Proxy Lab Recitation I

1
Proxy Lab Recitation I

• Monday Nov 20, 2006

2
Outline

• What is a HTTP proxy?
• HTTP Tutorial
• HTTP Request
• HTTP Response
• Sequential vs. concurrent proxies
• Caching

3
What is a proxy?
Proxy
Client
Server
Browser
www.google.com

• Why a proxy?
• Access control (allowed websites)
• Filtering (viruses, for example)
• Caching (multiple people request CNN)

4
Brief HTTP Tutorial

• Hyper-Text Transfer Protocol
• Protocol spoken between a browser and a
  web-server
• From browser ? web-server REQUEST
• GET http//www.google.com/ HTTP/1.0
• From web-server ? browser RESPONSE
• HTTP 200 OK
• Other stuff

5
HTTP Request
Request Type
Path
Host
Version

• GET http//csapp.cs.cmu.edu/simple.html HTTP/1.1
• Host csapp.cs.cmu.edu
• User-Agent Mozilla/5.0 ...
• Accept text/xml,application/xml ...
• Accept-Language en-us,enq0.5 ...
• Accept-Encoding gzip,deflate ...

An empty line terminates a HTTP request
6
HTTP Request

• GET http//csapp.cs.cmu.edu/simple.html HTTP/1.1
• Host csapp.cs.cmu.edu
• User-Agent Mozilla/5.0 ...
• Accept text/xml,application/xml ...
• Accept-Language en-us,enq0.5 ...
• Accept-Encoding gzip,deflate ...

The Host header is optional in HTTP/1.0 but we
recommend that it be always included
7
HTTP Request

• GET http//csapp.cs.cmu.edu/simple.html HTTP/1.1
• Host csapp.cs.cmu.edu
• User-Agent Mozilla/5.0 ...
• Accept text/xml,application/xml ...
• Accept-Language en-us,enq0.5 ...
• Accept-Encoding gzip,deflate ...

The User agent identifies the browser type. Some
websites use it to determine what to send. And
reject you if you say you use MyWeirdBrowser ?
Proxy must send this and all other headers
through
8
HTTP Response
Status

• HTTP/1.1 200 OK
• Date Mon, 20 Nov 2006 033417 GMT
• Server Apache/1.3.19 (Unix)
• Last-Modified Mon, 28 Nov 2005 233135 GMT
• Content-Length 129
• Connection Keep-Alive
• Content-Type text/html

Status indicates whether it was successful or
not, if it is a redirect, etc. The complete
response should be transparently sent back to the
client by the proxy.
9
HTTP Response

• HTTP/1.1 200 OK
• Date Mon, 20 Nov 2006 033417 GMT
• Server Apache/1.3.19 (Unix)
• Last-Modified Mon, 28 Nov 2005 233135 GMT
• Content-Length 129
• Connection Keep-Alive
• Content-Type text/html

This field identifies how many bytes are there in
the response. Not sent by all web-servers. DO
NOT RELY ON IT !
10
Concurrent Proxy

• Need to handle multiple requests simultaneously
• From different clients
• From the same client
• E.g., each individual image in a HTML document
  needs to be requested separately
• Serving requests sequentially decreases
  throughput
• Server is waiting for I/O most of the time
• This time can be used to start serving other
  clients
• Multiple outstanding requests

11
Concurrent Proxy

• Use threads for making proxy concurrent
• Create one thread for each new client request
• The thread finishes and exists after serving the
  client request
• Use pthread library
• pthread_create(), pthread_detach(), etc.
• Can use select() as well for adding concurrency
• Much more difficult to get right

12
Caching Proxy

• Most geeks visit http//slashdot.org/ every 2
  minutes
• Why fetch the same content again and again?
• (If it doesnt change frequently)
• The proxy can cache responses
• Serve directly out of its cache
• Reduces latency, network-load

13
Caching Implementation Issues

• Use the GET URL (host/path) to locate the
  appropriate cache entry
• THREAD SAFETY
• A single cache is accessed by multiple threads
• Easy to create bugs thread 1 is reading an
  entry, while thread 2 is deleting the same entry

14
General advice

• Use RIO routines
• rio_readnb, rio_readlineb
• Be very careful when you are reading line-by-line
  (HTTP request), versus just a stream of bytes
  (HTTP response)
• When to use strcpy() vs. memcpy()
• gethostbyname(), inet_ntoa() are not thread-safe!
• Path sequential ? concurrency ? caching