CSP 214 Written Homework 4 Solutions - Fall, 1997

1. Assume that node A is already using the address, when node B sends
out the "duplicate" ARP request. The neighboring caches will become
temporarily polluted since the ARP request from B host will cause all
adjacent node's to replace A's link-layer in their caches with B's
address. Thus, even if B chooses not to use the address after getting
a response to its ARP request, other nodes will send traffic intended
for A to B.  This will last until either node A sends out a new ARP
request (which will update all caches at once) or when replying to
individual nodes that ARP for A's address. So yes, by the time B
realizes that the IP address it wants to use is a duplicate, other
nodes will have incorrect entries in their caches.

2. If the address is nonexistent, the router on the remote network
would have to broadcast an ARP request for that address. Being unable
to receive a reply, the router would possibly retry for some times,
which generates further broadcast traffic.

3. The result will not be catastrophic. IP may process the packet
incorrectly, but the result of that processing is unlikely to do any
serious damage.  For example, a datagram might be delivered to the
wrong endpoint (if, for example, the destination address field became
corrupted). However, higher layer protocols are likely to reject such
packets due to higher layer checksums, etc.

Note also that in IPv4, checksums reduce the probability that errors
will slip byundetected. However, they do not reduce the probability to
0. Thus, if an erroneous packet could really cause serious problems,
then even IPv4 wouldn't be immune even with the checksums it uses.

4. If a unicast IP packet is sent using link-layer broadcast (or
multicast), it is likely that the same packet will be received by
multiple nodes. If all of those nodes generated ICMP messages at the
same time, a flood of messages (e.g., a "broadcast storm" -- if all
those error messages triggered the generation of ARP messages) would
result. Having one message trigger an explosion of response messages
is generally considered to be bad  thing.

5. Assigning a single CIDR address for NCREN is not possible as 60 is
not a power of two and hence the 60 B addresses cannot be represented
with a common prefix without also covering networks that do not belong
to NCREN. 

6. A host can connect to more than one physical link (network). Since
an IP address is comprised of network part and the host part, multiple
IP address is necessary for multiple network interfaces. Otherwise the
redirect mechanism cannot work correctly. Also, multicast routers use
different IP addresses to tell from which network a multicasting
packet is coming in order to suppress duplicated packets. A host that
responds to a multicast message sent to 'all host on the network' must
know from which interface it receives the message.

7. No, the placement is not arbitrary. A different version might have
completely different packet header layout (e.g., have a different
header length), so fixing the version field as the first field of the
header makes it possible to parse the version field without having to
understand any other field.f

8. When the IP layer receives the fragmented packets without the
fragment more bit set, it will just treat them as unfragmented
packets, strip off the IP header and relay the remaining data up to
the appropriate higher-layer protocol. The first fragment, which
contains the TCP segment header will be handed to TCP, but the TCP
layer will almost certainly reject it because the TCP checksum will
not be correct. Subsequent fragments will likely be rejected for
similar reasons.

9. No. ARP is used to determine the link-layer address of a unicast IP
address. When sending IP multicast traffic, the Ethernet address of an
IP multicast packet is determined algorithmically without sending
additional packets on the wire. In the Ethernet case, for instance,
the sender simply ORs the rightmost 23 bits of the IP multicast
address to the well-known Ethernet address 01.00.5e.00.00.00. The IP
multicast packet is then sent to the resultant link-layer addresss.