MIT OpenCourseWare http://ocw.mit.edu 6.033 Computer System Engineering Spring 2009 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.DNS, CDNs Lecture 14 Sam Madden DNS What is the relationship between a domain name (e.g., youtube.com) and an IPaddress? DNS is the system that determines this mapping. Basic idea: You contact a DNS server, give it a query It responds with the answer, or forwards your answer on to some other server Example: Whenever a server answers a query, it adds it to a cache When you get a response, you add it to a cache Each cached value has a duration after which you are supposed to lookup the name again At the top of the DNS hierarchy are root servers Then servers for .com, .edu, etc ... Then servers for MIT, youtube, etc....Show cached dig +norec csail.mit.edu A record vs NS Record TTL Observe that A records expire much sooner than NS Records Example: csail.mit.edu digdig [email protected] mit.edu @G.GTLD-SERVERS.NET.dig csail.mit.edu @bitsy.mit.edu. How do you get your DNS server initially? - Your admin tells it to you- DHCP (Domain Host Configuration Protocol) -- you send out a broadcast onthe local link -- DHCP server responds (via another broadcast message) with: - An IP address for you to use- The IP address of a "gateway" router for you to use- The IP address of a DNS server to use - .... DNS has some fancy features: - One physical machine can have multiple names - One name can correspond to multiple IP addresses DNS load balancing DNS server picks which name to return most DNS servers cycle through these in "round robin" fashion Example: dig google.com ;; ANSWER SECTION: google.com. 282 IN A 209.85.171.100google.com. 282 IN A 74.125.45.100 google.com. 282 IN A 74.125.67.100 Content delivery networks (CDNs -- e.g., Akamai)Use DNS to provide scalability and adapt to load Suppose I have some content that is accessed a lot -- e.g., a video on youtube I can balance load amongst my local servers using DNS load balancing, but I still haveto own the servers. Puts a huge load on my servers to deliver it; can't adapt to load spikes (e..g, the"slashdot" effect.) Also, for users that are far away (e.g., in Asia or Australia), they have to download thatcontent over long distance and thin pipes, and ISPs in Australia have to pay a lot forthan bandwidth. For content -- like this video -- that is accessed repeatedly, would be be better to nothave to go all the way to San Bruno CA everytime. Idea: create a local cache Solution 1: Proxy cache. For every URL, look up in a local cache (perhaps run byyour ISP) to see if the content is there. If it is, fetch it. Otherwise, get original data. Just like DNS requests, web pages can have cache lifetimes associated with them whichproxy caches respect. Problems: - Helps ISP but not necessarily content provider- requires clients to configure their browsers to use proxy caches- Doesn't address slashdot effect Exist so called "transparent proxies" that can do this filtering automatically, but this may be distasteful to users, especially if their ISP is doing it (companies do this all the time.) Solution 2: Content Distribution Networks (e.g., Akamai) Show diagram:Give example:nytimes.comdig graphics8.nytimes.com From both MIT network and cellular network Observe that TTL for answer is very short -- Why? -- Handle slashdot effect -- can dynamically start using more and more akamai servers for a particular request Akamai -- company Gives a way for content provider to offload load from server Also helps ISP if server is inside ISP (creates an incentive for ISPs to participate!) Akamai has thousands of caches all over the world When a request for content -- like from images8.nytimes.com arrives -- it uses dns to forward user to the nearest server. Determining the "nearest" server and how many servers to allocate is their secret sauce. For dynamic content, Akamai also works. Example -- show attempt to select best route back to Akamai, maintenance ofreachability info, etc.Akamai is a more general example of something called an overlay network. An overlay is a way to create a network with new features or a different structure bybuilding on top of an existing network. Akamai creates an overlay on top of the IP network that chooses the best route fromamongst a collection of IP servers. Overlays are used widely to extend the network with new features, for example : Provide a different topology (e.g., direct connection of clients)- E.g., for administrative reasons (VPNs)- Or for performance reasons (Akamai)Provide a different addressing mechanism (e.g., content addressability, P2P) VPN example: Companies want to allow remote users or sites to have access to theircorporate Internet sites Good old days might have done this with a modem, but that's slow, and a pain to run. VPN provides a way for remote users to appear to be on internal networkwhile actually being external. Idea is to "tunnel" traffic over public internet using an overlay (client andserver inside of network): Diagram: This is a simple example, but in principle can create complex
View Full Document
Unlocking...