Internet Scalability

Florin Coras talks in this lecture about the Inter-Domain Routing Problem and representative solutions. The Inter-Domain Routing Problem allows the exchange of data between peers along the best path that possibly crosses several transit provider domains and fulfills the routing policies of each domain independent of its network topology. Each peer is an Autonomous System. The Border Gateway Protocol (BGP) is a commom inter-AS routing protocol and his principal function is to exhange network reachability info with other BGP systems.

 Each AS announces only what it considers its best path and each domain is able to define its own routing policy thanks to BGP. In practice there are two policies: customer-provider peering and shared-cost peering. The decission process consists in select preferred routes (local-pref) manually configured, select shortest AS path route (topology dependent) and, In case of ties, use tie-breaking rules.

About the BGP Routing Tables we have the BGP Routing Information Base (RIB) that aggregates all BGP reachability announcements and saved in control plane memory. We also have the BGP Forwarding Information Base (FIB). The output of the BGP decision process ran on the RIB. It contains one route per destination prefix, used when forwarding packets and saved in data plane memory (fast memory).

The routers need to store information about all the destinations: prefixes and the AS. There so many prefixes due to the multihoming, the traffic engineering and the IANA allocation policies. With the existing mechanisms the ingress TE is problematic. Like the reachability is announced to all Internet the size of the FIB can produce a memory problem and the updates propagates to everybody can produce CPU limitations.

The Florin talks about the possible solutions: The Disruptive (clean-state) solutions and the evolutionary solutions.

- Clean State Architectures (NNC): the problems are the content availability, the security, the content location dependence, where vs what, named hosts vs named data and the host-to-host vs many-to-many.

The communications are built on named data with no notion of the hosts

The content node model has two types of packets: the interest and the data. Both are one for one and the data packets carry security information.

The forwarding engine: exists the Forwarding Information Base (FIB), the content store and the Pending Interest Table (PIT)

Transport: operates on top of unreliable packet delivery services, has a flow control and the CNN can take advantages of multiple interfaces. The CNN names are composed of components and have a relative access to data in totally ordered tree.

Routing (Intra-Domain): works with link state IGP (OSPF and IS-IS). Can customize the link-state and have a behavioral difference from IP. BGP has the equivalent of a IGP TLV and he topology at the AS rather than network prefix.

Implementation: is in C and Java, the interest and the data packets are sent over UDP and the VoCCN implementation is done with linphone. Conclusions: is less efficient than TCP but better than HTTPS CNN and HTTP.