1
Real-Time JavaWhat Could and Should it Mean?

• E. Douglas Jensen
• The MITRE Corporation
• jensen_at_real-time.org, jensen_at_mitre.orghttp//www.
  realtime-os.com

2
Real-Time JavaDifferent Strokes for Different
Folks

• Real-time Java, hard real-time Java, and
  soft real-time Java are ambiguous terms
• currently used with very different definitions
• often used without any definitions
• Could mean very different things
• cf. motorized road vehicle sports car,
  18-wheeler, pickup, van, motorcycle
• Different users need very different things

3
Outline

• Operate in real-time, real-time resource
  management
• Time constraints hard, soft
• Scheduling optimization criteria hard, soft
• Predictability and determinism
• Hard real-time Java, soft real-time Java
• Distributed Java, distributed real-time Java
• Real-time enhancements needed for Java

4
Real-Time Computing

• Entity operates in real-time
• degree to which the entity
• satisfies time constraints acceptably
• collective optimality (scheduling optimization
  criterion)
• predictability
• Entity does real-time resource management
• degree to which the entity
• explicitly takes time constraints into account
• (vs. resource supply gtgtgt resource demand)

5
Time Constraints

• Time constrained thread activities
• Utility for performing the activity in terms of
  when the activity is completed
• Completed no loss of generality
• Used by the scheduler
• (Progressive utility out of scope here)

6
Hard Deadline Time Constraint
7
Soft Deadline Not Hard

• In practice and in this model
• Conventional real-time fails to distinguish among
  different types of soft deadlines
• Different types have different attributes
• In practice, most deadlines are soft
• This model is more general and expressive

8
Scheduling Theory Deadline
9
Invert Lateness to Earliness
10
Time Constraints That Are Not DLs
11
General (Soft) Time Constraints
12
Time Constraint Scoped Attribute

• Scoped attribute of a thread
• static entire lifetime of the thread
• dynamic threads time constraints change
• Thread scheduling
• inside scope criterion includes time constraints
• outside scope only other factors in criterion

13
Scheduling Events

• Thread transition into time constraint scope
• Completion of time constrained activity
• Deadline or time constraint terminal times
• (Presumes dynamic preemptive scheduling)

14
Scheduling

• Contention for a serially reusable resource ?
  scheduling
• Scheduling optimization criterion
• scheduling algorithm/heuristic
• collective/aggregate timeliness
• multiple factors composite
• Real-time resource management to the degree that
  it includes time constraints

15
Other Optimization Factors

• Threads relative importance
• Thread dependencies
• precedence
• data consistency
• resource utilization efficiency

16
Optimal Schedule

• Each activity completes during interval of
  maximum utility
• Not feasible in general
• dynamics of application and system
• tractability of algorithm/heuristic
• In general, time constraint factors are
  necessarily sub-optimal

17
Hard Real-Time Scheduling

• One time constraint factor hard deadlines
• Only optimality criterion meet all deadlines
• No provisions for
• sub-optimality (so either feasible or incorrect)
• soft deadlines or general time constraints
• Presumes static enough execution context
• Most applications have softer time constraints,
  and dynamic contexts

18
Soft Real-Time Scheduling

• Soft real-time scheduling not hard
• time constraints may be any mix of
• hard deadlines
• soft deadlines
• general time constraints
• any optimality criterion
• provisions for sub-optimality
• any dynamic execution context
• Many different types of soft real-time scheduling
  with different attributes

19
Sub-Optimal for Hard Deadlines

• Sub-optimal hard deadline factors
• minimize the number of missed deadlines
• miss deadlines according to importance
• Composite hard deadline factors
• always meet these deadlines, and
• minimize the number of those deadlines that are
  missed
• miss deadlines according to importance

20
General Optimality Criteria

• Optimality criteria for soft deadlines
• minimize mean lateness
• minimize summed tardiness
• amortize tardiness according to importance
• Optimality criteria for general time constraints
• maximize the accrued (e.g., summed) utility

21
Fixed and Dynamic Priorities

• Fixed assigned á priori (programmer, tool)
• e.g., RMA
• Computing field allows run-time adjustment of
  priorities for scheduling anomalies
• Dynamic priorities may change at run-time
• e.g., earliest-deadline-first

22
Static Scheduling

• Schedule created á priori
• list ordered by static eligibility fixed
  priority
• highest priority ready thread granted the
  resource
• Computing field allows run-time adjustment of
  priorities for scheduling anomalies
• Rate monotonic analysis is normally performed
  statically

23
Dynamic Scheduling

• Schedule created at execution time
• Normally uses dynamic priorities
• e.g., according to earliest deadline first
• can use fixed priorities if
• other optimization factors (e.g., importance) are
  dynamic
• fixed priority (e.g., rate monotonic) scheduling
  performed at execution time

24
Predictability

• Something is predictable to the degree it is
  known in advance
• one end-point is determinism known absolutely
• other end-point is maximum entropy no knowledge
• The degree to which an outcome or behavior is
  predictable implies the degree to which its
  future context is predictable(the converse is
  not generally true)

25
Determinism in Hard Real-Time

• Hard real-time scheduling may have two different
  outcomes deterministically
• schedule optimality
• weaker but normally sufficient
• the most frequent objective
• completion times
• stronger but generally infeasible and unnecessary

26
Determinism in Soft Real-Time

• Soft real-time scheduling may have four different
  deterministic outcomes
• schedule optimality
• weakest but normally sufficient
• the most frequent objective
• utility value optimality
• utility values
• completion times
• strongest but generally infeasible and unnecessary

27
Predictability In Stochastic Cases

• Stochastic characteristics of the threads and
  their execution environments are considered to be
  random variables
• Scheduling is optimized in expectation
• A predictability measure coefficient of
  variation C? variance/mean2
• one endpoint deterministic distribution, C? 0
• the other endpoint EME distribution, C? ?

28
Predictability vs. Optimality

• Predictability of schedule optimality is
  orthogonal to schedule optimality per se
• They must be traded off against one another to
  suit specific application/situation requirements
  e.g., better ? but worse ?
• Hard real-time scheduling is a special case
  which, if feasible, is optimal in both dimensions

29
Hard and Soft Real-Time Java

• Hard
• deterministic timeliness
• deterministic execution context
• Soft
• everything else (but not que sera sera)
• non-deterministic predictability of timeliness
• non-deterministic execution context
• many different possible cases

30
Hard Real-Time Java

• Hard real-time scheduling
• hard deadlines
• deterministic optimal scheduling (meet all hard
  deadlines)
• Deterministic execution context
• current and future load characteristics
• threads, resource usage, dependencies, conflicts
• resource management
• memory, garbage collection, events, i/o

31
Soft Real-Time Java

• Soft real-time scheduling
• soft deadlines or general time constraints
• non-deterministic (or deterministic) optimal or
  sub-optimal scheduling
• Non-deterministic (or deterministic) dynamic
  execution context
• load
• resource management

32
Examples of Hard Real-Time Java

• JEM?
• More? to be completed . . .

33
JEM

• to be completed . . .

34
Examples of Soft Real-Time Java

• NewMonics PERC
• to be completed . . .

35
NewMonics PERC