Link-level vs end-to-end retransmission. Saltzer et al. [157] argue that a system with end-to-end error control needs no further error control. In this question we will investigate this argument. (a)...

Link-level vs end-to-end retransmission. Saltzer et al. [157] argue that a system with end-to-end error control needs no further error control. In this question we will investigate this argument.

(a) Consider an interconnection network that sends 1,024-bit packets divided into 64-bit flits. Each packet travels an average of 10 hops, and each link has a BER of 10⁻⁵. Link bandwidth is 1 Gbit/s and router latency is 20 ns per hop. The system checks each packet at the destination and requests retransmission if the packet is in error. To simplify analysis, assume that the error control code detects all possible errors and that delivery of the ack or nack message is error free. What is the probability of a packet being received in error? What is the resulting throughput and average latency for a packet?

(b) Now consider the same system, but with link-level retry in addition to endto-end retry. Each link checks each flit of each packet as it is received and requests a retry if the flit is received in error. How does link-level retry affect the throughput and average latency?