Dev Roadmap Milestones

Bloom Development Milestones, 2012. Rev 0.

Mainline Bloom Runtime Environments

Lead: Srinivasan

1. Completion of Bud v0.1.0. Target: SW release 2/1

   • current push-based implementation cleaned up, including docs, sandbox and benchmarks.
   • address NTT benchmark concerns
   • Note: I had originally thought that I'd optimize the current design, but I think it must be done in the new setting.

2. Design/prototype of new architecture for Bloom 1.0. Target: SW release 6/1

   • Restructured Bloom language as a standalone piece, not a DSL embedded in a host language. Will still follow Ruby conciseness.
     • Typed schemas for primitives, but "Object" type allowed. Explicit casting will of course have to be allowed.
     • type and monotonicity annotations for external and built-in functions
     • expression language.
     • better module composition.
   • multi-language target design, with well-defined Bloom-level keywords/functions and host-language access to collections.
     • first prototype target to Scala, just so that I am not spending inordinate amounts of time with two different language environments.
     • concentration on getting the design of Joins correct and performant
   • compiler written in Java/Scala.
   • rigorous performance goals and testing

3. Second language target for Bloom 1.1. Target: SW release 11/1

• Javascript backend, including the equivalent of the sandbox bundled as a package.
• One option is to have someone from IIT here do an internship and add stuff to the package.

CALM Program Analysis Toolkit.

Lead: Alvaro. Target: SW release and research paper, 3/1

1. Visual debugger

   • visual dataflow diagrams with monotonicity analysis (points of order)
   • visualization of trace-driven flow weights to drive program variants
   • visual timestep debugging

2. Code synthesis

   • augmentation of programs with customized coordination logic for points of order that minimizes the participants in the coordination
   • augmentation fo programs with taint-tracking logic and optional "boundary memory" for flows lacking CALM coordination

3. Case study

   • configurable-consistency key-value store or an HDFS-style filesystem

Bloom Core

Lead: Alvaro. Target: SW Release 2/1

Replaces bud sandbox with tested, documented suite of core distributed systems libraries. Building off work done in the Fall 2011 Program the Cloud course.

• delivery protocols
  • best-effort
  • reliable
  • ordered
  • reordered
  • drop-inducing
• membership
• multicast
• alarms
• heartbeats
• counters
• queues
• voting
• vector clocks
• leader election
• consensus protocols
  • 2-phase commit
  • multipaxos
• concurrency protocols
  • 2-phase locking
  • timestamp
• read-write quora with session guarantees
  • read-your-writes
  • monotonic reads
  • majority

Lattice-Enabled Bloom

Lead: Conway. Target: SW code freeze and research paper, 2/1.

1. Language and runtime

• Bud extensions to support arbitrary lattices and morphisms between them
• Suite of core lattice types including Sets, MaxInts, BooleanLattices and Vectors
• morphisms between key lattice types

2. CALM analysis via lattices

• overcome specific roadblocks in logic-based CALM analysis
  • aggregate threshhold logic
  • shopping carts example
• revisit Bloom Core with lattices
  • simplify, analyze, optimize

Haskell Bloom Runtime

Lead: Marczak. Target: unknown

Experimental runtime for lattice-enabled Bloom using rich Haskell type system and lazy evaluation.