Proxy Servers

1
Proxy Servers
2
What Is a Proxy Server?

• Intermediary server between clients and the
  actual server
• Proxy processes request
• Proxy processes response
• Intranet proxy may restrict all outbound/inbound
  requests the intranet server

3
What Does a Proxy Server Do?

• Between client and server
• Receives the client request
• Decides if request will go on to the server
• May have cache may respond from cache
• Acts as the client with respect to the server
• Uses one of its own IP addresses to get page
  from server

4
Usual Uses for Proxies

• Firewalls
• Employee web use control (email etc.)
• Web content filtering (kids)
• Black lists (sites not allowed)
• White lists (sites allowed)
• Keyword filtering of page content

5
User Perspective

• Proxy is invisible to the client
• IP address of proxy is the one used or the
  browser is configured to go there
• Speed up retrieval if using caching
• Can implement profiles or personalization

6
Main Proxy Functions

• Caching
• Firewall
• Filtering
• Logging

7
Web Cache Proxy

• Our concern is not with browser cache!
• Store frequently used pages at proxy rather than
  request the server to find or create again
• Why?
• Reduce latency faster to get from proxy so
  makes the server seem more responsive
• Reduce traffic reduces traffic to actual server

8
Proxy Caches

• Proxy cache serves hundreds/thousands of users
• Corporate and intranets often use
• Most popular requests are generated only once
• Good news
• Proxy cache hit rates often hit 50
• Bad news
• Stale content (stock quotes)

9
How Does a Web Cache Work?

• Set of rules in either or both
• Proxy admin
• HTTP header

10
Dont Cache Rules

• HTTP header
• Cache-control max-agexxx, must-revalidate
• Expires date
• Last-modified date
• Pragma no-cache (doesnt always work!)
• Object is authenticated or secure
• Fails proxy filter rules
• URL
• Meta data
• MIME type
• Contents

11
Getting From Cache

• Use cache copy if it is fresh
• Within date constraint
• Used recently and modified date is not recent

12
2. Firewalls

• Proxies for security protection
• More on this later

13
3. Filtering at the Proxy

• URL lists (black and white lists)
• Meta data
• Content filters

14
Filtering
label base
Web doc
URL lists
keywords
URLs
ratings
URLs
ratings
15
The Problem the Web

• 1 billion documents (April 2000)
• Average query is 2 words (e.g., Sara name)
• Continual growth
• Balance global indexing and access and
  unintentional access to inappropriate material

16
Filtering Application Types

• Proxies
• Black lists
• White lists
• Keyword profiles
• Labels

17
Black and White Lists

• Black list URLs proxy will not access
• White list URLs proxy will allow access

18
How Is Filtering/selection Done?

• Build a profile of preferences
• Match input against the profile using rules

19
Black and White Lists

• Black list of URLs
• No access allowed
• White list of URLs
• Access permitted

20
Lists in Action

• 1 billion documents!
• Who builds the lists
• Who updates them
• Frequency of updates

21
Labels

• Metadata tags
• Rule driven PICS rules for example
• Labels are part of document or separate
• Separate label bureau

22
Labels

• Metadata (goes with page)
• Label Bureau (stored separately from page)

23
Meta Data as part of HTML doc

• ltHTMLgt
• ltHEADgt
• ltMETA
• HTTP-EQUIVkeywords CONTENTfederalgt
  
• ltMETA
• HTTP-EQUIVkeywords
• CONTENTtaxgt
• lt/HEADgt
• lt/HTMLgt
• Browser and/or proxy interpret the metadata

24
Metadata Apart From Doc

• Label bureaus
• Request for a doc is also a request for labels
  from one or more label bureaus
• Who makes the labels
• Text analysis
• Community of users
• Creator of document

25
Labels Collaborative Filtering
Search Engine
Label Bureau B
Labels
Author Labels
Label Bureau A
Web Site
Rating Service
26
PICS and PICS Rules

• Tools for communities to use profiles and
  control/direct access
• Structure designed by W3 consortium
• Content designed by communities of users

27
PICS Rating Data

• (PICS1-1 http//www.abc.org/r1.5
• by John Doe
• labels on 1998.11.05
• until 2000.11.01
• for http//www.xyz.com/new.html
• ratings (violence 2 blood 1 language 4)
• )

28
Using a URL List Filtering

• (PicsRule-1.1
• (Policy (RejectByURL (http//www.xyz.com/)
• Policy (AcceptIf otherwise)
• )
• )

29
Using the PICS Data

• (PicsRule-1.1
• (serviceinfo (
• http//www.lablist.org/ratings/v1.html
• shortname PTA
• bureauURL http//www.lablist.org/ratings
• UseEmbedded N
• )
• Policy (RejectIf ((PTA.violence gt3) or
  (PTA.language gt2)))
• Policy (AcceptIf otherwise)
• )
• )

30
Example Medical PICS labels

• Su UMLS vocab word 0-9999999
• Aud- audience 1-patient, 3-para, 5-GP, etc.
• Ty-information type 5-scientist, 3-patient,
  4-prod
• C-country 1-Can, 4-Afghan, etc.
• Etc.
• Ratings(su 0019186 aud 35 Ty 3 C 1)

31
User Profiles for Labels

• Rules for interpreting ratings
• Based on
• User preferences
• User access privileges
• Who keeps these
• Who updates these
• How fine is the granularity

32
Labels and Digital Signatures

• Labels can also be used to carry digital
• Signature and authority information

33
Example

• (''byKey'' ((''N'' ''aba21241241'')
• (''E'' ''abcdefghijklmnop'')))
• (''on'' ''1996.12.02T2220-0000'')
• (''SigCrypto'' ''aba1241241''))
• (''Signature'' ''http//www.w3.org/TR/1998/REC-DS
  ig-label/DSS-1_0''
• (''ByName'' ''plipp_at_iaik.tu-graz.ac.at'')
• (''on'' ''1996.12.02T2220-0000'')
• (''SigCrypto'' ((''R'' ''aba124124156'')
• (''S'' ''casdfkl3r489'')))
  ))

34
Proxy level (hidden)
35
Text analysis of Page content

• Proxy examines text of page before showing it
• Generally keyword based
• Profile of black and/or white keywords

36
Profiles for Text analysis

• Keywords ( weights sometimes)
• Reflect interest of user or user group
• May be used to eliminate pages
• All but
• May be used to select pages
• Only those

37
Keyword matching algorithms

• Extract keywords
• Eliminate noisy words with stop list (1/3)
• Stem (computer compute computation)
• Match to profile
• Evaluate value of match
• Check against a threshold for match
• Show or throw!

38
Stop List (35)

• the for
• of on
• and is
• to with
• in by
• a as
• be this
• will are

• from that
• or at
• been an
• was were
• have has
• it
• (27 words)

39
Matching Profile to Page

• Similarity?
• How many profile terms occur in doc?
• How often?
• How many docs does term occur in?
• How important is the term to the profile?

40
Cosine Similarity Measurement

• Profile terms weighted PW (0,1) ? importance
• Document terms weighted TW (0,1)
• frequency in doc
• frequency in whole set
• Overall closeness of doc to profile
• ?(all profile terms)TW PW
• --------------------------------------------
• ?(?(all profile terms)TW2PW2)

41
What works well?
Nothing
42
Whats the problem?

• Site Labels
• Who does them?
• Are they authentic?
• Has the source changed?
• A billion docs?
• Black and White lists
• Ditto
• Text analysis of page contents
• Poor results