Chapter 2: The Application Layer

Question 1

How does a client-server application architecture work?

Answer 1

A server is an always-on host with a permanent IP address, typically in data centres.

A client cannot communicate to another client directly; instead it communicates to a server, which forwards the request to the other client.

Question 2

What is the peer-to-peer (P2P) architecture?

Answer 2

Arbitrary end systems communicate directly with each other. These (termed peers) request a service from other peers, while providing a service in return to other peers.

Peers change IP addresses and are intermittently connected. It still has ‘server’ and ‘client’ processes.

Question 3

What is a process?

Answer 3

A program within a host.

Question 4

What is a socket (or API)?

Answer 4

A software interface between two application layers, analogous to a letterbox.

Question 5

How are processes addressed?

Answer 5

Each process is associated with a port number. To communicate, that socket’s identifier is the IP address and port.

Question 6

What are the four main variables of data transmission that should be considered when choosing investigation transport service protocols?

Answer 6

Data Integrity
Throughput
Timing Guarantees
Security.

Question 7

Give the things that an application protocol defines.

Answer 7

Types of message exchanged (request, response).
Message syntax (how fields are delineated).
Message semantics (meaning of information in fields).
Rules for when and how processes send and respond to messages.

Question 8

Which of the following does TCP provide? Which does UDP provide?

reliable transport
timing
minimum throughput
congestion control
flow control
security
no set-up required

Answer 8

TCP: reliable transport, congestion control, flow control

UDP: No set up required

Question 9

Why is HTTP is a stateless protocol?

Answer 9

It doesn’t maintain any information about the client.
[If a client asks again for an item, the server doesn’t recall that it has already been sent.]

Question 10

Compare persistent and non-persistent connections.

Answer 10

In non-persistent connections, only one object can be sent over each HTTP connection. In persistent, there can be multiple object sent over the server.

Question 11

What are the steps in a non-persistent HTTP connection?

Answer 11

1. The HTTP client sends a TCP request to the server.
2. The server accepts.
3. The client sends the HTTP request to the server.
4. The server sends the html file.
5. The server sends a termination request, terminating the question
6. The client receives the file, realises that there are objects linked to.
   [Steps 1-5 are repeated for each object].

Question 12

What is the response time for a non-persistent connection, with a base page and 10 objects?

Answer 12

1 RTT for TCP initiation
1 RTT for HTTP request and first few bytes of HTTP response to object/file transmission.
Then there is file transmission time.

Thus, we have (2 RTT + file transmission) + 10 * (2 RTT + object transmission).

Question 13

What is the response time for a persistent connection, with a base page and 10 objects?

Answer 13

1 RTT for TCP set up
1 RTT for HTTP set up
11 * object transmission
= 2 RTT + 11 * object transmission

Question 14

Give a standard example of a HTTP
GET request and header line for a non-persistent connection.

Answer 14

GET /somedir/page.html HTTP/1.1
Host: www.someschool.edu
Connection: close
User-agent: Mozilla/5.0
Accept-language: en

Question 15

What are the HEAD and PUT HTTP methods used for?

Answer 15

HEAD is for debugging. PUT is used for uploading objects to a directory on a Server.

Question 16

Give a standard response HTTP message.

Answer 16

HTTP/1.1 200 OK
Connection: close
Date: Tue, 18 Aug 2015 15:44:04 GMT
Server: Apache/2.2.3 (CentOS)
Last-Modified: Tue, 18 Aug 2015 15:11:03 GMT
Content-Length: 6821
Content-Type: text/html

Question 17

What do the following sample codes represent?
200, 301, 400, 404, 505

Answer 17

200 - OK
301 - Moved Permanently
400 - Bad Request
404 - Not Found
505 - HTTP Version Not Supported

Question 18

What are the four components of cookies?

Answer 18

1. Cookie header line of HTTP response
2. Cookie header line in next HTTP request message
3. Cookie file kept on user’s host, managed by user’s browser
4. Back-end database at website

Question 19

Explain the steps of setting a cookie and using it at a later time.

Answer 19

1. The client sends a HTTP request msg to the server.
2. The server creates a cookie ID in its backend database.
3. The server sends a HTTP response with a set-cookie line (and cookie ID)
4. The client then sends a later message with that cookie ID and the server then knows the user.

Question 20

Give four uses of cookies.

Answer 20

Authorisation
Shopping carts
Recommendations
User session state (web email)

Question 21

Suppose a browser requests an image from a server with a proxy and cache hit. Describe the exchange.

Answer 21

1. The browser establishes a TCP connection to the proxy server and sends an HTTP request or the object to the proxy server.
2. The proxy server checks if it has the object and returns it within a HTTP response message.

Question 22

Suppose a browser requests an image from a server with a proxy and cache miss. Describe the exchange.

Answer 22

1. The browser establishes a TCP connection to the proxy server and sends an HTTP request or the object to the proxy server.
2. The proxy server checks if it has the object and opens a TCP connection to an origin server and sends an HTTP request.
3. The origin server sends the object within an HTTP response to the proxy server.
4. The proxy server returns it to the client in a HTTP response message.

Question 23

An institutional network is connected to the Internet with a 15 Mbps link. The average object size 1 Mb and the average request rate from the browsers to the origin servers is 15 requests per second.

Assume the request messages are negligibly small - so don’t contribute to traffic.

What is the traffic intensity on the access link?

Answer 23

15 requests/sec * 1 Mb/request * 1/15 s/Mb = 1

(Thus, this may begin growing too fast to handle)

Question 24

An institutional network is connected to the Internet with a 15 Mbps link. The average object size 1 Mb and the average request rate from the browsers to the origin servers is 15 requests per second.
There is also a proxy server in the institutional network, which is a 100 Mbps LAN.

Assume the request messages are negligibly small - so don’t contribute to traffic.

Assume the Internet delay is 2 seconds.

What is the average delay for a requests, if the cache hit rate is 0.4?

Answer 24

The traffic intensity is 0.6.

Then:
Cache hit time is approx 0.01s
Cache miss time is approx 2.01s

Thus, time = 0.4 * 0.01 + 0.6 * 2.01 s

Question 25

How does the conditional GET work and why is it needed?

Answer 25

It is needed to combat stale data in the web server. 1. When the proxy server forwards information from the origin server to the client, it stores the last modified date too. 2. When another browser requests that object later, the proxy issues a conditional GET by sending a GET requests with just a host and If-modified-since header line. 3a. If it has not been modified, an empty entity body is returned and status code of 304 (Not Modified) 3b. If it has been modified, the object is returned with a last modified date.

Question 26

How does HTTP/2 fix the HOL issue with HTTP/1.1?

Answer 26

HTTP/1.1 has a Head of Line (HOL) blocking problem: if there is a large object, like a video, due to the persistent connection, all later objects are delayed. Normally multiple parallel connections are then needed to reduce user-perceive delay. In HTTP/2, each message is split into small frames and interleave the requests and response message on the same connection.

Question 27

Aside from framing, give 2 other benefits of HTTP/2 over HTTP/1.1.

Answer 27

Ability to prioritise messages via weightings. Servers can send multiple responses for a single client request.

Question 28

Give a benefit of HTTP/3 over HTTP/2.

Answer 28

Security Per-stream flow control Low-latency connection establishment

Question 29

Give the services that DNS provides.

Answer 29

Host aliasing, mail server aliasing, load distribution, hostname to IP address translation.

Question 30

Why is it problematic to have a centralized design for DNS?

Answer 30

Single point of failure, overloads from traffic, distant central database, maintenance.

Question 31

Explain how DNS lookups work.

Answer 31

1. The browser gets the hostname from the URL and passes it to the client side of the DNS application. 2. The DNS client sends a query to the root DNS server, which replies with the top level domain address 3. The client recursively or iteratively tracks through the top-level, and then authoritative DNS servers hierarchy until a DNS server knows the IP address for it.

Question 32

What is a local DNS server?

Answer 32

A DNS server belonging to each ISP that acts like a proxy to send to the DNS hierarchy.

Question 33

What is the format of a DNS resource record for: type A? type NS? type CNAME? type MX?

Answer 33

(hostname, IP address, A, TTL) (alias, canonical name, CNAME, TTL) (domain, hostname of authoritative name server, NS, TTL) (name, name of mailserver, MX, TTL)

Question 34

What is the format for a DNS message?

Answer 34

12 bytes for the header section: - 2 bytes for the query ID - 1 bit for identifying query (0) or reply (1) - 1 bit for identifying if the reply is from an authoritative server. - 1 bit for setting if the client desires the recursive approach for finding records. - 1 bit for if recursion is available. - 4 number of fields that tell how much questions, answers, authority and additional info there is. - Name/type fields for a query - RRs in response to a query - Records for authoritative servers - Additional helpful info

Question 35

How are DNS records inserted in the first place?

Answer 35

The domain name is registered at a registrar, a commercial entity that verifies the uniqueness of the domain name, which enters it into the DNS database (and you pay them for this). You will provide the names and IP address of your primary and secondary authoritative DNS servers.

Question 36

Define distribution time.

Answer 36

The time taken get a file from one peer to all others who requires it.

Question 37

What is the distribution time for N peers needing a file of F size, if the server upload rate is Us and the download rate of peer i is d_i? Assume this is a client-server architecture.

Answer 37

The upload takes N*F/Us secs. The minimum distribution time is at least F/d_min. Then, we have Dcs = max{NF/Us, F/d_min}

Question 38

What is the distribution time for N peers needing a file of F size, if the server upload rate is Us and the download rate of peer i is d_i? Assume this is a P2P architecture.

Answer 38

The server must upload a copy to the internet, so F/Us is a minimum. The peer with the lowest download rate has F/dmin as a lower bound. The total upload capacity is the upload of the server plus all the uploads of the peers. We then have NF bits to upload, and so the minimum distribution time is NF/(U_s + (u_1 + u_2 + ... + u_n)) We take the max of these to get the rate.

Question 39

As the number of peers involved increases, what happens to the minimum distribution time of: a) client-server b) P2P

Answer 39

a) Grows linearly b) Grows logarithmically.

Question 40

In BitTorrent, how does a new peer receive a file?

Answer 40

A random subset of peers is chosen by the tracker from the set of participating peers and sends the IP addresses to the new peer. TCP connections are set up to as many of these are possible; and if they connect, we consider it a neighbouring peer - though these will fluctuate over time. At any point, each peer has a different subset of chunks (each chunk is 256 kB). Periodically, the new peer will ask all neighbouring peers for the chunks they have. The new peer then issues requests for chunks that are missing until the file is complete, with the rarest first. This means the rarest chunks get quickly redistributed, aiming to equalize the numbers of copies of each chunk in the torrent.

Question 41

How does a peer decide how to send chunks in a torrent?

Answer 41

Priority goes to the four peers supplying them with the highest rate of chunks, and sends these four peers (termed unchoked) chunks. Every 10 seconds, this is recalculated. Additionally, every 30 seconds, a random peer is sent chunks (termed optimistically unchoked). Effectively, if the peers are both satisfied, then they become unchoked peers.

Question 42

How are videos stored in bits?

Answer 42

A sequence of images displayed at constant rate (usually 24 frames per sec). Each image is an array of pixels. Bits are reduced spatially (per image) and temporal (storing only those that change). On encoding, we can have CBR (constant bit rate), so the video encoding rate is fixed or VBR (variable bit rate) where the encoding rate changes as amount of coding changes.

Question 43

Why was DASH set up for video transfer?

Answer 43

To allow clients to all have different encodings that fit their bandwidth.

Question 44

How does DASH (Dynamic Adaptive Streaming over HTTP) work?

Answer 44

Several encodings of a video are stored, and when a client's bandwidth is high, chunks from a high-rate version are selected and from a low-rate version when a client's bandwidth is low. The data about which file and versions exist is called a manifest file.

Question 45

What does a CDN (Content Distribution Network) allow for?

Answer 45

Widescale streaming of media.

Question 46

What is the Enter Deep philosophy for CDNs?

Answer 46

Deploy server clusters in access ISPs over the world to get close to end users. Maintaining and managing clusters becomes challenging.

Question 47

What is the Bring Home philosophy for CDNs?

Answer 47

Large clusters are built at a few IXPs. There is lower maintenance, but higher delay and throughput to end users.

Question 48

How does a client access a video at a CDN?

Answer 48

1. The user clicks on the link to the video (say http://video.netcinema.com/6Y7B23V) and a DNS query is sent for video.netcinema.com. 2. The local DNS relays this to an authoritative DNS server for NetCinema, which notices the 'video' in the hostname. 3. The authoritative DNS server 'hands over' the query to a CDN by returning a hostname in KingCDN's domain, say 'a1105.kingcdn.com'. 4. The local DNS then issues that request until an IP address of a content server is returned. 5. The LDNS returns this to the client. 6. A TCP connection is set up and a GET request issued - DASH may be used - and the video downloaded.

Question 49

How is the best cluster to direct to decided?

Answer 49

It could be geographical, but it is better to perform real-time measurements via probes to LDNS around the world (though some aren't configured to receive them.)

Question 50

How does Netflix employ CDNs?

Answer 50

When a user selects a movie to play, the Netflix software running in the Amazon cloud decides which of its CDN servers is the best to deliver the content. It sends the IP address to the client alongside a manifest file. The CDN server and client then directly interact using DASH to get chunks of the movie. [No DNS redirect is needed]