Focus on Distributed Hash Tables n Distributed hash
Focus on Distributed Hash Tables n Distributed hash tables (DHT) provide resource locating and routing in peer-to-peer networks – But, more than object locating and routing – Basically, a routing network n Outline – Three papers • Mobile Grid • Broadcast • An architecture based on DHT for grid resource location and allocation for massively distributed applications – One implementation • Grid resource location and allocation
Wireless Grid n Vision: provide a easy way to access resource in the grid – PUT and GET: put/get a resource from/into the grid – Fundamental service: routing! n Comprise of stationary Internet hosts and wireless devices – Wireless devices can change their network attachment points n A potential application: NCHC’s Eco. Grid – Internet severs + sensors in the wild n Our solution to provide a wireless grid – Based on distributed hash tables (DHT)
Wireless Grid (cont’d) Nodes can change their network addresses n Research problem n – Given the network bandwidth budget (Bi) of each node i to maintain the DHT – How to efficiently utilize Bi of each node i to provide efficient routing • By efficiency, the route takes nearly zero timeouts An example of DHT Failed route from 1 to 12 Successful route
Wireless Grid (cont’d) n Preliminary results – PHT: percentage of hop counts due to timeouts – PBT: percentage of maintenance bandwidth due to timeouts n Manuscripts in preparation – Mobility churn in DHTs – Nearly, optimal peer state supervision for mobile DHTs Our sol.
Broadcast in Dynamic P 2 P Networks n Broadcast is a basic building block for many P 2 P applications – – n Message broadcast Query routing in unstructured P 2 P networks Diagnostics for overlay network structure … Broadcast in P 2 P networks – Each message has a TTL value – TTL decreases when the message is forwarded every time – A message is not forwarded if TTL = 0
Broadcast … (cont’d) n Research problem – Given a TTL value – How to maximize the broadcast scope while to minimize the maintenance bandwidth? n Intuition behind: to maximize the broadcast scope requires the overlay network being maintained robustly!
Broadcast … (cont’d) n There is no easy way to maximize the broadcast scope and meanwhile to minimize the maintenance bandwidth – We investigate several scalable failure detection and recovery mechanisms in the context of DHT – Paper under reviewed: scoped broadcast in dynamic peer-to-peer networks
P 2 P Mechanism for Grid Resource Location and Allocation n Overview
P 2 P Mechanism for Grid Resource Location and Allocation (cont’d) n Organizes the allocated resources as a tree-like overlay n Submit tasks using Armada agents – Send the agents to the tree root – The tree root helps distribute the agent codes to its children
Implementation n Targeted environment: NCHC Uni. Grid n Status: – A DHT Prototype for a cluster environment is available – P 2 P mechanism for Grid resource location and allocation – Hopefully, can be tested in Uni. Grid
- Slides: 10