Peer Net Pushing PeertoPeer Down the Stack Jakob
Peer. Net: Pushing Peer-to-Peer Down the Stack Jakob Eriksson, Michalis Faloutsos, Srikanth Krishnamurthy University of California, Riverside
What is Peer. Net • an approach to separate address and identity – address reflects the node’s current location – identifier remains the same • a p 2 p-based network layer for large networks – an alternative to the IP layer
Peer. Net Goals • • targeting large, wireless and mobile networks emphasis on mobile nodes and p 2 p interactions minimizing the need for manual configuration favoring distributed and p 2 p solutions instead of centralized solutions and node specialization • localizing control overhead
. . . and Challenges • while location-based addressing simplifies routing, new challenges are created: – consistent dynamic address allocation – efficient node lookup service for finding addresses for given identifiers => the project proposes distributed solutions
Peer. Network layer (1/2) • dynamic addressing, depending on the node’s current location in the network • addresses organized as a binary tree • selecting addresses so that locality of communication is guaranteed (subtree/area)
Peer. Network Layer (2/2) • Area invariant – all nodes belong to a nested sequence of areas, one of each level – all nodes within an area share a unique address prefix – l bit addresses l area levels – violating node resigns & rejoins • Three major parts of the network layer – address allocation – routing – node lookup
Address Allocation in General • addresses assigned dynamically, preserving the invariant • goals: – balancing the tree & effective utilization of address space – good physical connectivity within an area – improving scalability & minimizing control overhead
Address Allocation: Joining the Network • • physical connection requesting address registering to node lookup service moving node => new request & registration
Address Allocation: Details • each node controls a range of addresses • responding node splits range in half – joining node gets the lower half – joining node’s address is the lowest address in that range – nodes get evenly distributed in the address space
Address Allocation: Challenges • 1. Keeping the tree balanced – local congestion leads to address unavailability – solution: proactively migrating nodes • 2. Maximizing intra-area connectivity – solution: nodes within an area have to be wellconnected by physical links => improves routing performance and tolerance to link failures, desirable in mobile networks
Routing • sender only needs to know the identifier of the receiving node • address provided by the lookup service • packets include both the identifier and last known address – address rechecked along the route if necessary
Routing: Details • routing one level at a time • each step takes closer to destination in both topology and address tree • a distance vector type approach: – each node has l=log N entries, one for each level – each entry stores the next hop to corresponding sibling subtree • destination address compared to node address bit by bit, MSB first, packet then forwarded to corresponding sibling subtree
Node Lookup Service (1/3) • • distributed node lookup for finding addresses for given node identifiers Storing lookup entries (id, addr) – choosing a node with minimum xor-distance (address XOR identifier) • Challenges – node movement => address change => lookup entries need to be updated or moved
Node Lookup Service (2/3) • Solutions: – Preserving locality of lookups • multiple locations for lookup tables, chosen so that local queries get local answers • iterating from local scope to larger subtrees – Creating locality of updates • moving nodes change their low bits more often Þstoring LSB locally and MSB in remote nodes Þremote nodes need less updating ÞMSB found first, query then forwarded to corresponding area
Node Lookup Service (3/3) • Communication abstractions – identifier mapping to multiple addresses Þeffective implementation of multicast/anycast For example: – multicast group id & group of receiving nodes – service id & group of providing nodes
Further Directions of the Project • • • security performance issues tree balancing and local connectivity portability leveraging various infrastructures – tunneling through the Internet to connect Peer. Nets – interconnection with Internet nodes – TCP/IP emulation on Peer. Net nodes • actual data / results / comparison ?
- Slides: 16