Run-Time Models for Measurement

1
Run-Time Models for Measurement Control
Systems and Their Support in Ptolemy II
Agilent Technologies Research Intern Report

• Jie Liu
• EECS, UC Berkeley
• liuj_at_eecs.berkeley.edu
• 9/13/2000

2
Outline

• Overview and Classification of Run-Time Models
  for MC systems
• Run-time models in Ptolemy II
• Synchronous Dataflow
• Finite State Machine
• Real-Time Processes
• Time-Synced Discrete Event
• Composing run-time models
• Demos

new
new
3
Measurement and Control Systems are Distributed,
Real-Time, Reactive

• Distributed
• Sensor nodes
• Computational nodes
• Actuator nodes
• Communication system
• Reactive
• React to its environment at the speed of the
  environment
• Real-Time
• Directly Interact with Physical World
• Constrains on response delays

4
Run-Time Software in Computational Nodes

• Aggregation of interacting software components
• A model of run-time software defines
• What the components are
• How they execute
• How they exchange messages
• Models provide properties that can be used to
  reason about safety, liveness, performance, and
  scalability.

5
Messages in MC Systems

• Message Source
• Internal
• External
• Acquisition Style
• Push
• Pull
• Message Semantics
• Event Every event matters.
• State Only the newest state matters.

6
Event-Triggered and Time-Triggered Architectures

• What triggers a reaction?
• Event
• Unpredictable
• Interrupts
• Easy to distribute

• Time
• Predictable
• Polled
• Hard to distribute

system load
ETA
TTA
of events/second
H. Kopetz, Real-Time Systems Design Principles
for Distributed Embedded Applications
7
Scheduling in Real-Time Systems

• Static Scheduling
• Fixed order of execution (non-prioritized)
• Predictable response time
• Urgent events may be delayed
• Dynamic Scheduling
• Prioritized execution
• Static priority v.s. dynamic priority
• Preemptive or Non-preemptive

8
Run-Time Models in Ptolemy II
model message semantics trigger schedule preemptive timed
SDF event time static - no
FSM event time/ event dynamic no no
RTP state time/ event dynamic yes no
TSDE event event dynamic no yes
9
Synchronous Dataflow (SDF)
model message semantics trigger schedule preemptive timed
SDF event time static - no
FSM event time/ event dynamic no no
RTP state event dynamic yes no
TSDE event time/ event dynamic no yes
10
Synchronous Dataflow

• Components Functional blocks
• Communication FIFO queue
• Requirement Fixed consumption and
• production rate
• Execution Static scheduled
• (AAACBBD)

1
B
D
2
2
1
2
2
C
1
3
A

• Analysis
• Match well with time-triggered approach
• Not so expressive
• Hard to handle emergent events

safety liveness bounded memory response time
? ? ? ?
11
Finite State Machine (FSM)
model message semantics trigger schedule preemptive timed
SDF event time static - no
FSM event time/ event dynamic no no
RTP state event dynamic yes no
TSDE event time/ event dynamic no yes
12
Finite State Machine

• Components states
• Communication transitions
• Requirement finite states,
• atomic transitions
• Execution events trigger transitions

guard/action
A
B
C

• Analysis
• Match well with both ET and TT architectures
• Not so expressive
• Sequential

safety liveness bounded memory response time
? ? ? some
13
Real-Time Processes (RTP)
model message semantics trigger schedule preemptive timed
SDF event time static - no
FSM event time/ event dynamic no no
RTP state event dynamic yes no
TSDE event time/ event dynamic no yes
14
Real-Time Processes

• Components processes
• Communication state semantics
• Requirement static priorities
• blocking read
• Execution preemptive, event driven

B
D
C
A

• Analysis
• Match well with ET architectures
• Easy for handling urgent events
• Nondeterministic, Not predictable.

safety liveness bounded memory response time
? ? ? some
15
Time-Synced Discrete Event (TSDE)
model message semantics trigger schedule preemptive timed
SDF event time static - no
FSM event time/ event dynamic no no
RTP state event dynamic yes no
TSDE event time/ event dynamic no yes
16
Discrete Event (DE)
C
A
B

• Global notion of model time
• Components functional blocks react to input
  events
• Communication event (time_tag, data_token)
• Require Components are causal
• Execution Event-driven execution
• Global event queue, sorting events in
  their chronological order

17
Faster-Than-Real-Time Computation

• Not all events have real-world counter parts
• Map between model time and real time only when
  necessary
• If we have
• Global notion of the real time (time-sync
  protocol)
• Time-stamped sensor data
• Time-bomb feature
• We benefit
• Tolerance to communication and computation
  jitters
• Easiness of distributing and scaling up
• Possibility of distributed synchronized operations

Sensor
Actuator
Computer
x
x
18
Causality Subtlety

• Event in the past!

Sensor
Actuator
Computer
x
x

• Conditions to resolve the causality subtlety
• Synchronous/Reactive assumption
• Predictable inputs assumption
• Side-effect-free assumption
• Rollbackable computation assumption

19
Time-Synced Discrete Event

• Analysis
• Match with ET and TT architectures
• Directly reason about time
• Need infrastructure support
• Have causality subtlety

safety liveness bounded memory response time
? ? ? some
20
Example Discrete Event Control
N
NCAP
NCAP

• Excite the beam using zero-crossing events
• Time-stamped event triggering
• Time-Synced sensor, computation, and actuator

21
Example Control Law

• Time-stamped sensor data
• Estimate the peak time.
• Control magnitude by setting time bombs
• Adaptive to change of physical dynamics
• Tolerate communication and computation latency

up edge
down edge
?
?
sensor
control law
?/2
?/2
actuator
22
Composing Multiple Models
controller
actuator
sensor
b
c
smoother
actuator
a
d
mode d controller
23
Example A Data Acquisition Analysis System
N
B
A
NCAP
NCAP
NCAP

• Time-triggered and event-triggered sequential
  operations
• Time-synced sensor data acquisition
• Composition of timed and untimed models

24
Example Top-level sequential operations
ready
Settling
Data Acquisition
finish
Analysis
complete
25
Example Settling Mode

• SDF untimed model
• Streamed-data as fast as it can
• Best-effort computation
• Event detection

SDF
sensor1
max-minltd

sensor2
max-minltd
ready
26
Example Acquisition Mode

• TimeSyncDE
• Synchronized data acq
• Faster-than-real-time computation
• Time-bombed reader and writer

GlobalTime
suffix
ReadBurst1
ReadBurst2
D1
D2
D3
TimeBomb
complete
TSDE
27
Example Analysis Mode

• SDF
• Implicitly timed
• Equidistance-sampled data
• signal processing

SDF
log1
FFT
512
ramp
1
64
scope
?
log2
FFT
512
512
finish
64
28
Conclusion

• There are a variety of run-time software models
• Real-time software ? prioritized preemptive
  multitask.
• Time-Synced Architecture opens new opportunities
• Choosing models are application dependent
• Usually need to compose more than one model
• Ptolemy II is a laboratory for exploring the
  models and composition

29
Acknowledgement

• Agilent Systems and Solutions Lab
• Stan Jefferson Steve Greenbaum
• John Eidson Randy Coverstone
• Stan Woods Hans Sitte
• Jeff Burch Bruce Hamilton
• Jerry Liu
• Ptolemy Group

THANK YOU!