Please add to and reorganize this file!

Future Work Ideas from P2 Papers

SIGCOMM '05

• Use query optimization ideas to inform new routing protocols
• Automated (multi-)query optimization
• Expressing BGP inter-domain routing policies
• Expressing overlay networks

SOSP '05

• Other overlays beyond Chord: epidemics, link-state, path-vector, other DHTs.
• Sharing & multiquery optimization in the context of overlays
• Transport layer optimizations
• Online debugging (P2 is "astonishingly capable" of doing it!)
• Express security invariants for each element in isolation, and compose them into global properties for the whole system.
• Make the execution of an overlay tamper-evident
• Reason about safety, termination, and other program properties
• Language extensions for high-level specification of sharing, transport, debugging and security logic.
• An overlay-specific language, rather than general Datalog.

SIGMOD '06 Submission

• Cost-based optimizations including random walks
• Adaptive query processing to react to network dynamics
• Multiquery optimization for overlays
• Incorporate negation into our language model and implementation (affects on pipelining and dynamic data?).
• Static program checks

Integration of existing group research

• Integration of Maelstrom-style induced churn or other secure routing
• Implementation of ProofSketch aggregation
• Implementation of SIGMOD06-style pipelined semi-naive evaluation
• Distributed triggers a la Ankur Jain
• Gossipy things like Dahlia's

Applications, etc.

• Traditional protocol implementations in P2 (BGP, ISIS, OSPF, TCP, etc.)
• Ephemeral overlays
• Distributed relational operators (join, agg, etc.)
• P2P queries over WinFS
• Distributed triggers
• Simple firewall sharing
• Host firewall clustering
• Guestrin/Paskin Belief Propagation (or Loopy Belief Propagation) with application to knowledge-plane-like inference
• Rebuild PHI demo

Query Optimizations

• An extensible, cost-based optimizer architecture
• Automatic rewrites including aggregate selections, magic sets, predicate reordering
• Cache selection
• Cost-based optimizations based on neighborhood functions and/or expander properties

Engine Architecture

• P2 w/ Capriccio lightweight threads
• Coordination extensions to the language
• P2 for clusters: MapReduce and the like
• On-line reconfigurability of dataflow (e.g., to accommodate on-line optimizations).

Crazy Ideas

• Aspect-oriented distributed systems. Write the security logic, the transport logic, the storage logic, etc., in isolation and have the system weave them together into a single dataflow graph. Might be related to Synthesis.
• On-line checking for protocol compliance. For example, check at runtime that a particular BAN logic is being followed in the authentication aspect of the system.
• Distributed taint tracking. Newsome and Song's NDSS 05 paper on taint checking looks for the ways that untrusted inputs propagate through program code (using valgrind), and then tracks when those values get used in suspicious ways. Observation 1: this is a transitive closure computation that could be easily expressed in datalog, as in Whaley and Lam's bdddb system. Observation 2: one could imagine combining this with a distributed propagation facility across machines, ala P2. This could be done either via forward tracking in real time (to confirm suspicious behavior) or via backtracking into historical logs (in the spirit of the Xie, et al work on Random Moonwalks from Oakland 05