Real-time Programming in Java

Real-time Programming in Java 12.11.2008 Czech Java User Group Prague, Czech Republic Aleš Plšek [email_address]

INRIA INRIA The French National Institute for Research in Computer Science and Control 8 research centers 1800 scientists, 1000 PhD students, 100 post-docs 150 joint research project-teams 186 million Euros budget, 20% from research contracts Industrial Relations 790 active reseach contracts 89 companies

INRIA Nord Europe, team ADAM A daptive D istributed A pplications and M iddleware Component Oriented Programming Fractal Component Model Model-Driven Engineering Service Oriented Architectures WebServices Ubiquitous computing Mobile Computing Context Oriented Programming The team 3 Profs, 2 Asist.Prof 2 Post-Doc 7 PhD students 6 R&D engineers

Me… Past (2001-2006) Master Studies, DSRG, Charles University in Prague Model Checking of Software Components Present – Since 2006 3 rd year PhD Student, INRIA ADAM Research Interests Component-Oriented Programming Real-time Java Programming Model Checking

Outline Realtime Programming What is real-time ? RT OS Developing RT Applications Realtime Programming in Java RTSJ - Realtime Java Specification Realtime Garbage Collection Future of RT Java

Real-time “ Real-time is not real-fast !” Realtime is any information processing system which has to respond to externally generated input stimuli within a finite and specified period. the correctness depends not only on the logical result but also the time it was delivered Real-time is ubiquitous Tiny (sensors), Small (pacemakers), Medium (phone), Large (avionics) Single processor vs. Multi-core vs. Distributed

Why Real-Time? Real-time Programming A little interest in Real-time from the mainstream software engineering community Deadlines, interruption handling, too low-level… Real-Time Systems Trends Large-scale, heterogeneous systems Dynamically highly adaptable systems Systems composed from hard-, soft-, and non-real-time units Many software engineering techniques can be applied in real-time domain Component oriented programming, Code generation, Model Driven Engineering, Formal Verification, etc.

Successful Stories Shipboard computing US navy Zumwalt-class Destroyer 5mio lines of Java code Red Hat Linux, RT GC the key part Avionics 787 Dreamliner saves 900kgs of weight A380 saves a half of the processing units Financial Information Systems 100ms deadlines

Hard-, Soft- and Non- Real-time Non Real-time Soft Real-time Hard Real-time “ Hard real-time is hard but soft real-time is harder.”

RT System Categories Hard vs Soft real-time Embedded systems (cars, trains, sensors) Event-triggered vs. Time-triggered Safety-critical vs. Mission-critical Systems

Let’s get Real Systems are designed for throughput , not for determinism . Problems causing non-determinism Garbage collection Failure to respect application priorities Priority inversions Uncontrolled load Methods Normal program optimization is not helpful. What we need: RT OS Deterministic programming model Assembler, Ada, C, … ?

Real-time OS RT OS features Fully preemptible kernel Fixed priority pre-emptive scheduling FIFO scheduling Different priority levels (at least 38) Priority Inheritance High-Resolution timers Real-life Example RT Linux IBM Blades hardware patch needed Priority Inheritance Illustration

Developing Real-time Systems Schedulability Analysis List all the tasks the system must perform: When they will be released for execution The deadline The cost (CPU time) The resources (including locks) that they need Task Execution Time Period Priority 1 50 100 ? 2 90 300 ? 3 55 350 ?

Rate Monotonic Analysis Schedulability Analysis 2. Analyze the system to find the priority assignment (if any) that will let the system meet all the deadlines. Rate Monotonic Analysis Task Execution Time Period Priority 1 50 100 High 2 90 300 Medium 3 55 350 Low

Developing Real-time Systems II. Programming Realtime Systems Worst Case Execution Analysis E.g. quicksort vs. mergesort Deterministic Programming model Garbage collection Thread model for preemptive scheduling C, C++, AdA

RT Systems Intricacy ABS-Systems What really happened on Mars? Mars, Pathfinder, July 4, 1997 Lost contact due to embedded software failure Ariane 5 French Guyana, June 4, 1996 $800 million embedded software failure

RT Summary Realtime Predictability Performance is not important Real-time is everywhere RealTime OS Priority Fixed Preemptive Scheduling Priority Inversion Developing Realtime Systems Schedulability Analysis Worst Case Execution (WCE) analysis Deterministic programming model C, C++, Ada

Why Java? Java Easy to use Familiar and widely used language. Libraries Well-defined semantics. Portable across platforms. Efficient development oriented.

Predictable? Not really. Hard real-time often requires microsecond accuracy.

Making Java Predictable…? Real-time Specification for Java (RTSJ, 2000) Real-time Garbage Collection (RTGC, 2006?)

Real-time Specification for Java Java-like programming model Compliant to Java SE RTSJ features: Fixed priority and round robin scheduling Realtime Threads and inter-thread communication (e.g. semaphores) User-defined interrupt handlers and device drivers — including the ability to manage interrupts (e.g., enabling and disabling) Timeouts and aborts on running threads However, Feasibility analysis : The RTSJ does not mandate any particular schedulability analysis technique

Realtime Threads Real-Time threads 28 Real-time priorities Can start at a specific or relative time / run on specified intervals or periods Runaway Threads Asynchronous event handlers

Memory Management Memory Areas Heap memory Immortal Memory Scoped Memory Immortal Memory Never released during the lifetime of the application Scoped Memory a well-defined life time enter and leave a scoped memory area all memory allocations performed from the scoped memory the allocated memory is reclaimed when no thread in scope

Scoped Memory - Example Application Logic Runnable Cross-Scope Communication

Non-Heap Real-time Thread (NHRT) NoheapRealtime threads Can not be preempted by Garbage Collector No heap memory access Immortal or Scope Memory

Benefits of Real-Time Java Rule of thumb: 1% hard real-time, 9% soft real-time, 90% non-real-time Standard Java Advantages The main reason of RTSJ’s success is that it allows mixing freely real-time and timing-oblivious code in the same platform

RTSJ Limitations Complexities Error-prone process Non-intuitive rules and restrictions Introducing a new programming style Standard Java Libraries can not be used in RT Software Engineering Aspect Ad-hoc approach No reuse, verification, formalization, etc. No adaptability, distribution support

RTSJ vs. C++ Project Golden Gate RTSJ on a Mars Rover RTSJ vs. C++ C++ : memory management, … RTSJ: scheduling API Java performance not important Bottom line… Essential vs. incidental choices Separation of concerns needed Framework: Essentials specificated Generation of language and platform- specific incidentals

RTSJ Implementations SUN RTSJ Solaris 10 or Linux with RT-Preempt patch Java RTS IBM IBM Blades Hardware supported IBM BladeCenter Red Hat Enterprise Linux, Version 4 (POSIX conformant) TimeSys, Aonix, JamaicaVM, Ovm, … Usually not fully compliant to RTSJ

RT Garbage Collection Don’t fear the garbage. Automatic defragmentation for long lived apps. Avoid manual memory management unless absolutely necessary.

RT Garbage Collection Traditional Garbage Collection Real-time Garbage Collection Time-based collectors can easily bound pauses to one millisecond. Slack-based collectors (SUN’s) ensure RT code never preempted by GC.

RT Garbage Collection Metronome Maximum pause times down to 250 microseconds Blue Wonder 150 microseconds worst case latency 15 microseconds WCL for NonHeap Solaris OS

RTGC Limitations Ensuring that the GC can keep up requires some understanding of whole-program allocation behavior.

Conclusion Real-time Determinism RT OS Programming Realtime Scheduability Analysis Worst Case Execution Analysis Real-time Programming in Java RTSJ Manual memory management Real-time Threads RT Garbage Collection Overhead, limitations

Future of Real-time Java RTSJ JSR-282 Fixing the problems of RTSJ JSR-302 Safety-Critical Java A strict subset of RTSJ Certified for Safety-Critical systems RT Garbage Collection The most practical solution However, mind the limitations Overhead Allocation behavior of applications

Further reading…. [1] Bollela, Gosling, Brosgol, Dibble, Furr, and Turnbull : The Real-Time Specification for Java . 2000 [2] Peter C. Dibble Real-Time Java Platform Programming: Second Edition, 2008 [3] Alan Burns and Andy Wellings : 2001 [4] New Adventures in Real-Time: An Interview with Greg Bollela, 2008 ( http://java.dzone.com/articles/new-adventures-in-real-time ) [5] IBM : Real-time Java Series , 2007 ( http://www.ibm.com/developerworks/views/java/libraryview.jsp?search_by=Real+time+Java+Part )

Acknowledgements Slides based on material by Jan Vitek and Andy Wellings Further information based on [1,2,3,4,5] and many others… Last but not least, the author acknowledges the following colleagues for their comments and consultations : Frederic Loiret , Michal Malohlava, Lionel Seinturier, Philippe Merle, Carlos Noguera, Martin Decky and Marek Prochazka

Real-time Programming in Java

Related slideshows

More Related Content

Real-time Programming in Java