6.033 Reading Report IV
Chi a (Janet) Wu

Contrast Ethernet's scheme for packet routing on a non-switched ethernet and the packet routing scheme presented in the "High Speed Switch Scheduling for LAN" paper.

The major aspect in which Ethernet differs from other communication systems for carrying digital data packets, namely the high speed switch scheduling method, is the scheme for packet routing. Each offers advantages over another even though the high speed switch network proves decisive in shaping the future of local-area networking technology because it creates a brand-new class of distributed networks that can be more closely coupled than ever.

The nature of Ethernet's topology is based on a single broadcast multi-access channel. Since in 1976 it was not yet cost-effective to implement a reliable intermediate 'router' which would require redundant and dynamic connections, Ethernet assures reliability by decentralizing control. A sender station with control to the 'ether' is responsible for preventing/recovering from problems such as packet contention and physical errors. A station recovers from a detected collision by abandoning the attempt and retransmitting the packet after some dynamically chosen interval. A degree of cooperation among the stations is required to share the ether fairly, otherwise a station is capable of reducing efficiency to virtually zero by not adjusting its retransmission interval with increasing traffic or by sending very large packets. This way failure of a single active element will not propagate to affect the communications of other stations. Its generality allows potential for convenient gradual extension without side effect of incommensurate scaling. However, despite all its modular simplicity, Ethernet has an ultimate constraint on performance- the need for the entire channel during a single transmission.

High Speed Switch Scheduling, on the other hand, is built based on the recent advent of CMOS and FPGA technologies, which makes possible the implementation of a dynamic router such as a banyan switch. Packets placed into a banyan network are automatically delivered to the correct output based solely on the information in the cell header. A Batcher sorting network used in combination with a normal banyan network would guarantee that a packet may be sent from any input to any output provided no two packets are destined for the same output. In the case where more than one packets are destined for the same output, the Batcher schedules one of them to be passed to the Banyan and places the rest in a queue waiting for the Banyan to complete current transmission. Hence scheduling and queuing are mandatory issues. Internal buffers is preferred over retransmission because it allows the sender to transmit to another destination while its previous packet is still in the buffer. This indicates a potential for a later packet to arrive early. To summarize, High Speed Switching network has the potential for lower latency by shortening path lengths and eliminates the need to acquire control over the entire network in order to begin transmitting. The high availability of channels and the existence of multiple paths between hosts are results of an asynchronous router tightly organized to maximize performance.

Overall, Ethernet and High Speed Switching are two distinct forms of distributed system. Even though the packet delay for both systems becomes unbounded as utility increases, this rate of increase is much higher for Ethernet. The parallelism and flexible approach of High Speed Switching is a concluding factor for the Switch to support a large number of hosts. Ethernet, however, is capable of offering similar performance at less cost as long as the number of hosts is moderate. In the near future, the majority of small networks would be likely to continue to use Ethernet.