Lesson 12 (CDNs and Overlay Networks)

Question 1

What are the two approaches to CDN server placement?

Answer 1

1) Enter Deep - Deploy smaller server clusters “deep” into the access networks around the world.
Pro: minimizes distance between a user and the closest server cluster
Con: More difficult to manage and maintain
2) Bring Home - Place fewer larger server clusters at key points (typically in IXPs).
Pro: Less server clusters to manage/maintain
Con: User experience higher delay and lower throughput

-CDNs can also employ a hybrid approach.

Question 2

How does a CDN operate?

Answer 2

1) User visits a content provider’s website and navigates to the web page hosting the target video.
2) User clicks on the video link and the user’s host sends a DNS query for the applicable domain (eg “video.examplemovies.com”)
3) The DNS query goes to the user’s local DNS server (LDNS) which issues an iterative DNS query to the authoritative DNS server owned by the content provider. The authoritative DNS server knows where the requested domain is stored in the CDN and sends back the host name in the CDN’s domain (eg “1130.examplecdn.com”).
4) The user’s LDNS performs an iterative DNS query for the provided CDN host name which returns an IP address.
5) The users LDNS returns the CDN provided IP address to the user.
6) The user’s client directly connects via TCP to the IP address provided by the user’s LDNS and then send an HTTP GET request for the video.

Question 3

What are 2 cluster selection strategies?

Answer 3

1) Pick the geographically closest cluster

2) Pick based on real-time measurements vs static cluster selection policy

Question 4

What is the policy for server selection once a cluster has been selected?

Answer 4

• Map the requests based on the content.

- Requests for the same piece of content can be mapped to the same machine by using some sort of content-based hashing.

Question 5

What is consistent hashing and why is it relevant for server selection?

Answer 5

• A distributed hash table
• Balances load by assigning roughly the same number of content IDs, and requires relatively little movement of these content IDs when nodes join and leave the system.
• The servers and content objects are mapped to the same ID space so if one server is offline, the next server can deliver the content.
• Provide to be optimal which means that the least number of keys need to be remapped to maintain load-balance on an average.

Question 6

What are the network protocols used for cluster/server selections?

Answer 6

1) DNS - Recursive and iterative DNS queries are used by a client to find the applicable IP address/server within a CDN cluster. 1st query is recursive and the remaining queries are iterative.
2) IP Anycast - Routes a client to the “closest” server as determined by BGP. In different clusters, they all have a server with the same IP address.
3) HTTP Redirection - Server redirects a GET request. Helpful for when a content server is receiving a large amount of requests.