C. Kim, M. Caesar, and J. Rexford. Floodless in SEATTLE: a scalable ethernet architecture for large enterprises. In SIGCOMM, 2008.
This paper introduces SEATTLE which is an advanced scalable layer-2 network architecture with the plug-and-play and manageability of Ethernet and the scalability of IP. The Ethernet is not an scalable structure; it uses flooding for locating end-hosts, uses spanning-tree basesd routing and relies on broadcast for bootstarping protocols (ARP and DHCP). The usage of IP on top of Ethernet in order to emulate the extension of layer2 hasn't been really working well since IP addresses are hierarchical and topology dependent. A time-varying topology requires continuous monitoring and modification of routers and subnetting configurations. Non-mobile IP does not support seamless mobility!
The SEATTLE architecture utilizes the difference in the timescales of dynamics of the system. Switches are quite static and much fewer in number rather than end-hosts. Therefore SEATTLE uses a link-state protocol to enable shortest path routing between switches. For address resolution and location resolution a one hope DHT is used to provide the directory service. Furthermore observing that end-hosts usually connect to few others only, authors claim that caching is a good strategy in order to keep the routing tables limited in size. The results are verified through simulation and limited deployment.
I found this paper to be quite an interesting one with full description of the methods they have used to eliminate the causes of non-scalability of Ethernet. The idea of using DHT is quite interesting even though it is not obvious what can happen if one of the switches are hacked! it seems like there is too much trust given to possibly more than hundreds of routers involved in this architecture. It will also be interesting to discuss the similarities and differences between the mobile IP approach to seamless transition versus and SEATTLE approach.
This paper introduces SEATTLE which is an advanced scalable layer-2 network architecture with the plug-and-play and manageability of Ethernet and the scalability of IP. The Ethernet is not an scalable structure; it uses flooding for locating end-hosts, uses spanning-tree basesd routing and relies on broadcast for bootstarping protocols (ARP and DHCP). The usage of IP on top of Ethernet in order to emulate the extension of layer2 hasn't been really working well since IP addresses are hierarchical and topology dependent. A time-varying topology requires continuous monitoring and modification of routers and subnetting configurations. Non-mobile IP does not support seamless mobility!
The SEATTLE architecture utilizes the difference in the timescales of dynamics of the system. Switches are quite static and much fewer in number rather than end-hosts. Therefore SEATTLE uses a link-state protocol to enable shortest path routing between switches. For address resolution and location resolution a one hope DHT is used to provide the directory service. Furthermore observing that end-hosts usually connect to few others only, authors claim that caching is a good strategy in order to keep the routing tables limited in size. The results are verified through simulation and limited deployment.
I found this paper to be quite an interesting one with full description of the methods they have used to eliminate the causes of non-scalability of Ethernet. The idea of using DHT is quite interesting even though it is not obvious what can happen if one of the switches are hacked! it seems like there is too much trust given to possibly more than hundreds of routers involved in this architecture. It will also be interesting to discuss the similarities and differences between the mobile IP approach to seamless transition versus and SEATTLE approach.
Fair criticism, but what happens if an ethernet switch is hacked in an existing environment? Given that switches flood, participate in global algorithms like spanning trees, it seems to me that a rogue switch can have a big impact on the enterprise network.
ReplyDelete