1 / 9

Overlay Networks

Overlay Networks. Outline Overlay networks overview Routing overlays Resilient Overlay Networks Content Distribution Networks. Overlay Networks Overview. Logical network on top of the physical network Underlying idea: Let the applications make their own forwarding decisions

vallejoc
Download Presentation

Overlay Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Overlay Networks Outline Overlay networks overview Routing overlays Resilient Overlay Networks Content Distribution Networks CS 640

  2. Overlay Networks Overview • Logical network on top of the physical network • Underlying idea: Let the applications make their own forwarding decisions • Enables us to introduce new functionality into the network • How it works: • Each node processes and forwards packets in an application specific way • The links are implemented as tunnels *Image from: http://www.cs.virginia.edu/~mngroup/hypercast/general.html CS 640

  3. Routing Overlays • Can we try routing algorithms not supported by router vendors? • YES! Using Overlays. • Experimental Versions of IP • IP multicast deployed in Mbone and 6BONE as an overlay • Multicast-aware routers tunnel through legacy routers • End System Multicast • Lets the end hosts implement their own multicast trees • The goal is to find the embedded multicast tree that spans all group members CS 640

  4. End System Multicast Cont. • Using unicast naïvely is not efficient: • Optimal, by DVMRP: • The best we can get with end system multicast: A A C C R1 R1 R2 R2 B B D D A C R1 R2 B D CS 640

  5. Managing the Overlay • For joining the network, the host needs to know the IP address of at least one other node in the network • A connected node sends periodic “keep alive” message to its neighbors • Each node exchanges the list of all nodes that it knows of with its neighbors • These messages propagate through the mesh similar to the Distance Vector algorithm • Before leaving the mesh, a node sends a “leave mesh” message to its neighbors • Once a node detects a partition, it creates a new edge by sending a “join mesh” message • End hosts measure the round-trip latency to other nodes • Add new links or drop links based on these measurements CS 640

  6. Resilient Overlay Networks • Idea: To find alternative routes for traditional unicast applications • Why do we need alternative routes? • BGP Does not guarantee shortest paths • Tradeoff: • Getting optimal routes vs. scalability • BGP scales well, but does not give optimal routes • RON can give optimal routes, but does not scale well • The bad: Needs active monitoring of links -> Expensive • Monitors Latency, available bandwidth, and loss probability • The good: Fast recovery from network failures CS 640

  7. Content Distribution Networks • CDNs use replication to overcome bottlenecks in the underlying system • Geographically distribute server surrogates that cache pages in backend servers • Helps increase response time in flash crowds • Data distribution • Servers proactively replicate their data across surrogates • If a surrogate does not have a page, it asks the backend server • Only static and not dynamic pages are distributed CS 640

  8. Redirectors • Redirectors forward client requests to the appropriate server. • Objective is to • Reduce response time • Maximize system throughput: # Requests satisfied in a specific period • Implementing Redirection • DNS-based: Augment DNS to return different addresses. • This requires use of short TTLs. • Http redirect: The server responds with a new/better server • Incurs an additional round-trip latency • Servers can be overloaded with the redirection task CS 640

  9. Policies • Naïve: Round Robin, or select servers randomly • Network proximity: Minimize response time • One way to implement is to choose servers from the same ISP as the client • Balance the load: Maximize system throughput • Locality: Select a server that is likely to have the page in cache • One way to implement is to use hashing to map URLs to servers • Finding the right balance between these criterion is a subject of ongoing research. CS 640

More Related