Question: Consider a node-by-node window scheme. In an effort to reduce the required buffering, the designers associated the windows with destinations rather than with virtual circuits. Assume that all virtual circuit paths to a given destination j use a directed spanning tree so that each node i≠ j has only one outgoing link for traffic to that destination. Assume that each node i≠ j originally has permits for two packets for j that can be sent over the outgoing link to j. Each time that node i releases a packet for j into its DLC unit on the outgoing link, it sends a new permit for one packet back over the incoming link over which that packet arrived. If the packet arrived from a source connected to i, the permit is sent back to the source (each source also originally has two permits for the given destination).
(a) How many packet buffers does node i have to reserve for packets going to destination j to guarantee that every arriving packet for j can be placed in a buffer?
(b) What are the pros and cons of this scheme compared with the conventional node-by-node window scheme on a virtual circuit basis?