Chapter 9

9.1	Why do you think both types of ICMP redirects ” network and host ” exist?
A:	When the ICMP standard was first specified, RFC 792 [Postel 1981b], subnetting was not in use. Also, using a single network redirect instead of N host redirects (for all N hosts on the destination network) saves some space in the routing table.
9.2	In the routing table for `svr4` shown at the beginning of Section 9.2, is a specific route to the host `slip` (140.252.13.65) necessary? What would change if this entry were removed from the routing table?
A:	The entry is not required, but if it is removed, all IP datagrams to `slip` are sent to the default router ( `sun` ), which then forwards them to the router `bsdi.` Since `sun` is forwarding a datagram out the same interface on which it was received, it sends an ICMP redirect to `svr4.` This creates the same routing table entry on `svr4` that we removed, although this time it is created by a redirect instead of being added at bootstrap time.
9.3	Consider a cable with both 4.2BSD hosts and 4.3BSD hosts. Assume the network ID is 140.1. The 4.2BSD hosts only recognize a host ID of all zero bits as the broadcast address (140.1.0.0), while the 4.3BSD hosts normally send a broadcast using a host ID of all one bits (140.1.255.255). Also, the 4.2BSD hosts by default will try to forward incoming datagrams, even if they have only a single interface. Describe the events that happen when the 4.2BSD hosts receive an IP datagram with the destination address of 140.1.255.255.
A:	When the 4.2BSD host receives the datagram destined for 140.1.255.255 it finds that it has a route to the network (140.1) so it tries to forward the datagram. To do this it sends an ARP broadcast looking for 140.1.255.255. No reply is received for this ARP request, so the datagram is eventually discarded. If there are many of these 4.2BSD hosts on the cable, every one sends out this ARP broadcast at about the same time, swamping the network temporarily.
9.4	Continue the previous exercise, assuming someone corrects this problem by adding an entry to the ARP cache on one system on the 140.1 subnet (using the `arp` command) saying that the IP address 140.1.255.255 has a corresponding Ethernet address of all one bits (the Ethernet broadcast). Describe what happens now.
A:	This time a reply is received for each ARP request, telling each 4.2BSD host to send the datagram to the specified hardware address (the Ethernet broadcast). If there are k of these 4.2BSD hosts on the cable, all receive their own ARP reply, causing each one to generate another broadcast. Each host receives each broadcast IP datagram destined to 140.1.255.255, and since every host now has an ARP cache entry, the datagram is forwarded again to the broadcast address. This continues and generates an Ethernet meltdown. [Manber 1990] describes other forms of chain reactions in networks.