|
Slide 49 -
|
The Challenges of Embedded System Design Edward A. Lee
Robert S. Pepper Distinguished ProfessorUC Berkeley
Invited Talk
Xilinx Emerging Technology Symposium (ETS)
San Jose, CA
February 1, 2012 Key Collaborators on work shown here:
Steven Edwards
Jeff Jensen
Sungjun Kim
Isaac Liu
Slobodan Matic
Hiren Patel
Jan Reinke
Sanjit Seshia
Mike Zimmer
Jia Zou Abstract All widely used software abstractions lack temporal semantics. The notion of correct execution of a program written in every widely-used programming language today does not depend on the temporal behavior of the program. But temporal behavior matters in almost all systems, particularly in networked systems. Even in systems with no particular real-time requirements, timing of programs is relevant to the value delivered by programs, and in the case of concurrent and distributed programs, also affects the functionality. In systems with real-time requirements, including most embedded systems, temporal behavior affects not just the value delivered by a system but also its correctness.
This talk will argue that time can and must become part of the semantics of programs for a large class of applications. It will argue that temporal behavior is not always just a performance metric, but is often rather a correctness criterion. To illustrate that this is both practical and useful, we will describe recent efforts at Berkeley in the design and analysis of timing-centric software systems. In particular, we will focus on two projects, PRET, which seeks to provide computing platforms with repeatable timing, and PTIDES, which provides a programming model for distributed real-time systems. Courtesy of Kuka Robotics Corp. Cyber-Physical Systems (CPS):Orchestrating networked computational resources with physical systems Courtesy of Doug Schmidt Power generation and distribution Courtesy of
General Electric Military systems: E-Corner, Siemens Transportation
(Air traffic control at SFO) Avionics Telecommunications Factory automation Instrumentation
(Soleil Synchrotron) Daimler-Chrysler Automotive Building Systems Claim For CPS, programs do not adequately specify behavior. A Story The Boeing 777 was Boeing’s first fly-by-wire aircraft, controlled by software. It is deployed, appears to be reliable, and is succeeding in the marketplace. Therefore, it must be a success. However…
Boeing was forced to purchase and store an advance supply of the microprocessors that will run the software, sufficient to last for the estimated 50 year production run of the aircraft and another many years of maintenance.
Why? Lesson from this example: Apparently, the software does not specify the behavior that has been validated and certified!
Unfortunately, this problem is very common, even with less safety-critical, certification-intensive applications. Validation is done on complete system implementations, not on software. Problems that complicate analysis of system behavior: Structure of a Cyber-Physical System Messages from different sources interleave nondeterministically Sensors may be locked out for an indeterminate amount of time Plat Variability of execution times affects results (not just WCET) Interrupt-driven I/O disrupts timing Platforms’ measurements of time differ A fault in a remote component may disrupt a critical local activity A fault in a remote component may go undetected for a long time Etc… A Key Challenge:Timing is not Part of Software Semantics Correct execution of a program in C, C#, Java, Haskell, OCaml, etc. has nothing to do with how long it takes to do anything. All our computation and networking abstractions are built on this premise.
Programmers have to step outside the programming abstractions to specify timing behavior. Execution-time analysis, by itself,does not solve the problem! Analyzing software for timing behavior requires:
• Paths through the program (undecidable)
• Detailed model of microarchitecture
• Detailed model of the memory system
• Complete knowledge of execution context
• Many constraints on preemption/concurrency
• Lots of time and effort
And the result is valid only for that exact hardware and software!
Fundamentally, the ISA of the processor has failed to provide an adequate abstraction. Wilhelm, et al. (2008). "The worst-case execution-time problem - overview of methods and survey of tools." ACM TECS 7(3): p1-53. Our first goal is to reduce the problem so that this is the only hard part. Part 1: PRET Machines PREcision-Timed processors = PRET
Predictable, REpeatable Timing = PRET
Performance with REpeatable Timing = PRET = PRET + Computing With time // Perform the convolution.
for (int i=0; i<10; i++) {
x[i] = a[i]*b[j-i];
// Notify listeners.
notify(x[i]);
} Dual Approach Rethink the ISA
Timing has to be a correctness property not a performance property.
Implementation has to allow for multiple realizations and efficient realizations of the ISA
Repeatable execution times
Repeatable memory access times
Example of one sort of mechanism we would like: tryin (500ms) {
// Code block
} catch {
panic();
} jmp_buf buf;
if ( !setjmp(buf) ){
set_time r1, 500ms
exception_on_expire r1, 0
// Code block
deactivate_exception 0
} else {
panic();
}
exception_handler_0 () {
longjmp(buf)
} If the code block takes longer than 500ms to run, then the panic() procedure will be invoked.
But then we would like to verify that panic() is never invoked! Pseudocode showing the mechanism in a mix of C and assembly. Extending an ISA with Timing Semantics [V1] Best effort: set_time r1, 1s
// Code block
delay_until r1 [V2] Late miss detection set_time r1, 1s
// Code block
branch_expired r1,
delay_until r1
set_time r1, 1s
exception_on_expire r1, 1
// Code block
deactivate_exception 1
delay_until r1
[V3] Immediate miss detection [V4] Exact execution: set_time r1, 1s
// Code block
MTFD r1 To provide timing guarantees, we need implementations that deliver repeatable timing Fortunately, electronics technology delivers highly reliable and precise timing…
… but the overlaying software abstractions discard it. Chip architects heavily exploit the lack of temporal semantics. // Perform the convolution.
for (int i=0; i<10; i++) {
x[i] = a[i]*b[j-i];
// Notify listeners.
notify(x[i]);
} To deliver repeatable timing, we have to rethink the microarchitecture Challenges:
Pipelining
Memory hierarchy
I/O (DMA, interrupts)
Power management (clock and voltage scaling)
On-chip communication
Resource sharing (e.g. in multicore)
Hardware
thread Hardware
thread Hardware
thread Our Current PRET ArchitecturePTArm, a soft core on aXilinx Virtex 5 FPGA Hardware
thread registers scratchpad memory I/O devices Interleaved pipeline with one set of registers per thread SRAM scratchpad shared among threads DRAM main memory, separate banks per thread memory memory memory Note inverted memory compared to multicore!
Fast, close memory is shared, slow remote memory is private! Multicore PRET In today’s multicore architectures, one thread can disrupt the timing of another thread even if they are running on different cores and are not communicating!
Our preliminary work shows that control over timing enables conflict-free routing of messages in a network on chip, making it possible to have non-interfering programs on a multicore PRET. Status of the PRET project
Results:
PTArm implemented on Xilinx Virtex 5 FPGA.
UNISIM simulator of the PTArm facilitates experimentation.
DRAM controller with repeatable timing and DMA support.
PRET-like utilities implemented on COTS Arm.
Much still to be done:
Realize MTFD, interrupt I/O, compiler toolchain, scratchpad management, etc. A Key Next Step:Parametric PRET Architectures ISA that admits a variety of implementations:
Variable clock rates and energy profiles
Variable number of cycles per instruction
Latency of memory access varying by address
Varying sizes of memory regions
…
A given program may meet deadlines on only some realizations of the same parametric PRET ISA. set_time r1, 1s
// Code block
MTFD r1 Realizing the MTFD instruction on a parametric PRET machine The goal is to make software that will run correctly on a variety of implementations of the ISA, and that correctness can be checked for each implementation. set_time r1, 1s
// Code block
MTFD r1 PRET Publications S. Edwards and E. A. Lee, "The Case for the Precision Timed (PRET) Machine," in the Wild and Crazy Ideas Track of DAC, June 2007.
B. Lickly, I. Liu, S. Kim, H. D. Patel, S. A. Edwards and E. A. Lee, “Predictable programming on a precision timed architecture,” CASES 2008.
S. Edwards, S. Kim, E. A. Lee, I. Liu, H. Patel and M. Schoeberl, “A Disruptive Computer Design Idea: Architectures with Repeatable Timing,” ICCD 2009.
D. Bui, H. Patel, and E. Lee, “Deploying hard real-time control software on chip-multiprocessors,” RTCSA 2010.
Bui, E. A. Lee, I. Liu, H. D. Patel and J. Reineke, “Temporal Isolation on Multiprocessing Architectures,” DAC 2011.
J. Reineke, I. Liu, H. D. Patel, S. Kim, E. A. Lee, PRET DRAM Controller: Bank Privatization for Predictability and Temporal Isolation (to appear), CODES+ISSS, Taiwan, October, 2011.
S. Bensalem, K. Goossens, C. M. Kirsch, R. Obermaisser, E. A. Lee, J. Sifakis, Time-Predictable and Composable Architectures for Dependable Embedded Systems, Tutorial Abstract (to appear), EMSOFT, Taiwan, October, 2011 http://chess.eecs.berkeley.edu/pret/ Part 2: How to get the Source Code? The input (mostly likely C) will ideally be generated from a model, like Simulink or SCADE. The model specifies temporal behavior at a higher level than code blocks, and it specifies a concurrency model that can limit preemption points. However, Simulink and SCADE have naïve models of time. Problems that complicate analysis of system behavior: Recall Structure of a Cyber-Physical System Messages from different sources interleave nondeterministically Sensors may be locked out for an indeterminate amount of time Plat Variability of execution times affects results (not just WCET) Interrupt-driven I/O disrupts timing Platforms’ measurements of time differ A fault in a remote component may disrupt a critical local activity A fault in a remote component may go undetected for a long time Etc… Ptides: First step: Time-stamped messages. Messages carry time stamps that define their interleaving Ptides: Second step: Network time synchronization GPS, NTP, IEEE 1588, time-triggered busses, etc., all provide some form of common time base. These are becoming fairly common. Assume bounded clock error Assume bounded clock error e Assume bounded clock error e Ptides: Third step:Bind time stamps to real time at sensors and actuators Input time stamps are ≥ real time Input time stamps are ≥ real time Output time stamps are ≤ real time Output time stamps are ≤ real time Messages are processed in time-stamp order. Clock synchronization gives global meaning to time stamps Global latencies between sensors and actuators become controllable, which enables analysis of system dynamics. Ptides: Fourth step:Specify latencies in the model Model includes manipulations of time stamps, which control latencies between sensors and actors Actuators may be designed to interpret input time stamps as the time at which to take action. Feedback through the physical world Ptides: Fifth stepSafe-to-process analysis (ensures determinacy) Safe-to-process analysis guarantees that the generated code obeys time-stamp semantics (events are processed in time-stamp order), given some assumptions. Assume bounded network delay d Assume bounded clock error Assume bounded clock error e An earliest event with time stamp t here can be safely merged when real time exceeds t + s + d + e – d2 Assume bounded clock error e Assume bounded sensor delay s Application specification of latency d2 Ptides Schedulability AnalysisDetermine whether deadlines can be met Schedulability analysis incorporates computation times to determine whether we can guarantee that deadlines are met. Deadline for delivery of event with time stamp t here is t – c3 – d2 Deadline for delivery here is t Assume bounded computation time c1 Assume bounded computation time c3 Assume bounded computation time c2 PtidyOS: A lightweight microkernel supporting Ptides semantics PtidyOS runs on
Arm (Luminary Micro)
Renesas
XMOS
Occupies about 16 kbytes of memory.
Luminary Micro
8962 An interesting property of PtidyOS is that despite being highly concurrent, preemptive, and EDF-based, it does not require threads. A single stack is sufficient! The name “PtidyOS” is a bow to TinyOS, which is a similar style of runtime kernel. Renesas 7216 Demonstration Kit XMOS development board with 4 XCores. Workflow Structure
HW Platform Software Component Library Ptides Model Code Generator
PtidyOS Code Plant Model Network Model HW in the Loop Simulator Causality Analysis Program Analysis Schedulability Analysis Analysis Mixed Simulator Ptolemy II Ptides domain Ptolemy II Discrete-event,
Continuous, and
Wireless domains Luminary Micro
8962 IEEE 1588 Network time protocol A Typical Cyber-Physical SystemPrinting Press Application aspects
local (control)
distributed (coordination)
global (modes)
Open standards (Ethernet)
Synchronous, Time-Triggered
IEEE 1588 time-sync protocol
High-speed, high precision
Speed: 1 inch/ms
Precision: 0.01 inch
-> Time accuracy: 10us Bosch-Rexroth Goal: Orchestrated networked resources built with sound design principles on suitable abstractions 32 Source: http://offsetpressman.blogspot.com/2011/03/how-flying-paster-works.html Example – Flying Paster Source: http://offsetpressman.blogspot.com/2011/03/how-flying-paster-works.html Flying Paster Printing Press – Model in Ptolemy II Model by Patricia Derler Plant model
+
Distributed Controllers 5 Siemens CKI Project Review Patricia Derler Printing Press – Model in Ptolemy II Model by Patricia Derler Printing Press – Model in Ptolemy II Model by Patricia Derler Platform independent model of functional and
timing behavior Determinate timing at sensors and actuators Platform independent model of functional and
timing behavior Determinate timing at sensors and actuators XMOS
Predictable timing
Multiple cores
No analog I/O
No FPU
No hardware clock Renesas
PHY chip for accurate timestamping of inputs,
Analog I/O
Renesas vs. XMOS: Measured I/O timing Simulation Renesas XMOS Oscilloscope traces on GPIO pins Simulation Renesas XMOS input input output Renesas vs. XMOS:I/O timing Oscilloscope traces on GPIO pins Renesas vs. XMOS: Busy vs. Idle Time Simulation Renesas XMOS Oscilloscope traces on GPIO pins Ptides Publications Y. Zhao, J. Liu, E. A. Lee, “A Programming Model for Time-Synchronized Distributed Real-Time Systems,” RTAS 2007.
T. H. Feng and E. A. Lee, “Real-Time Distributed Discrete-Event Execution with Fault Tolerance,” RTAS 2008.
P. Derler, E. A. Lee, and S. Matic, “Simulation and implementation of the ptides programming model,” DS-RT 2008.
J. Zou, S. Matic, E. A. Lee, T. H. Feng, and P. Derler, “Execution strategies for Ptides, a programming model for distributed embedded systems,” RTAS 2009.
J. Zou, J. Auerbach, D. F. Bacon, E. A. Lee, “PTIDES on Flexible Task Graph: Real-Time Embedded System Building from Theory to Practice,” LCTES 2009.
J. C. Eidson, E. A. Lee, S. Matic, S. A. Seshia and J. Zou, “Time-centric Models For Designing Embedded Cyber-physical Systems,” ACES-MB 2010.
J. C. Eidson, E. A. Lee, S. Matic, S. A. Seshia, and J. Zou, Distributed Real-Time Software for Cyber-Physical Systems, To appear in Proceedings of the IEEE special issue on CPS, December, 2011. http://chess.eecs.berkeley.edu/ptides/ Conclusions Today, timing behavior is a property only of realizations of software systems.
Tomorrow, timing behavior will be a semantic property of programs and models. Raffaello Sanzio da Urbino – The Athens School Overview References:
Lee. Computing needs time. CACM, 52(5):70–79, 2009
Eidson et. al, Distributed Real-Time Software for Cyber-Physical Systems, Proc. of the IEEE January, 2012. A Test Casefor PtidyOS Tunneling Ball Device
sense ball
track disk
adjust trajectory
This device, designed by Jeff Jensen, mixes periodic, quasi-periodic, and sporadic real-time events. Tunneling Ball Device in Action Tunneling Ball Device – 10 rps Tunneling Ball Device Mixed event sequences Periodic Events Quasi Periodic Events Sporadic Events Distributed PTIDES Relies on Network Time Synchronization with Bounded Error This may become routine!
With this PHY, clocks on a LAN agree on the current time of day to within 8ns, far more precise than older techniques like NTP.
A question we are addressing at Berkeley: How does this change how we develop distributed CPS software? Press Release October 1, 2007
|