Chapter 9 HTTP, Caching, Load Balancing

1
Chapter 9HTTP, Caching, Load Balancing

• Professor Rick Han
• University of Colorado at Boulder
• rhan_at_cs.colorado.edu

2
Announcements

• HW 6 due next week, Thursday April 17
• Programming Assignment 4 soon
• No OH Today, Wed OH 1-3 pm
• Attending a conference on Tuesday April 22
• Next, Application Layer

3
Recap of Previous Lecture

• Domain Name Service
• Translate/resolve a name to an IP address
• www.cs.colorado.edu 128.9.17.42
• Hierarchical name space
• Hierarchical name servers
• Root name servers about a dozen
• Then, .edu, .com, .gov, .mil, .org, .net,
• Local name server
• Authoritative name server gives back final IP
  address
• Recursive vs. iterative queries
• Caching

4
DNS Lookup Example
root edu DNS server
www.cs.colorado.edu
www.cs.colorado.edu
NS colorado.edu
colorado.edu DNS server
Local DNS server
NS cs.colorado.edu
Client
cs.colorado.edu Authoritative DNS server
wwwIPaddr
Courtesy Srini Seshan
5
More on DNS

• In addition to name translation, DNS helps with
• Host aliasing
• DNS supports multiple host names for a single IP
  address, e.g. yahoo.com and www.yahoo.com
• Load distribution
• Instead of HTTP Redirect, use DNS!
• A busy site like cnn.com will have multiple
  replicated Web servers, each with a different IP
  address
• A set of IP addresses associated with cnn.com
• DNS can return multiple records that match a
  single name
• Order of replicated server addresses is rotated

6
DNS Message Format
Identification
Flags
No. of Questions
No. of Answer RRs
12 bytes
No. of Authority RRs
No. of Additional RRs
Name, type fields for a query
Questions (variable number of answers)
Answers (variable number of resource records)
RRs in response to query
Authority (variable number of resource records)
Records for authoritative servers
Additional Info (variable number of resource
records
Additional helpful info that may be used
7
More on DNS (2)
Server Farm
DNS
Client

• DNS helps with
• Load distribution (cont.)
• DNS round robin to N servers
• Akamaizing Smarter than DNS round robin
  choose the server closest to you better
  response time!
• Akamai serves a subset of cnn.com
• Each URL in Akamai subset has a name for which
  the Akamai DNS server is authoritative
• www.cnn.com/foo.gif ? a128.g.akamai.net/foo.gif

8
More on DNS (3)

• RFC 1794, DNS Support for Load Balancing
• DNS also helps with
• Mail server aliasing
• Given hotmail.com, return the specific host name
• BIND is a popular name server for Unix

9
More on DNS (4)

• Dynamic DNS
• Mapping your well-known Web name to a dynamic IP
  address (from DHCP)
• Suppose youve reserved a hostname www.myweb.org
  to serve Web pages from your home PC
• Each time your PC connects via cable/DSL, your
  ISP assigns your PC a different dynamic IP
  address via DHCP
• Users wont know your dynamic IP address but may
  remember your Web address. How can they reach
  you?
• Solution your PC includes code snippet to update
  DNS each time your PC gets a new IP address via
  DHCP
• Your PC must have authorization at a DNS server
  to update its DNS record
• Dynamic DNS services are being offered on the
  Web, some for free, others you pay

10
HyperText Transfer Protocol (HTTP)

• Basis for Web
• Application-layer protocol built on top of TCP
• Request-Response type of protocol
• Request e.g. GET URL HTTP_version_
• Response from server
• Requests and responses are encoded in text
• Stateless after request and response, no further
  state maintained
• Cookies maintain session state outside of HTTP

11
HTTP Request

• Request headers
• Authorization authentication info
• Acceptable document types/encodings
• From user email
• If-Modified-Since ? return page only if mod
  after date
• Referrer what caused this page to be requested
• User-Agent client software
• Blank line
• Body

12
HTTP Request Example GET

• GET / HTTP/1.1
• Accept /
• Accept-Language en-us
• Accept-Encoding gzip, deflate
• User-Agent Mozilla/4.0 (compatible MSIE 5.5
  Windows NT 5.0)
• Host www.seshan.org
• Connection Keep-Alive

13
HTTP Response

• Headers
• Location for redirection
• Server server software
• WWW-Authenticate request for authentication
• Allow list of methods supported (get, head,
  etc)
• Content-Encoding E.g x-gzip
• Content-Length ? bytes in content
• Content-Type ? MIME type
• Expires ? when contents become stale
• Last-Modified ? time contents last mod by servr
• Blank-line
• Body

14
HTTP Response Example

• HTTP/1.1 200 OK
• Date Tue, 27 Mar 2001 034938 GMT
• Server Apache/1.3.14 (Unix) (Red-Hat/Linux)
  mod_ssl/2.7.1 OpenSSL/0.9.5a DAV/1.0.2
  PHP/4.0.1pl2 mod_perl/1.24
• Last-Modified Mon, 29 Jan 2001 175418 GMT
• ETag "7a11f-10ed-3a75ae4a"
• Accept-Ranges bytes
• Content-Length 4333
• Keep-Alive timeout15, max100
• Connection Keep-Alive
• Content-Type text/html ? MIME Type
• ..

15
HTTP 0.9/1.0

• One HTTP 1.0 request/response per TCP connection
• Simple to implement
• Disadvantages
• Multiple connection setups ? three-way handshake
  each time
• Several extra round trips added to transfer
• Netscape browser opens up to 4 parallel HTTP 1.0
  connections
• Multiple slow starts

16
HTTP 1.0 Interaction With TCP
Server
SYN
0 RTT

• Client

SYN
Client opens TCP connection
1 RTT
ACK
DAT
Client sends HTTP request for HTML
ACK
Server reads from disk
DAT
FIN
2 RTT
ACK
Client parses HTML Client opens TCP connection
FIN
ACK
SYN
SYN
3 RTT
ACK
DAT
Client sends HTTP request for image
Server reads from disk
ACK
4 RTT
DAT
Image begins to arrive
Courtesy Srini Seshan
17
More HTTP 1.0 TCP Interaction Problems

• Lots of extra connections
• Increases server state/processing
• Server also forced to keep TIME_WAIT connection
  state for dead TCP connections
• Tends to be an order of magnitude greater than
  of active connections

18
HTTP 1.1 Persistent Connection Solution

• Multiplex multiple requests onto one open TCP
  connection ( multiple responses in reverse
  direction)
• Serialize transfers ? client makes next request
  only after previous response
• Reduce slow start latency
• Reduce amount of TCP state at both endpoints
• Reduce overhead
• HTTP 1.1 adds complexity because multiple
  requests (and responses) have to be multiplexed
  and demultiplexed

19
HTTP 1.1 Persistent Connection Example
Server

• Client

0 RTT
DAT
Client sends HTTP request for HTML
ACK
Server reads from disk
DAT
1 RTT
ACK
Client parses HTML Client sends HTTP request for
image
DAT
Server reads from disk
ACK
DAT
2 RTT
Image begins to arrive
Courtesy Srini Seshan
20
Web Caching Proxies

• Place a Web caching proxy in the network between
  Web client and Web server
• Reduces client response time
• HTTP GET only goes as far as intermediate cache,
  rather than all the way to server
• Reduces network bandwidth usage
• HTTP GET doesnt travel over wide area from
  caching proxy to server
• Reduces server load
• HTTP GET never reaches server

21
Web Proxies

• Used for Caching
• Improved response time, etc. from previous slide
• Provides a centralized coordination point to
  share cached information across all of a
  companys client hosts
• Also used for security
• Proxy for a company can be the only host that can
  access Internet
• Administrators makes sure that it is secure
• Used for protocol translation
• Translate HTTP 1.0 to/from HTTP 1.1. Enables old
  HTTP 1.0 clients to connect to HTTP 1.1 servers,
  and benefit from HTTP 1.1 performance boosts

22
Designing Caching Proxies

• How much can/should be cached?
• How large a cache is necessary?
• On disk vs. in memory ? typically on disk
• What are the cache hit rates?
• If user behavior is uncorrelated, have to cache a
  lot of data to improve response time, resulting
  in small cache hit rate
• If user behavior is correlated, i.e. everyone
  visits only a few Web sites, then cache less data
  and still improve response time (high cache hit
  rate)

23
Designing Caching Proxies (2)

• What can be cached?
• Cache first-time unknown documents/objects
• Non-cacheable documents
• CGI-scripts
• Personalized documents (cookies, etc)
• Encrypted data (SSL)
• Document should no longer be cached if
  updated/expired before reuse

24
Designing Caching Proxies (3)

• Performance
• How many TCP connections can the proxy handle?
• How to efficiently index into database/cache?
• Early caches used file system to find file
• Metadata now kept in memory on most caches
• Prefetching combine with caching to reduce
  response time
• Proxy parses a Web page and prefetch its
  hyperlinked objects before the client asks for
  them
• Example when a proxy fetches a Web page on
  behalf of a client, the proxy will parse and
  cache the Web page returned by the server, and
  then prefetch all links before client requests
  them
• Not widely used due to poor hit rates?

25
Caching Policy at Proxy

• Relevant HTTP fields
• Request
• If-Modified-Since
• Response
• Last-Modified
• Expires
• Caching proxy doesnt cache pages with
• Pragma no-cache header field
• WWW-Authenticate or Authorization headers
• Server and proxy clocks must be reliable

26
Caching Policy at Proxy (2)

• Browser has its own browser cache.
• Browser sends a conditional GET with
  If-Modified-Since header field when
• a user hits Reload, or
• a page expired in browser cache, or
• browser set to always ask for a page,
• A conditional GET will only succeed in returning
  a page if that page has been modified since the
  If-Modified-Since date.
• Otherwise, get back a status code 304 Not
  Modified
• Caching proxy receives a conditional GET what
  is its policy?

27
Caching Policy at Proxy (3)

• If page not in proxys cache, or cached page has
  expired, or if cached page was Last-Modified
  earlier than requests If-Modified-Since date,
  then
• Forward Conditional GET to server. If server
  finds its page was Last-Modified earlier than
  If-Modified-Since date, then servers response to
  proxy is status 304 Not Modified
• Proxy returns status 304 Not Modified to client
  
• Else, server returned fresher page, so proxy
  caches it and returns it to client
• Else
• return cached page to client (its not expired
  and its freshly modified)

28
Caching Policy at Proxy (4)

• Summary
• cached page returned only if not expired and new
  enough (recently modified), otherwise return
  fresher page from server or status 304 message
• Expires header may be missing, so proxy has
  to guess a probable expiration date
• If Last-Modified is recent, then guess the page
  is changing frequently, so choose a quick
  expiration date
• Chaining of caches is allowed, e.g. Browser cache
  chained with Proxy caches
• SQUID caching proxies are common freeware
• based on Harvest caches developed in part at the
  University of Colorado, see http//www.squid-cache
  .org

29
More on HTTP

• Configure the Web browser to access the Web via
  the HTTP proxy
• Internet Explorer Tools ? Internet Options ?
  Connections ? LAN Settings
• Netscape Edit ? Preferences ? Advanced ? Proxies
  ? Manual
• DNS lookup by HTTP Proxy
• Given URL http//www.cs.colorado.edu/index.html,
  then proxy must call DNS to translate
  www.cs.colorado.edu to 128.138.242.195
• Then, proxy establishes HTTP over TCP connection
  to 128.138.242.195 to retrieve URLs page

30
More on HTTP (2)

• Most browsers now support HTTP 1.1
• Compatibility with HTTP 1.0 is expected but not
  mandated
• Eased via HTTP proxies
• Load balancing via HTTP Redirect
• In response to a GET request, a server can return
  an HTTP Redirect Response
• Server selects another server that is less loaded
• Client is redirected to again send GET request to
  less loaded server

Redirect (2)
HTTP Server 2
HTTP Server 1
GET (1)
GET (3)
Client
31
More on HTTP (3)

• GET
• Retrieve document, No payload
• One-step roundtrip process
• Incorporate parameters via long URL
• The server returns a response file with a MIME
  header identifying the type of file.
• MIME was developed for email, but is reused by
  HTTP
• POST sent from Client to Server
• typically used by HTML to send data to a
  back-end CGI script
• Two-roundtrip process contact form-processing
  server, then send data
• Give information to a server, has payload
• Expect a response

32
More on HTTP (4)

• POST vs. GET
• Use POST instead of GET if you want to send
  complex long text fields/parameters to server
• PUT sent from Client to Server
• Store document at server under specified URL,
• May be disabled at server to avoid modifying
  files
• Receive a response Created, Modified,
• POST vs. PUT
• POST URL specifies the CGI process that will
  handle the enclosed form
• PUT URL specifies the enclosed document to be
  created/stored

33
Load Balancing Techniques

• HTTP Redirection
• DNS Load Balancing
• Router-based
• Zany idea 1 N servers each advertise the same
  IP address. Let IP shortest-hop routing
  determine the nearest server.
• Hopefully no loops.

34
Load Balancing Techniques (2)

• Router-based
• Better idea Place an IP router in front of N
  servers the router balances the load
• Example each server has different IP address,
  and router substitutes IP address of lightest
  loaded server
• If a TCP connection is established to a specific
  server X, router must remember to route packets
  for this TCP connection to server X only
• router cant just choose most lightly loaded
  server L, because L might not be X, so server L
  would not be expecting server Xs TCP packets
• NAT-Based (see NAT section)