1. Two primary goals: low-effort interconnection and generality.  The phone
network was designed for exactly one application: phone calls, and didn't
emphasize low-effort interconnection. This was mostly because the phone network
was owned and operated by one major entity (AT&T).

2. Packet switching because during the 50% of the time that you application is
dormant, you can use the network's links to transmit data belonging to other
applications. This leverages the fact that not all applications want to send
data at the same time (statistical multiplexing).

3. Circuit switching. For an application like audio, the throughput
requirements are much less demanding, but it's more important that you have
predictability (i.e., no queueing).  Circuit switching provides this
predictability by guaranteeing that if you're using a circuit on the network,
no one else is.

4. Application layer: DNS lookup to find IP address from domain name. Getting
data from the application (such as the search query typed into the browser) and
sending it over packets using a socket API call.
Transport layer: Connection establishment (using accept/listen) and reliable
and in-order delivery of HTML/CSS/JavaScript pages.
Routing layer: At the end hosts, the routing layer inserts the destination
address into packets and sends it to the default router. Every router then
consults its forwarding table to determine who to forward the packet to next
(i.e., the next hop) until it eventually reaches the destination.
Link layer: Once a laptop/desktops knows its first/default router, it needs to
actually get packets to this first/default router using either WiFi or
Ethernet. The link layer handles this.
Physical layer: For the  link layer to work, you need a cable or antenna, and a
way to translate bits into voltages on a wire/antenna. This is the domain of
the physical layer.

5. Stop And Wait: 1/RTT_min packets per second.
Sliding Window: W/RTT_min packets per second.
Multiply the two quantities by packet size if you want the answer in bits per
second.

6. It's not valid in Stop And Wait because the sender will never move on to
transmitting the 5, 6, and 7th packet (and hence the receiver won't receive the
7th packet) until the sender receives an ACK for the 5th one. It's valid in
Sliding Window assuming a window size that is at least 5 packets. The reason
for this is that right before 12 is transmitted at the sender, the sender has
not yet received ACKs for 5, 6, 9, and 11. So the sender's window must be at
least 5: 4 to account for the 4 packets (5, 6, 9, and 11) that haven't yet been
received and hence ACKed and 1 more to allow the packet with sequence number 12
to be transmitted.

7. Everyone uses a large window: congestion collapse.
Everyone uses a small window: underutilization if the number of people is
small.
Eventually, this will also lead to congestion collapse but at a later
point because each person's individual window is small.

8. Any of the examples from section 3, 4, or 5 from lec6.pdf is a valid answer.

9. Look at question 14 in
https://cs.nyu.edu/courses/fall18/CSCI-UA.0480-009/exams/midterm_solutions.pdf

10. For a small file, the fetch time is dominated mostly by the minimum
round-trip time because the file fits into a packet. So it's the time taken to
send and receive a single packet. For a large file, the fetch time is dominated
by the transmission delay (there are thousands of packets to be transmitted),
which is in turn dominated by capacity.

So for a small file, I would bring the server closer to the user. For a large
file, I would add more capacity to the client-to-server link.

11. See Section 5 of lec9.pdf for one example for what happens when a BGP
advertisement is false. We could ask similar questions inspired by the YoutTube
outage discussed in section 5.

12. Won't be tested.

13. Hierarchy allows DNS to scale to a large number of domain names (and
corresponding IP addresses). For instance, the TLD server for the .com TLD is
only responsible for domain names that end with .com. Similarly, a TLD server
for the .edu TLD is only responsible for domain names that end with .edu. In
other words, hierarchy creates a division of responsibility where each entity
is only responsible for a portion of domain names.

14. Domain Name System, Public vs. Private IP addresses.

15. A large alpha makes the timeout calculation converge faster to a new set of
RTT values, while a small alpha is more sluggish.