1 - Server/Client

Question 1

What are the 2 broad models of a distributed system?

Answer 1

1. Server with many client (can be thick or thin)
2. Peer to peer systems (clients are servers)

Question 2

What really is the cloud? What’s an elastic cloud?

Answer 2

The cloud is literally just a host. Elastic Cloud is just adding more hosts to a distributed system.

Question 3

What are the ways that we can find and contact resources?

Answer 3

1. Just hard code it (put in IP)
2. Configuration file (hostname)
3. Announce yourself (DHCP)
4. Lookup service (domain names!)

Question 4

What are the 4 resource location requirements?

Answer 4

1. Location transparency
2. Global uniqueness
3. Ability to access names from all locations
4. Protocol identification - speak same language

Question 5

What is DNS?

Answer 5

Domain Name System:
- hierarchical lookup, translates domain names to IP addresses

Question 6

What are 3 issues with client/server architectures?

Answer 6

1. Division of labour (how to split up work?)
2. Reliability of transmission
3. Is it the right solution?

Question 7

What are some important considerations for server design of multiple machines?

Answer 7

1. How to manage state between clients/server, server/client?
2. Concurrency
3. Work distribution

Question 8

Who holds state in a thin server? What about thick?

Answer 8

Thin - client (stateless)
Thick - server (stateful)

Question 9

For an NFS like aviary, argue for both sides of why stateless or stateful would be better

Answer 9

Stateful - need to lock files to prevent concurrent writes

Stateless - invalid locks on crashes, stateless would be better

Question 10

Give the pros and cons of thin clients

Answer 10

Pros:
cheap clients

Cons:
beefy servers, can’t do anything if not connected to internet

Question 11

Give the pros and cons of thick clients

Answer 11

Pros:
- offload work to client, server has less load
- reduce cost if lots of stuff is done on client-side
- client can work even if not connected (offline netflix)

Cons:
- maintenance is hard because you have to push it out to every client

Question 12

When sending data, what does it mean to “push”?

Answer 12

Send an announcement, data to peers and no response is needed

Question 13

When sending data, what does it mean to “pull”?

Answer 13

Request/response protocol, (e.g. HTTP), you ask for a response and you get that resource

Question 14

When sending data, what does it mean to “push/pull”?

Answer 14

Do both push/pull, can announce and respond to requests (P2P)

Question 15

When would you block?

Answer 15

When you need to know the response before continuning:

• check if logged in
• check if the write worked

Question 16

What’s a rendezvous?

Answer 16

Synchronizes client and server

• client blocks while it waits for a response
• client and server have the same data for a moment

Question 17

What’s a callback?

Answer 17

Client sends a message and says where it is to send a message back

• client goes on doing whatever, server eventually sends a message

Question 18

What’s a relay/buffer?

Answer 18

• Send asynchronously to a middle host
• sender does a push, receiver can be pushed-to, or pull from middle (message queue!)

Question 19

How does P2P flooding work?

Answer 19

Receive a message, forward message to every peer except the one that sent it (message is received by a peer multiple times)

Question 20

How does P2P gossiping work?

Answer 20

Flooding, but only to some neighbours which reduces the avalanche of methods