• An Observation:
  • The same information gets transmitted over the network many times
• Some Solutions:
  • Cache at the client
  • Cache at the server
  • Cache at intermediaries

• The Idea:
  • Keep a local copy
  • Check source to see if the content has changed
• Questions:
  • How much to cache?
  • When to check? (e.g., every request, every day)

• The Idea:
  • Keep frequently requested items in memory
• Questions:
  • How much to cache?
  • Cache replacement (e.g., least recently used, least frequently used)
  • When to update?

• Proxy Caching:
  • Reverse Proxies/Accelerators
    • Close to the server
  • Forward Proxies
    • Close to the client
    • Often implemented by "enterprises" (e.g., JMU)
    • Reduce traffic and/or decrease latency
• Content Distribution Networks:
  • Replicate content at multiple locations

• Free:
  • Coral CDN
  • FreeCast
  • PeerCast
  • SquidCache
  • Xunlei
• Commercial:
  • Akamai
  • CacheFlow
  • EdgeCast
  • Inktomi
  • SkyCache

• The Concept:
  • Replicate content at multiple locations to improve performance and ensure that the content is "always" available
• The Challenges:
  • Where to locate replicas?
  • How to replicate content?
  • How to find content (and choose between replicas)?

• Lowest Load?
  • Balances load on servers
• Best Performer?
  • To improve performance
• Geographically Close?
  • To improve performance
• Any Live Server?
  • To improve fault tolerance

• Routing (e.g., IP anycast):
  • Advantages: Transparent
  • Disadvantages: Complex, doesn't scale
• Application Layer (e.g., HTTP redirect):
  • Advantages: Fine-grained control
  • Disadvantages: Additional load, application-specific
• Naming/Directory Based (e.g., DNS selection):
  • Advantages: Reduces RTTs
  • Disadvantage: Domain-based not URL-based

1. Client fetches HTML document from source
   • GET index.html from jmu.edu
2. Appropriate replicating server is identified
   • va.mycdn.com
3. URLs of replicated content are replaced in response
   • jmu.edu/images/quad.gif is replaced with va.mycdn.com/jmu.edu/images/quad.gif
4. Client fetches va.mycdn.com/jmu.edu/images/quad.gif
   • The image may or may not have been cached on va.mycdn.com
   • If the image is not on va.mycdn.com, it acts as a proxy and requests it from jmu.edu

• Canonical Name Records:
  • CNAME records indicate that a domain name is an alias
• How They Can Be Used:
  • Delegate name resolution to a CDN that can find a nearby replicating server

1. Root server gives a nameserver (NS) record for Akamai
   • akamai.net
2. The akamai.net nameserver returns an NS record for a nearby client
   • b.akamai.net
3. The b.akamai.net nameserver returns an appropriate nearby server
   • a1105.b.akamai.net

dig www.bestbuy.com
;; ANSWER SECTION:
www.bestbuy.com.  3600     IN  CNAME   www.bestbuy.com.edgesuite.net.
www.bestbuy.com.edgesuite.net. 21600 IN   CNAME   a1105.b.akamai.net.  
a1105.b.akamai.net.   20   IN   A   198.7.236.235  
a1105.b.akamai.net.   20   IN   A   198.7.236.240  
;; AUTHORITY SECTION:  
b.akamai.net.   1101  IN   NS   n1b.akamai.net.  
b.akamai.net.   1101  IN   NS   n0b.akamai.net.  
;; ADDITIONAL SECTION:  
n0b.akamai.net.   1267 IN  A    24.143.194.45  
n1b.akamai.net.   2196 IN  A    198.7.236.236