Two layers of networks:
Two layers of networks Peer to Peer overlay nodes
Unmanaged, very large collection of nodes
AS domain-level overlay nodes
Managed (?) collection of nodes per AS domain
Used to short-circuit sub-optimal overlay network routing between AS domains
Peer to Peer Nodes:
Peer to Peer Nodes Roughly comparable capabilities
Interested in various kinds of links (latency/BW)
Abilene, CENIC, Internet to wide-area to DSL/Cable
Nodes provide traffic generator function
Minimal requirements
Need JVM environment
1 to many virtual nodes per physical node
Some JNI (MDW's nbio/sandstorm)
Storage requirements (100 Mbyte to 1 Gbyte/node)
AS-level Overlay Nodes:
AS-level Overlay Nodes 1-2 per domain? More?
“Overlay-on-overlay” network between AS domains
Better managed/stable
Introduction function, primary persistent storage
Minimal requirements (more powerful!)
More memory/CPU
More persistent storage (order 10-100 GBytes/node)
High BW network connection (route at wire speed)
More than one physical network connection
Computational resources
Content transcoding, storage mgmt algorithms
Crypto hardware?
Same execution env as lower-level P2P nodes
Application Set:
Application Set OceanStore: Global information storage
Multimedia information dissemination
Sahara: Telco-type services
Chat/IM/Telephony
1-way/2-way broadcast (Bayuex, Scribe)
Lightweight data storage (Mnemosyne, Past)
Policy-based overlay routing (Brocade)
i3: Internet indirection infrastructure
Basic comm primitives: multicast, anycast, mobility, service composition
Mobile nodes (Mobile Tapestry)
PDAs, phones, etc.
Some Experiments:
Some Experiments P2P clients
Scalability measurements
Topology discovery (Latency/BW usage)
Introduction mechanisms
Supernode election protocols
Fault-tolerance
Node faults, link faults, P2P lifetimes
Data availability/reliability/reachability
Load balancing
Flash crowds on one or many nodes
Rebalancing after faults
Mobility
Insertion/appearance times
Rapidly moving clients (single and groups)
Management Issues:
Management Issues Distributed software update/startup
Rapid wipe and re-install
Millennium rootstock-model?
Single floppy (+network access) to recreate
Statistics/trace gathering/collection
Privacy, legal issues?
Distributed debugging?
Routing policies
BW limits vs peer
QoS
Minimal resource guarantees?
Perhaps there are not a lot of packets of type X floating around, but want P2P-X to start whenever packet arrives?