Network Programming

Systems Programming Exam 2007 Sample Answers
The answers underneath are VERY short sample answers of the June 2007 Exam. These might come in handy for revision, to check your answers or to consider before applying for a revision of paper. Note that sometimes more than one version of an answer might be correct, I am just listing one of these versions. Also I am not going in the depth expected during the exam. I am just highlighting the main concepts expected. I will not answer to personal queries requesting information on their particular answer since that would break confidentiality. Also please note that my results have already been submitted to administration.
(a) Obviously its 576+38=614 bytes. (b) Since an Ethernet frame is 1500 bytes, then max data allowed in ethernet with encapsulated IP is 1550-20-38 = 1442. Thus max throughput is 1442/1500 * 100 = 96.13Mbps. (c) Traceroute, and it works by sending an IP or ICMP packet with TTL equal to 1. Then when one gets an ICMP back from a router (TTL expired or something similar), one sends another packet with TTL=2 and so on. (d) Start sending one segment, when you receive acknowledgement, send two, wait for ACKs, then send 4, and so until you start losing packets. This algorithm allows transmission rate to be very close to maximum allowed by network without causing congestion. (a) (b) ARP is used to map an IP address to a MAC address. PC keeps ARP cache, if IP is not matched, a broadcast is sent on the Physical Layer with ARP request for MAC required by specifying the IP address. The machine with that IP will reply back with its own MAC address. (c) Distance Vector sends its vector tables periodically to all other connected routers. The vector table consists mainly of a destination route, the router to send a packet to and the cost of reaching the destination. It can get into trouble when redundant links exist because of slow convergence time and also due to counting to infinity. Its very efficient computationally and does not have moments of network saturation. Link state works by keeping internally a view of the network with itself as the route. It also stores the cost of each link (unlike distance vector which just stores total cost to destination). For each packet it then calculates the shortest path to the destination. Unlike distance vector, when a network change occurs, an advertisement of the change is sent immediately. Each router will do the same so one can get flooding of packets on any change (not a nice thing when one has just lost an important link and OSPF is just flooding you with packets about itself). Yet convergence is much quicker than distance vector and it can deal perfectly with redundant links. One uses distance vector in smallish networks which are well structured and do not have redundant links. Also where bandwidth is not such a problem. Link state networks is used is more complicated scenarios where reliability is paramount and one can afford the router initial cost. (a) Since vector routing algorithms rely on periodic updates, having a change trigger the updates will make convergence much quicker. Thus any changes will be propogated very quickly throughout the whole network. (b) Something like split horizon that should never send information back to an interface from which that information was learnt in the first place. Also split horizon with poison reverse is very nice where you do advertise a route from whom you just learnt it, but you mark that route as unreachable through yourself. Also if you receive a route which is unreachable, you immediately delete that entry from your table. Limiting the number of hops does not really solve the counting to infinity, it just limits the erroneous paths from keeping on increasing their cost (a bit like saying that giving up on educating a student solves the problem of low education levels). In turn, the router might believe a path is unreachable when it actually is. NAT is one way. You can do this either by having a pool of IP addresses and when a connection is made, you assign a mapping from public IP to private IP at the NAT server. Even better you can use port mapping (a form of NAT), where you map a public IP address AND port number to the private IP address and the original port number. Such a technique only works by keeping a table containing private IP/private port/public port. The public IP is only stored when using multiple public IP address but then you would need a multi-homed PC! Limitations including IPSEC and that private IP PCs cannot be servers. (d) It sends more than one packet before expecting an acknowledgement, thus it uses up the network much more efficiently (is latency insensitive). Also TCP windowing is more robust to lost acknowledgement packets.

Systems Programming Exam 2007 Sample Answers

The answers underneath are VERY short sample answers of the June 2007 Exam. These might come in handy for revision, to check your answers or to consider before applying for a revision of paper. Note that sometimes more than one version of an answer might be correct, I am just listing one of these versions. Also I am not going in the depth expected during the exam. I am just highlighting the main concepts expected. I will not answer to personal queries requesting information on their particular answer since that would break confidentiality. Also please note that my results have already been submitted to administration.

- (a) Obviously its 576+38=614 bytes.
- (b) Since an Ethernet frame is 1500 bytes, then max data allowed in ethernet with encapsulated IP is 1550-20-38 = 1442. Thus max throughput is 1442/1500 * 100 = 96.13Mbps.
- (c) Traceroute, and it works by sending an IP or ICMP packet with TTL equal to 1. Then when one gets an ICMP back from a router (TTL expired or something similar), one sends another packet with TTL=2 and so on.
- (d) Start sending one segment, when you receive acknowledgement, send two, wait for ACKs, then send 4, and so until you start losing packets. This algorithm allows transmission rate to be very close to maximum allowed by network without causing congestion.
- (a)
- (b) ARP is used to map an IP address to a MAC address. PC keeps ARP cache, if IP is not matched, a broadcast is sent on the Physical Layer with ARP request for MAC required by specifying the IP address. The machine with that IP will reply back with its own MAC address.
- (c) Distance Vector sends its vector tables periodically to all other connected routers. The vector table consists mainly of a destination route, the router to send a packet to and the cost of reaching the destination. It can get into trouble when redundant links exist because of slow convergence time and also due to counting to infinity. Its very efficient computationally and does not have moments of network saturation.
  Link state works by keeping internally a view of the network with itself as the route. It also stores the cost of each link (unlike distance vector which just stores total cost to destination). For each packet it then calculates the shortest path to the destination. Unlike distance vector, when a network change occurs, an advertisement of the change is sent immediately. Each router will do the same so one can get flooding of packets on any change (not a nice thing when one has just lost an important link and OSPF is just flooding you with packets about itself). Yet convergence is much quicker than distance vector and it can deal perfectly with redundant links.
  
  One uses distance vector in smallish networks which are well structured and do not have redundant links. Also where bandwidth is not such a problem. Link state networks is used is more complicated scenarios where reliability is paramount and one can afford the router initial cost.
- (a) Since vector routing algorithms rely on periodic updates, having a change trigger the updates will make convergence much quicker. Thus any changes will be propogated very quickly throughout the whole network.
- (b) Something like split horizon that should never send information back to an interface from which that information was learnt in the first place. Also split horizon with poison reverse is very nice where you do advertise a route from whom you just learnt it, but you mark that route as unreachable through yourself. Also if you receive a route which is unreachable, you immediately delete that entry from your table. Limiting the number of hops does not really solve the counting to infinity, it just limits the erroneous paths from keeping on increasing their cost (a bit like saying that giving up on educating a student solves the problem of low education levels). In turn, the router might believe a path is unreachable when it actually is.
- NAT is one way. You can do this either by having a pool of IP addresses and when a connection is made, you assign a mapping from public IP to private IP at the NAT server. Even better you can use port mapping (a form of NAT), where you map a public IP address AND port number to the private IP address and the original port number. Such a technique only works by keeping a table containing private IP/private port/public port. The public IP is only stored when using multiple public IP address but then you would need a multi-homed PC!
  Limitations including IPSEC and that private IP PCs cannot be servers.
- (d) It sends more than one packet before expecting an acknowledgement, thus it uses up the network much more efficiently (is latency insensitive). Also TCP windowing is more robust to lost acknowledgement packets.