CSE 291 System Services for the World Wide Web

1
CSE 291System Services for the World Wide Web

• Winter 2000
• Geoffrey M. Voelker

2
Class Goal

• Provide background for doing experimental systems
  research in wide-area systems such as the Web
• Architectures for wide-area distributed systems
• What are good models for structuring systems to
  support apps?
• Understanding system behavior
• What are the workloads?
• Enhancing performance
• Caching, prefetching
• New opportunities
• Multimedia, XML
• Read, evaluate, present a variety of research
  papers
• Do projects on interesting problems in area

3
Today

• Course structure
• Presentations
• Evaluations
• Projects
• Course contents
• Web intro
• Topics and papers

4
Presentations

• We will present papers in class for discussion
• Roughly 2 papers per 1.5 hour class (sometimes 3)
• Everyone presents (including me)
• How many depends on students registered
• I will present Thursdays papers and do my share
  during the rest of the quarter
• Look over schedule between now and Thursday
• Well allocate papers on Thursday

5
Evaluations

• You must submit evaluations of papers
• Email them to me by noon of day of class
• No evals if you have to present
• Brief (½ page)
• Summary of paper (research problem, conclusions)
• What you learned
• Any ideas that occurred to you
• Your frank opinion of topic and/or work
• If this gets to be too burdensome, might notch it
  down to one evaluation per day

6
Class Participation

• The presentations are for fostering discussion
• so I expect you to participate in discussions
• Presenters
• Come prepared with discussion questions
• Rest of us
• Use your evaluations as a basis for discussion

7
Class Project

• For those signed up for four units
• Work in pairs
• Schedule in handout, on the Web
• Roughly 5 weeks setup, 5 weeks working
• Start thinking about what you might want to work
  on, who you might want to work with
• Ill have a list of topics, but I also encourage
  you to use your own
• Your final will be a class presentation on your
  project

8
Course Contents

• Topics
• Wide-area system architectures
• Naming
• Scalable servers
• Workload characterizations
• Caching, prefetching
• Protocols
• Security
• Emerging applications
• Overview of Web to help put things into context

9
How does the Web work?

• The canonical example in your Web browser
• Click here
• here is a Uniform Resource Locator (URL)
• http//www-cse.ucsd.edu
• It names the location of an object on a server

10
In Action
http//www-cse.ucsd.edu
HTTP

• Client resolves name of server (www-cse.ucsd.edu)
• Establishes a connection with the server
• Sends the server the name of the object (null)
• Server returns the object

11
Naming

• How should objects be named?
• URLs name locationsif an object moves, the URL
  breaks
• Location-independent names seem like the obvious
  way to go
• Why dont we use them (e.g., URIs)?
• How do we make them work, esp. in the face of
  mobility?
• How it works now, how it might work in the future
• DNS Mockapetris88
• DNS for URIs Daniel96
• Names as programs Vahdat99
• Finding replicas Guyton95, vanSteen98

12
Communication
http//www-cse.ucsd.edu
HTTP

• Communication between the client and server is
  done via HTTP over TCP/IP

13
Protocols

• What kind of transport protocol should the Web
  use?
• HTTP 1.0
• One TCP connection/object
• Complaints inefficient, slow, burdensome
• HTTP 1.1
• One TCP connection/many objects (persistent
  connections)
• Solves all problems, right? Huge amount of
  complexity
• Clients, proxies, servers
• How do they compare?
• Protocol differences Krishnamurthy99,
  performance comparison Nielsen97, effects on
  servers Manley97, overhead of TCP connections
  Caceres98

14
Scalable Servers
Server

• Of course, you are not the only person accessing
  the server

15
Scalable Servers

• How do you build servers to handle millions of
  hits a day?
• Web servers Flash Pai99, scheduling
  Crovella99
• Mail servers EarthLink Christenson97, Saito99
• Principles Transcend, HotBot Fox97
• Techniques Load balancing Pai98

16
Web Caching

• Gee, is there some way to offload those busy
  servers?
• Use caches to exploit reference locality among
  clients

17
Caching

• How should we build caching systems for the Web?
• Seminal paper Chankhunthod96
• Proxy caches Duska97
• Akamai hack Karger99
• Cooperative caching Tewari99, Fan98, Wolman99
• Popularity distributions Breslau99

18
Prefetching

• The fastest way to download a page is to fetch it
  before it is accessed
• How do you know what will be accessed?
• How much bandwidth can you afford for mistakes?
• Performances bounds Kroeger97
• Survey paper w/ practical approach Duchamp99

19
Security

• We cant just assume were in a back room anymore
• How do we secure access to resources?
• Infrastructure SDSI Rivest96
• Wide-area service CRISIS Belani98
• Computational grids Globus Foster98
• E-commerce SSL Wagner96
• Downloaded code Java Wallach97

20
Workload Characterizations

• How can you fix it if you dont look inside?
• Is the Web slow because of the network, the
  server, CPU-hogging browsers?
• What is the behavior of clients, proxies, and
  servers?
• Golden fleece Invariants across populations and
  time
• E.g., Zipf-like distribution of object
  popularities
• How can we use workloads to shape the systems we
  design?
• Characterization survey Pitkow98
• Rate of change Douglis97
• Key to data dissemination (caching, prefetching,
  etc.)

21
Emerging Applications

• HTML, gif, jpeg, etc. are all old news
• Whats the new, cool stuff?
• Multimedia
• Streaming multimedia next big thing (20 b/w in
  UW traces)
• Workloads Mena00, tools Caceres99, delivery
  Eager00
• XML
• I dont know what it is, so Id like to learn
• Papers TBD

22
Architectures

• The Web is basically a simple read-only data
  access system
• Click, fetch, click, fetch, click, fetch
• Why not fully generalize it into a universal
  wide-area distributed system?
• The Web as an operating system WebOS Vahdat98
• Wide-area computational grids Legion Lewis96,
  Grimshaw98, Globe vanSteen97, Globus
  Foster97, Foster98

23
Hits
Misses
Misses
Clients
Proxy Cache
Servers