Unmanaged Internet Protocol
Download
Report
Transcript Unmanaged Internet Protocol
Unmanaged Internet Protocol
Taming the Edge Network Management crisis
Bryan Ford
Presented by –Lizhao Ding
Today’s Internetworking challenges
Federation of edge networks
Ubiquitous network
What it means
User’s own environment
wherever and whenever
Appropriate security precautions
Require: Self-management
Why it doesn’t work yet
IP routing need skilled management
Ad-hoc routing doesn’t scale well
Motivations
Problems with IP
Hierarchical address architecture
Routable addresses allocation
Nodes addresses assignment
Static
DHCP
Topology-dependent
ABR: Good scalability, but bad
usability
Motivations
Problems with Ad-Hoc routing
Manages topology changes well
But doesn’t scale (hundreds of nodes)
Storage limits
Bandwidth overhead limits
Good self-management, but bad
scalability
Architecture
What we need --UIP
ad-hoc’s self-management
IP’s scalability
Architecture – “3.5 layer”
Node identities
Cryptographic hash of a node’s Public key
Globally unique
Stable
Self-authenticating
Topology-independent
Multiple identifiers-Virtual hosts
UIP Routing
Less important to find best possible route, but
more important to be scale, robust, and fully selfmanaging
O(logN) storage and update requirements
Average found routes within 2x best possible path
Routing algorithm
Extends Kademlia DHT, Virtual link
Analogous to aggregation
Routing mechanism
Sorts neighbours into buckets (proximity):N(10110010)
Maintain a connectivity invariant
At least one neighbour in Bi, if there is any node M in
the network: proximity (N, M)=i
Exmaples: 10111101, 10110110, Nt(10101010) ?
Check N1(10101100)’s neighbour table ’s B5
If not empty(N2), build virtual link to N2 via N1,and
establish the neighbourship between N and N2
Recursively searching until setting up the
neighbourship between N and Nt
Conclusion
Ubiquitous networks need a protocol that works well not
only in small edge networks, but also in the big
interconnected federation of edge networks
Hierarchical address architecture and ABR are unsuitable
due to the management expense.
Ad-hoc routing doesn’t scale well
Scalable IBR should be feasible, but still in research
More about UIP routing in another paper: Scalable
Internet Routing on Topology-Independent Node
Identities
URL: http://www.brynosaurus.com/pub/os/uip.pdf