Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 1 ECSE-6600: Internet Protocols Exam 2 Time: 80 min (strictly enforced) [Hint: spend time roughly in proportion to the points allocated to questions] Points: 50 YOUR NAME: Be brief, but DO NOT omit necessary detail {Note: Simply copying text directly from the slides or notes will not earn (partial) credit. Brief, clear and consistent explanation will.}

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 2 [5 pts] I. The IP address, Area ID, AS number are all identifiers. Unlike ATM networks, IP does not encode the Area ID and AS number into the IP address. What are the benefits and drawbacks of such encoding (of AreaIDs & ASNs in IP addresses) in 32 bit IPv4 addresses? [REMINDER: Be brief]

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 3 [4 pts] II. What is the key difference between I-BGP and an IGP protocol (like OSPF)? Why are I-BGP mechanisms preferred to OSPF’s AS-external-LSA (type 5 LSA) mechanism? [REMINDER: Be brief]

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 4 II. [6 pts] III. TCP congestion control has four key pieces: self-clocking, slow start phase, congestion avoidance phase and fast retransmit/recovery. a) Explain the difference between slow-start and congestion-avoidance phases. Why do we need both phases (why not just one of the two)? [REMINDER: Be brief] b) Under what circumstances does fast retransmit actually avoid an expensive timeout? When could it fail (and TCP will timeout) ? [REMINDER: Be brief]

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 5 [10 pts] IV. Briefly explain the differences (if any) in: How the OSPF building blocks (Hello exchange, database synchronization, LSA type & who generates LSAs for the link), are mapped to various underlying networks: - point-to-point links, - broadcast subnets & - NBMA subnets (the pt-mpt model ONLY, not the NBMA model). Hello Exchange: (3pts) Database Synchronization: (3pts)

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 6 LSA Type & Who generates it: (4 pts)

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 7 q III. [12 pts] A large enterprise (AS1) has 125 LANs with 250 nodes each. It has the address prefix /16 from ISP1. It has two connections to ISP1, and one connection to ISP2 [See Figure below]. It does business with AS2 and AS3 (with prefixes /16 and /20). Which of the following actions are possible? What BGP-related mechanisms would you use (briefly explain)? q A) Send the outbound traffic to AS2 via Link A and the outbound traffic to AS3 via Link B. q B) For all other incoming traffic, (assuming traffic is uniformly distributed to all of its address space) it would like to load-balance and get roughly 1/6th of all traffic from Link C. q What implications does AS1’s actions have on the global routing mesh ? Internet Routing Mesh Link B AS /16 Link A Link C ISP /11 ISP /10 AS /16 AS /20

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 8 q

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 9 q III. [13 pts] Consider a long TCP session (large file transfer) from Server A to Client B running over an asymmetric last mile link. Assuming 1500 byte packets and 40 byte acks (ignore other headers), and no loss anywhere, q What is the maximum TCP throughput possible from A->B? Why ? q What would happen if the packet size were 8000 bytes or 500 bytes instead? [ Hint: think about how self-clocking determines TCP throughput ] q Suppose, operator Frank Hacker can: q i) modify the interface at node B to suppress a subset of TCP ACKs q ii) regenerate the suppressed TCP ACKs at node interface I (towards node A) q Hacker is not allowed to break the byte-stream and end-to-end reliability semantics of TCP. (eg: node I cannot generate acks beyond the ack number received from node B). q Briefly explain how and why these actions could affect the I->B TCP dynamics and throughput (if at all)? [ Hint: think about what would happen if an ack were lost (or suppressed), how can a lost ack be regenerated, how these actions affect TCP self-clocking & hence the TCP throughput ] AIB Internet (no congestion, bandwidth > 6Mbps) 6Mbps 16kbps

Shivkumar Kalyanaraman Rensselaer Polytechnic Institute 10