers still---with only a few dozen packet buffers in each router, at the expense of 100% link utilization. There are interesting practical consequences if this is correct. It could remove one major roadblock to building optical packet-switched routers. Recent advances in technology (at UCSB) also make possible optical FCFS packet buffers that can hold a few dozen packets in an integrated opto-electronic chip. Larger all-optical buffers remain infeasible, except with unwieldy spools of optical fiber (that can only implement delay lines, not true FCFS packet buffers).We are interested in exploring the feasibility of an operational all-optical network with just a few dozen optical packet buffers in each router, and studying the boundary conditions where our theoretical results hold. Previous attempts in this direction include experiments performed in limited laboratory settings. Limited number of routers and switches (usually one or two), fixed traffic patterns, trivial dumbbell-shaped topologies, ... are a few of the shortcomings of such experiments. In this project, we propose a broader set of experiments to explore the boundaries of the validity of our buffer sizing results. To this end, we are going to use 4-port Gigabit Ethernet routers, called Controllable and Observable Buffer (COB) routers, recently designed in Stanford University. COB routers are built on the NetFPGA platform, have configurable buffer sizes, and are able to monitor and report the buffer occupancy with a time resolution of 16ns (less than the arrival time of one packet). This will allow us more control and observability than any commercial router, in addition to the possibility of building large topologies of COB routers in our lab.