Implementing declarative overlays

1
Implementing declarative overlays

• Boom Thau Loo
• Tyson Condie
• Joseph M. Hellerstein
• Petros Maniatis
• Timothy Roscoe
• Ion Stoica

2
Presentation

• Introduction
• P2 approach
• Overlog examples (NARADA mesh)
• Implementation
• Performance
• Conclusions

3
INTRODUCTION

• Overlay networks are used in variety of
  distributed systems.
• P2 a facility, deployable as a service or
  library, for the declarative construction,
  maintenance, and sharing of an overlay network
• Applications concise a logical description of an
  overlay network, and P2 executes this to maintain
  routing data structures, perform resource
  discovery and forwarding functionality.
• Goal simplify the process of selecting,
  implementing, deploying and evolving an overlay
  network design.

4
INTRODUCTION

• P2 does not aim as optimized performance as e.g.
  C, C or Java implementations.
• CHORD 47 logical rules (full Overlog
  specification), MACEDON 320 statements.
• P2 Simplified approach (16 rules, NARADA)
• Target is to demonstrate that declarative overlay
  descriptions can be implemented by P2 with
  acceptable performance.

5
P2 approach

• A combination of protocol- and structure centric
  approach that would be most attractive for
  overlay specification and runtime.
• P2 employs 2 types on relations
• Soft-state tables (network state)
• Streams of transient tuples (stream query
  engines)

6
Overlog

• P2 specification language for overlays.
• Overlog designed for distribution of context data
  and messages.
• Overlog P2 adds constructs to specify physical
  distribution properties.
• Tuples are physically generated, stored and sent
• Queries over streams
• Deletion of tuples

7
NARADA

• Summary of functionality (16 rules)
• Each node maintains a set of neighbors, and a set
  of members in the group.
• Each member epidemically propagates keep-alive
  messages for itself
• (Associated to monotonically increased
  sequence number)
• Neighbors exchange information about membership
  liveness and sequence numbers.
• (members learn the liveness of each member)
• If member fails to hear direct neighbor for a
  period, it declares neighbor dead.
• (updates its own membership state and
  propagates the rest population)
• Periodical updates to measure roundtrip latency.
• If latency time improves -gt node A adds node B to
  its neighbor set
• If latency time exceeds -gt node A removes node B
  from its neighbor set

8
Example of NARADA based Overlog declarations

• Overlog syntacts ltruleIDgtltheadgt - ltbodygt
• Neighbor and member are tables whose tuples are
  retained for 120 seconds
• materialize(neighbor, 120, infinity, keys(2)).
• materialize(member, 120, infinity, keys(2)).
• materialize(sequence, infinity, 1, keys(2)).
• Rule (R1)
• R1 refreshEvent(X) - periodic(X, E, 3).
• Table refreshEvent has a row with value (X),
  for any X, if table periodic has a row with value
  (X, E, 3), for some E.

9
P2 Implementation

• Platform Linux
• Code C
• Codelines 20k ( 3rd party support)

10
Block diagram of a P2 node

11
Performance (FIG. 3)

12
Performance (FIG. 4)

13
Conclusion

• The P2 approach looks promising
• Overlays can be specified concisely
• Performance is acceptable
• Overlog allows easy modification

14
Thank You!