Lab 2 (January 1, 2018)

Socket Programming

What is the model for network programming?
- Client-server

Abstraction Level

Clients and servers are both programs at the app layer
Transport layer is responsible for providing communication services

Application Layer Models

Application Architectures
- Client-server model: Web (TCP), FTP (TCP), E-mail (TCP), DNS (UDP/TCP), RTP
- Peer-to-peer: BitTorrent (TCP), Tor (AKA "Onion Routing", TCP)
- Hybrid model: Skype (TCP & UDP), GTalk (TCP & UDP)

Web and HTTP

Web page consists of objects, addressable by a URL
HTTP: Hypertext transfer protocol
- Client/server model
- Uses TCP, a stateful protocol
- HTTP itself, however, is stateless (use cookies if we want stateful service)
- Note: UDP is stateless

Response Time

RTT (round-trip delay time): time for a small packet to travel from a client to server and back
One RTT to initiate TCP connection
One RTT for HTTP request an first few bytes of HTTP response to return
File transmission time
Total = 2RTT + Transmission Time

Nonpersistent vs. Persistent

For parallel TCP connections where n is the number of objects to transport and m is the number of parallel TCP connections:
- ⌈n/m⌉ = number of RTTs required

Nonpersistent HTTP

Requires 2 RTTs per object
OS overhead for each TCP connection
Browsers often open parallel TCP connections to fetch referenced objects

Persistent HTTP

Server leaves connection open after sending response
Subsequent HTTP messages between same client/server sent over open connection
Client sends requests as soon as it encounters a referenced object
As little as one RTT for all the referenced objects