Synthesis of Coordination Controllers for Distributed Embedded Systems

1
Synthesis of Coordination Controllers for
DistributedEmbedded Systems

• Pai H. Chou
• Dept. of ECE
• University of CaliforniaIrvine, CA

Gaetano Borriello Dept. of CSE University of
Washington Seattle, WA
2
Complex embedded systems

• Increasing complexity
• cars have 50-100 networked processors
• cost conscious
• Need abstraction
• components IP reuse
• compatible protocols at different levels
• Approaches
• embedded platforms
• interface-based

3
Matching protocols

• Match protocols
• insert glue for protocol translation
• modify component
• Protocol stack
• physical, datalink, network,transport,
  session...
• coordination

easy to decouplefrom component
hard to decouple
4
Coordination behavior

• Explicitly described
• signals betw. processes
• synchronization, mode change,arbitration, state
  coherence ...
• Implicitly defined
• (formal) models of computation
• Dataflow, CSP, ...
• platform (implementation-defined) semantics
• operating system, virtual machine, ..

5
Problem statement

• Component model
• adaptable components
• composition operators (coordination protocols)
• Synthesis of composition mechanism
• coordination controllers ( scheduler)
• target distributed architecture
• Objectives
• optimization for communication efficiency
• minimize controller complexity

6
Previous work

• Component model dac98
• modal process
• modes as part of composition interface
• adaptable to different coordination protocols
• Definition of coordination protocols (ACTs)
• functions on mode changes
• one mode change implies other changes
• constrain the state space spanned by modes
• Composition iccad98
• mode manager synthesis

7
Mode manager

• Run-time support
• maintains configuration bit vector
• handlers vote, evaluates ACTs
• generates mode entry/exit events

mode manager
modal processes
8
Architectural mapping

• Single processor or multiple processors
• Multiple mappings to an architecture

modal processes
9
Distributed mode managers

• Automatically partitioned among processors
• synthesized control communication
• comm. tradeoffs synchronization, replication

modal processes
10
Distributed coordination

• Minimum replication
• lower bound on controller size
• more communication
• Maximum replication
• upper bound on controller size
• fail to exploit dont cares
• Optimal -- somewhere in between
• least communication traffic
• smallest controller size

11
Projection optimization

• Interprocessor ACTs
• modes on processors
• max-flow min-cut

send B
12
Conclusions

• Coordination controllers
• for integration of complex components
• separate coordination from component behavior
• Distributed coordination
• minimize communication traffic
• reduce controller complexity
• On-going work
• application to component-based power management

13
(No Transcript)
14
Summary

• Specification by composition
• modal processes adaptable components
• ACTs declarative constraints on modes
• Synthesis of mode manager
• run-time mechanism for ACTs
• handles distributed coordination
• Supports design space exploration

15
Future work

• Specification
• integrate control w/ dataflow
• catalog of ACTs
• Implementation
• code generation, optimizations
• hardware mode manager
• Formal verification
• high-level knowledge (ACTs) composition

16
Conclusions

• Modal process model
• design by control composition
• separate app-specific reusable behavior
• enables synthesis of error-prone detail
• Design space exploration
• heterogeneous distributed architectures
• choice of synchrony and communication

17
Interprocessor ACT replication

• Min replic. gt frequent communication
• Max replic. gt wider channel

modal processes
18
Approaches to components

• Platform based
• HW boards, system architecture
• SW OS, middleware
• Embedded platforms
• tied to implementation ideosyncrasy
• Interface based
• protocol determines interoperability
• amenable to optimizations

19
Related work

• Data composition
• Ptolemy heterogeneous dataflow domains
• Control (reactive) specification
• Statechart, Esterel synthesis, target HW/SW
• no reuse of control behavior
• Distributed objects
• RMI, shared data, synchronization
• no pluggable coordination protocol

20
Our approach

• Interface-based component model
• decouples coordination from components
• more reusable
• Synthesis of coordination controller
• distributed target architecture
• optimize for given processor allocation
• communication efficiency
• minimize controller complexity

21
Abstract control types (ACTs)

• Constraints on modes
• one mode change implies other changes
• constrain the state space spanned by modes
• Usage
• intra-component behavior (reactive)
• adaptation (wrapping)
• inter-component coordination (integration)

22
ACT example
p
m1
m3

• parent(p, m1N)

x
m2
m2
m2
input condition
input votes
output votes

-p
-m1N
case 1
p
mi
p
case 2
p
m2
m3
m1
23
Propagating full replication
D
D
Joy
E
E
M
M
M
Pilot
A
A
S
F
Bumper
R
T
P
P
Sonar
O
O
Fd
Fd
Ht
Ht
Wheels
Ml
Ml
Rv
Rv
Tn
Tn
24
Unnecessary comm in full rep
D
D
Joy
-D
E
E
M
M
Pilot
A
A
wasted communication sending -D
without changing mode on processor 2
S
S
F
F
Bumper
R
R
T
T
P
P
Sonar
O
O
Fd
Fd
Ht
Ht
Wheels
Ml
Ml
Rv
Rv
Tn
Tn