Web Caching

1
Web Caching

• By
• Amisha Thakkar
• Alpa Shah

2
Overview

• What is a Web Cache ?
• Caching Terminology
• Why use a cache?
• Disadvantages of Web Cache
• Other Features
• Caching Rules

3
Overview

• Caching Architectures
• Comparison of Architectures
• Cache Deployment Scheme
• Client Side Cache Cooperation
• Active Caching

4
What is a Web Cache ?

• Cache is a place where temporary copies of
  objects are stored
• Cached information is generally closer to the
  requester than the permanent information is
• Objects -HTML pages, images, files

5
What is a Web Cache?
6
Caching Terminology

• Client - An application program that establishes
  connections for sending requests
• Server- An application program that accepts
  connection to service requests by sending back
  responses
• Origin Server-The server on which the given
  resource resides or is to be created

7
Caching Terminology

• Proxy- An intermediary program which acts both as
  a server and a client which requests on behalf
  of the other clients
• Proxy is not necessarily a cache
• Proxy does not always cache the replies
  passing through it
• It may be used on a firewall to monitor
  accesses

8
Why use a cache ?

• To reduce latency
• To reduce network traffic
• Load on origin servers will be reduced
• Can isolate end users from network failures

9
Disadvantages of Web cache

• With cached data there is always a chance of
  receiving stale information
• Content providers lose access counts when cache
  hits are served
• Manual configuration is often required
• Operation of cache requires additional resources
• In some situations the cache can be a single
  point of failure

10
Other Features

• Depending on the perspective the following may be
  good or bad
• Cache requests on behalf of clients the
  servers never see the clients IP addresses
• Cache provides an easy opportunity to
  monitor and analyze browsing activities
• Cache can be used to block certain requests

11
Types of Web Caches

• Proxy caches
• Serve a large number of users
• Large corporations and ISPs often set
  
• them up on the firewalls
• They are type of shared caches
• Browser caches
• Use a section of the computers hard disk
• to store objects that you have seen

12
Caching Rules

• Rules on which caches work -
• Some of them set in protocols
• Some are set by cache administrator
• Most common rules
• If the object is authenticated or secure it
• wont be cached
• Objects headers indicate whether the
• object is cacheable or not

13
Caching Rules

• Object is considered fresh when -
• ? It has an expiry time or other age
• controlling directive set is still
  
• within the fresh period
• ? If the browser cache has already seen
• the object has been set to check
• once a session

14
Caching Rules

• ? If a proxy cache has seen the object
• recently it was modified relatively
• long ago
• Fresh documents are served directly from the
• cache without checking with the origin server

15
Caching Rules

• For a stale object , the origin server will
• be asked to validate the object , or tell
  the
• cache whether the copy is still good
• The most common validator is the time
• that the object was last changed

16
Caching Architectures Hierarchical /Simple Cache

• Browser-cache interaction is same as browser
  -host interaction, i.e. a TCP connection is made
  item requested
• If not found send request to parent cache
• Hierarchy built up - each level serving
  indirectly a wider community of users

17
Caching Architectures Hierarchical /Simple Cache
18
Caching Architectures Distributed /Co-operating
Cache

• Decentralized(Cache Mesh)
• Multiple servers cooperate in such a way that
  they share their individual caches to create a
  large distributed one
• Simply put caching proxies communicating with
  each other to serve different users
• On a cache miss, it checks with other proxy
  caches before contacting the origin server

19
Caching Architectures Distributed /Co-operating
Cache

• Caches communicate amongst themselves using a
  protocol like ICP (Internet Cache Protocol)
• Caches can be selected on the basis of
• Distances from the end user
• Specialize in particular URLs(location hint).

20
Caching Architectures Distributed /Co-operating
Cache

• Why Distributed - limitations of hierarchy
• Width of cache in hierarchy caches at same
  level are inaccessible to each other
• LRU policy implies sufficient disk space
• Cost in replication of disk storage
• Amount of disk space reqd. depends on number
  of users served breadth of reading

21
Caching Architectures Distributed /Co-operating
Cache

• More the users ? more disk space higher in the
  hierarchy
• Exponential growth of number of documents on
  WWW

22
Caching Architectures Distributed /Co-operating
Cache

• Caching close to user - more effective, higher
  the level lower the efficiency
• Can be created for load balancing
• Most effective when serving a community of
  interests

23
Caching Architectures Distributed /Co-operating
Cache

• First an UDP packet sent for cache inquiry.
• Cache selection decision is determined by RTT
• Potential problem -network congestion because of
  UDP
• In favor-
• UDP exchange 2 IP packets, TCP at least 8
  packets

24
Caching Architectures Distributed /Co-operating
Cache

• UDP reply from cache can indicate
• a. Presence
• b. Speed
• c. Availability of requested documents

25
Caching Architectures Hybrid Cache

• Note ICP

26
Comparison of Architectures

• Hierarchical caches placed at multiple levels
• Distributed caches only at bottom level no
  intermediate caches

27
Comparison of Architectures

• Performance parameters.
• ? Connection time (Tc)is defined as the time
  since the document is requested first data byte
  is received
• ? Transmission time (Tt)is defined as the time
  taken to transmit the document
• ? Total latency Tc Tt .
• ? Bandwidth usage

28
Comparison of Architectures

• Fig 3 -Connection time for different documents
  popularity

29
Comparison of Architectures

• For unpopular documents high connection time
• No of requests increases ? avg.. connection time
  decreases
• For extremely popular documents distributed has
  smaller connection times

30
Comparison of Architectures

• Fig 4 Network traffic generated

31
Comparison of Architectures

• On lower levels, distributed caching practically
  double the network bandwidth usage
• Around the root node in national network, the
  network traffic is reduced to half
• Distributed caching uses all possible network
  shortcuts between institutional caches,
  generating more traffic in the less congested low
  network levels

32
Comparison of Architectures

• Fig 5 a, Not congested national network

33
Comparison of Architectures

• The only bottleneck on the path from the client
  to the origin server is the international path.
  Hence transmission times are similar for both

34
Comparison of Architectures

• Fig 5 b Congested National Networks

35
Comparison of Architectures

• Both have higher transmission times compared to
  the previous case
• Distributed caching gives shorter transmission
  times than hierarchical because many requests
  travel through lower network levels

36
Comparison of Architectures

• Fig 6 Average total latency

37
Comparison of Architectures

• For large documents transmission time is more
  relevant than connection times
• Hierarchical caching gives lower latencies for
  documents smaller 200 KB due to lower connection
  times
• Distributed caching gives lower latencies for
  larger documents due to lower transmission times

38
Comparison of Architectures

• The size- threshold depends on the degree of
  congestion in national network
• Higher the congestion, lower is the size-
  threshold
• Distributed caching has lower latencies than
  hierarchical

39
Comparison of ArchitecturesWith Hybrid Scheme

• Fig 7 connection time

40
Comparison of ArchitecturesWith Hybrid Scheme

• Fig 8.

41
Comparison of ArchitecturesWith Hybrid Scheme

• In the hybrid scheme if the number of
  cooperating caches (kc) is very small , the
  connection time is high
• When number of cooperating caches increases, the
  connection times decreases up to a minimum
• If the number increases over the threshold , the
  connection time increases very fast

42
Comparison of ArchitecturesWith Hybrid Scheme

• Fig 9 Transmission time

43
Comparison of ArchitecturesWith Hybrid Scheme

• For un-congested n/w the no.of coop caches (kt)
  at every level hardly influences Tt
• If no. of coop caches is very small , high Tt
  vice -versa
• If the no increases above the threshold the Tt
  increases
• Optimum no. of caches depends on the no of caches
  reachable avoiding congested links

44
Comparison of ArchitecturesWith Hybrid Scheme

• Fig 10

45
Comparison of ArchitecturesWith Hybrid Scheme

• Fig 11 total latency

46
Comparison of ArchitecturesWith Hybrid Scheme

• The no. of coop caches(kopt) at every level
  depend on the document size to minimize the total
  latency
• For small documents the optimum no. is closer to
  kc
• For large documents the the optimum no. is closer
  to kt

47
Comparison of ArchitecturesWith Hybrid Scheme

• Fig 12

48
Comparison of ArchitecturesWith Hybrid Scheme

• For any document the optimum kopt that minimizes
  the total latency is such that kc? kopt?kt

49
Cache Deployment Schemes

• Proxy caching

50
Cache Deployment Schemes

• Advantages
• ? Clients point all web requests directly to
  cache no effect on non web traffic
• ?Cost of upgrading h/w s/w is limited
• ? Administration on caches limited to basic
  configuration

51
Cache Deployment Schemes

• Disadvantages
• ?Every browser must be configured to point to
  the cache
• ?Each client can hit only one cache
• ?Single point of failure
• ? Unnecessary duplication of data
• ? Bottleneck in cases where content is otherwise
  available in LAN

52
Cache Deployment Schemes

• Transparent Proxy caching

53
Cache Deployment Schemes

• Advantages
• ?No browser configuration
• ?Cost of upgrading h/w s/w is limited
• ?No administration of intermediate systems
  required

54
Cache Deployment Schemes

• Disadvantages
• ? Each client can hit only one cache
• ?If cache goes down internet as well as
  intranet access lost
• ? Negative impact on non web traffic
• ? Cache has to route non web traffic
• ? Routing ,packet examination n/w addr.
  translation steal CPU cycles from the main cache
  serving function

55
Cache Deployment Schemes

• Transparent proxy caching with web cache
  redirection.

56
Cache Deployment Schemes

• Advantages
• ?Switch/ router examines the packets
• ?Minimal impact on non-web traffic
• ?Frees up CPU cycles for the web cache
• ? Allows client load to be dynamically spread
  over multiple caches
• ? Eliminates single point of failure especially
  if redundant redirectors are used

57
Cache Deployment Schemes

• Disadvantages
• ?Additional intermediate systems must be
  deployed
• ? Increases expense

58
Client Side Cache Cooperation.
59
Active Caching

• Current problem unable to cache dynamic documents
• Caching Dynamic contents on the web using active
  web
• Cache applet is server supplied code that is
  attached with an URL , or collection of URLs
• Applet is written in platform independent language

60
Active Caching

• On a user request the applet is invoked by the
  cache
• The applet decides what is to be sent to the user
• Other functions of the applet-
• Logging user accesses
• Checking access permissions
• Rotating advertising banners

61
Active Caching

• The proxy has the freedom to not invoke the
  applet but send the request to the server
• Proxy promises to not send back a cached copy
  without invoking the applet
• If applet too huge ,send request to server
• Proxy not obligated to cache any applet , in that
  case agrees to not service the request for that
  document

62
Active Caching

• Proxy can devote resources to the applets
  associated with the hottest URLs to its user
• Proxy that receives the request is typically the
  proxy closest to the user , the scheme
  automatically migrates the server processing to
  the nodes that are close to users
• Thus increasing the scalability of web based
  services