Java Framework for Distributed Real-Time Embedded Systems

Arya MirΛογισμικό & κατασκευή λογ/κού

28 Μαρ 2012 (πριν από 5 χρόνια και 3 μήνες)

554 εμφανίσεις

This paper presents the evaluation of a multithread distributed real-time object-oriented platform. A communication API was developed to increase functionalities of an API that implements the Real- Time Specification for Java standard, extending it to cover embedded distributed applications. Real-time restrictions of the communication are ensured by a time-out mechanism.

Java Framework for Distributed Real-Time Embedded Systems
Elias Teodoro Silva Jr
; Edison Pignaton Freitas
; Flávio Rech Wagner
Fabiano Costa Carvalho
; Carlos Eduardo Pereira
Instituto de Informática, Federal University of Rio Grande do Sul, Brazil
{etsilvajr, epfreitas, flavio, fccarvalho}
on leave from Federal Center of Technological Education of Ceará, Brazil
Electrical Engineering Department, Federal University of Rio Grande do Sul, Brazil
This paper presents the evaluation of a multithread
distributed real-time object-oriented platform. A
communication API was developed to increase
functionalities of an API that implements the Real-
Time Specification for Java standard, extending it to
cover embedded distributed applications. Real-time
restrictions of the communication are ensured by a
time-out mechanism. The API can be adapted to be
used with different underlying network and physical
mediums. The development focused on restrictive
embedded platforms with low performance and small
memory. An evaluation in terms of the fulfillment of
timing constraints, and memory footprint is presented
for a CAN-bus network. The results also demonstrate
the timely correctness provided by the communication
API running over an RTSJ implementation.
1. Introduction
Embedded real-time systems are becoming more
complex, requiring distributing facilities in order to put
processing units where their services are demanded,
thus turning control and command activities more
efficient. A lot of examples can be quoted, like
automobile control (steer-by-wire), airplane control
(fly-by-wire), or sensor networks. In this context, it is
not enough to have physically-distributed processing
units, but they also need to be able to communicate, to
solve the problem in a cooperative way.
Over the last years, Java gained popularity as a
suitable programming language for embedded and
real-time systems development. The definition of the
Real-Time Specification for Java (RTSJ) standard [1]
is the most prominent example of such popularization
in the real-time domain. The RTSJ defines an
Application Programming Interface (API) for the Java
language that allows the creation, verification,
analysis, execution, and management of real-time
threads, whose correction also depends on the
fulfillment of timing requirements. However, it does
not take in account Java distributed programming
Consequently, a Distributed RTSJ (DRTSJ) Expert
Group has been set up under the Java Community
Process [2]. An initial framework by the expert group
for integrating the RTSJ with RMI describes three
levels of integration [3]. At Level (0), real-time Java
virtual machines (RT-JVMs) communicate via
standard RMI. No guarantee of timely delivery of a
remote request can be assumed, and the programmer
must explicitly pass scheduling and release parameters
with each call. This requires no extension of either
RMI or the RTSJ. At integration Level (1), the notion
of a real-time remote object is introduced and
supported by a real-time RMI that provides timely
invocation guarantees. Level (2) of integration
augments Level (1) with distributed thread model
semantics. Borg and Wellings [4] explore facilities that
must be provided by a real-time RMI (RT-RMI),
focusing on integration level (1), as defined in [3].
Their work differs from that presented in this paper in
that they assume a real-time network and consider the
real-time aspects at a higher level, focusing on the
remote invocation of threads. Our work, in turn,
considers facilities at a lower abstraction level,
providing a unicast/broadcast mechanism to exchange
messages meeting time restrictions. Moreover, our
development is focused on embedded platforms with
restricted performance and tight memory resources,
such as those used in control applications, while RT-
RMI does not consider these restrictions.
The goal of this paper is to present a framework for
real-time communication using Java, aimed at
applications that run over an embedded platform with
restricted resources. This platform implements RTSJ
and natively executes Java bytecodes. A case study in
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
vehicle automation was chosen, since it requires
distribution to fulfill functional requirements. This
work does not intend to provide a full solution to
automobile industry. Its main interest is to explore the
resources provided by a real-time Java platform,
analyzing latencies introduced by the communication
API, the fulfillment of timing constraints, and memory
usage, which are important requirements in many
embedded applications. This work is part of a larger
research effort that intends to provide a flexible and
reusable network support to higher abstraction layers
The remaining of this paper is organized as follows.
Section 2 gives a brief overview of the hardware and
software aspects of the development platform, which is
based on the FemtoJava processor, a customizable RT-
Java processor. Section 3 presents the proposed
communication architecture. Section 4 presents a case
study in the automotive automation domain. Section 5
shows experimental results from simulations providing
latencies and memory measures. Finally, Section 6
draws the main conclusions of the paper and discusses
future work.
2. Development platform
2.1. JAVA-RT configurable processor
The development platform used in this work is the
FemtoJava processor [5], a stack-based microcontroller
that natively executes Java bytecodes, whose major
characteristics are a reduced and configurable
instruction set, Harvard architecture, and small size. It
implements an execution engine for Java in hardware,
through a stack machine that is compatible with the
specification of the Java Virtual Machine (JVM). A
compiler that follows the JVM specification is used
and allows the synthesis of an ASIP (application-
specific integrated processor) version of FemtoJava.
For real-time applications, a multi-cycle version of
FemtoJava is used. It implements a subset of the JVM
bytecodes, with 68 instructions. The supported
instructions are basic integer arithmetic and bitwise
operations, conditional and unconditional jumps,
load/store instructions, stack operations, and two extra
bytecodes for arbitrary load/store. In this processor, all
instructions are executed in 3, 4, 7, or 14 cycles,
because the microcontroller is cacheless and several
instructions are memory bound. In order to support
multithread applications, the instruction set of
FemtoJava was expanded, with the inclusion of
[6]. Additionally,
two pseudo-bytecodes,
were created to provide context switching [7].
Enhancements in FemtoJava performance are
obtained with pipelined and VLIW versions of the
processor [8] or even using resources implemented in
hardware [9].
2.2. Design and simulation tools
The Sashimi environment [5] is used to generate
customized code for the application. The code includes
the VHDL description of the processor core and ROM
(programs) and RAM (variables) memories and can be
used to simulate and synthesize the application.
Sashimi is an example of JVM optimization for
embedded systems. It provides a powerful and easy-to-
use development environment for embedded systems
that has been successfully applied to different case
The Sashimi environment has been extended to
incorporate an API [10] that supports the object-
oriented specification of concurrent tasks and allows
the specification of timing constraints, implementing
the RTSJ standard. These facilities increase the code
abstraction level and optimize the development of real-
time embedded systems. The intent is to minimize
architecture-dependent characteristics within the
scheduling algorithms, thus making the framework as
general as possible.
The RTSJ-API uses the concept of schedulable
objects, which are instances of classes that implement
interface, for instance the
. It also uses a set of classes to store
parameters that represent a particular resource demand
from one or more schedulable objects. The
class (superclass of
), for example, includes
several useful parameters for the specification of real-
time requirements. Moreover, the API supports the
expression of the following elements: absolute and
relative time values, timers, periodic and aperiodic
tasks, and scheduling policies. The term ‘task’ derives
from the scheduling literature, representing a
schedulable element within the system context. It is
also a synonym for schedulable object.
3. Communication API
In order to provide communication facilities, an
API (APICOM) was developed for the real-time
FemtoJava processor, providing an interface to the
application layer.
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
The communication system was proposed to
provide message exchange among applications running
in different FemtoJava processors. The API allows
applications to establish a communication channel
through the network, which can be used to send and
receive messages. The service allows the assignment
of different priorities to messages and can run in a
multithread environment. From the application point-
of-view, the system is able to open and close
connections, in a client-server mode, or even to run in
a publish-subscribe mode.
Figure 1 shows the overall platform architecture,
which includes the APICOM. The APICOM works
together with the RTSJ-API, using the FemtoJava
features to provide communication via a network
Communication Bus
Communication Bus
Figure 1: General Platform Architecture
3.1. Requirements for the APICOM
There are two main viewpoints that can be
considered when requirements for the APICOM are
1 – Resource availability
Distributed embedded real-time systems have a
strong demand for performance, in order to meet the
timing requirements. In the other hand, they do not
dispose of powerful processing resources, since they
must have limited area and small memory and meet
power consumption restrictions.
Another important factor is the underlying network.
The developer would like to send messages,
abstracting the network frame format and physical
layer details.
2 – Application constraints
Embedded applications are growing in complexity,
becoming multithread and needing real-time support.
Multithreading and real-time are provided by the RT-
FemtoJava (FemtoJava processor with RTSJ-API).
However, communication features are also important,
since threads can be distributed in several processors.
Some distributed applications have a large number
of control packets exchanged during their execution.
Some packets are addressed to a specific host. This
feature supposes that packets contain their destination
address. This way, the API should provide a
connection-oriented communication. There are also
packets addressed to all hosts in the network, and so
the API should support broadcast too.
3.2. The APICOM model
The class diagram of the APICOM is presented in
Figure 2. The class Transport represents the front-end
that provides the interface to final applications. This
class is also responsible for breaking messages into
packets. The class Message represents the information
that comes from the application (or will be received
from a remote connection). TransportConnection is the
class that makes possible the individualization of each
connection between two hosts. It keeps the host logic
addresses and the ports of the connections.
The class Network works like a filter of packets. It
selects packets that are addressed to the local host and
redirects them to the class Transport. The received
packets that are not addressed to the local host are
Figure 2: Class diagram of the APICOM
The class DataLink is responsible for the final
packet transmission. To perform this, it breaks the
message into frames that are specific for each network
type that is used in the physical layer. In order to re-
use the APICOM over different network technologies,
this class was modeled as an abstract class. Therefore,
by extending this class it is possible to implement a
dedicated class for a certain type of network. Classes
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
DataLinkCan and DataLinkFieldBus provide the
specific implementation of CAN (Control Area
Network) and Field Bus network, respectively.
To define Pack objects for different network
standards the same approach of class DataLink is used.
Pack is an abstract class that must be extended to
implement specific aspects of the network being used.
In the example shown in Figure 2, CAN-bus and Field
Bus are provided as extensions.
A general description of the services that are
provided by the communication API is given in Table
Table 1: Services provided by communication API
Service Description
Applications can request and wait for connections.
The API provides a code that identifies the
connection and should be used to send and/or
receive messages.
Applications exchange information by sending and
receiving messages, which are sequences of up to
20 bytes.
Establish a
logic local
Applications can set their own addresses, which
will be used to identify stations.
Messages can be sent directly to a specific host,
through a predefined connection, or broadcasted in
the network. This option is made by calling different
primitives of the API when sending a message. A
host needs to perform a subscription in order to
receive broadcast messages.
3.3. The message exchanging algorithm
To perform message exchanging, a set of basic
operations was defined.
Sending a message requires the following steps:
- The message is fragmented into packets (Pack
objects), according to the packet length defined in
DataLink classes.
- For each packet to be sent, bit manipulation is
performed, setting network frame (data inserted in the
physical layer) attributes, like priority, packet
sequence, and addresses.
An important aspect of this operation is that the
application can set a time-out object. It defines the
maximum time that the application can wait sending a
message. If the API fails, it returns an error code and
the application can process the exception.
To receive a message there are two flows. The first
one is started when a frame arrives at the network
interface. The following steps are performed:
- A Pack object is filled.
- Fragmentation is detected and the message is
- A flag indicating a ready Message is activated.
The second flow is started by the application and
follows these steps:
- Set a waiting message flag, indicating the
intention to receive a message in a specified
- Test if there is a ready message in that connection.
- Receive the message in a specified message
4. Case study
4.1. Steer-by-wire system
Nowadays there is a trend in the automotive
industry to design vehicles with an embedded
electronic steer-by-wire system. This reduces the
weight of the vehicle, by the substitution of the
hydraulic column, which connects the steering wheel
to the road wheels, by an electronic system. In short,
the main idea is to replace hydraulic mechanic devices
by sensors and actuators linked by wires [11].
Figure 3 shows a simple model of a steer-by-wire
system based on the one presented in [12]. In this
example, only the variables involved in the axis
control are shown. Angular sensors capture the
steering wheel position, and an actuator is responsible
for the road wheels’ motion. In order to impose the
right position to the road, the local controller (ECU –
Electronic Control Unit) is responsible to process the
information that comes from the main controller and
the current position at a certain moment, using a PID
(Proportional Integral Derivative) algorithm to drive
road to the desired position. It means that the road
wheel position is captured and used by the controller to
calculate the value to be sent to set the road wheel
driver, providing feedback control to the road wheel
Steering Wheel
Main Controller
Dash Board
Figure 3: A simple model of a steer-by-wire system
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
4.2. Implementation of the control system
The control system model is shown in Figure 4. It
demonstrates the implementation of the structure
presented in Figure 3 and is related only to front road
Having in mind that there are spatial distribution
requirements imposed by the processing elements in an
application like that, the use of a multi-processor
system was considered. Moreover, since time-to-
market is a more important cost than the processors, it
is worthwhile to invest in a multiprocessor solution
that provides reduction in development time.
Each element (road wheels, steering wheel and dash
board) has it own controller that makes possible the
local data processing. A processor to control a road
wheel is directly linked to its activation point, being
the “set point” defined remotely by the Main
A CAN-bus network interconnects controllers.
With the abstractions provided by the APICOM,
however, it would be very easy to replace the network
type, while keeping most of the application code.
CAN - bus
Driver &
Processor 1
(PID controller)
Driver &
Processor 2
(PID controller)
Processor 3
(Main Controller)
Processor 4
(Dash Board)
Driver &
Processor 0
Figure 4: Architecture of the distributed control system
Road Wheel Controllers are responsible for the data
acquisition from angular sensors. The processor
calculates the right force that the actuator has to apply
over the road wheel, in order to establish a new
position. Besides that, it has to send position and
diagnostic data to the main controller.
The Steering Wheel Controller is in charge of
acquiring the steering wheel position and calculating
the feedback that has to be applied, based on car speed
information supplied by the Main Controller.
The Dash Board Controller processes the data that
comes from the Main Controller and displays this
information in a user-friendly way.
The Main Controller processes the data provided by
the Steering Wheel Controller and coordinates road
processors. In a complete steer-by-wire system, this
unit is supposed to perform other managing and
control functions.
Processors 1 and 2 (road wheel processors) run a
periodic RealtimeThread, called Controller, whose
period, imposed by system properties, is 8 ms, and
whose WCET (Worst Case Execution Time) is 0.9 ms.
The RealtimeThread, running in the Steering Wheel
Controller, has the same period (8 ms) and is in charge
of sending the wheel position to the Main Controller
and applying a feedback force in the steering wheel.
There are two main RealtimeThreads running in the
Main Controller. The first one has a period of 8 ms and
receives the steering wheel position and defines set
point value to the Road Wheel Controllers. The second
one sends the car speed to the Steering Wheel
Controller with a period of 25 ms.
Besides the main threads, each processor runs a
diagnostic thread that has a period of 100 ms and needs
two occurrences to complete one turn.
Timing constraints defined for each thread impose
maximum latencies of communication and scheduling
services. These restrictions will be used to evaluate the
APICOM in the next section.
5. Experimental results
The experiments were simulated with a cycle-
accurate performance and power simulator, called
CACO-PS [13]. The clock rate of the processors was
20 MHz. The CAN 2.0A [14] specification was
adopted, with eight bytes of data and operating at 1
Mbps. Two sets of experiments were evaluated. The
first one is a stressing test used to measure APICOM
properties. The second one uses the case study
presented in Section 4.
5.1. APICOM evaluation
In order to evaluate the APICOM, a benchmark
suite was applied. A producer-consumer application
was developed to send 20 messages whose lengths
vary from 1 up to 20 bytes. Time spent sending and
receiving messages is shown in Figure 5, with the x-
axis indicating the length of messages.
One remarkable aspect in Figure 5 is the difference
between latencies to send and to receive a message. To
send a message, a reference to a Message object is
passed to the API. However, to receive a message, the
content of the receiver object is copied into a Message
object, which belongs to the application. In a standard
JVM, the receiver object would be dynamically
created. In this work, since the platform supports only
static objects, the object already exists, thus remaining
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
only the cost associated to the copy, which is lower
than the cost to create it.
Other important aspect in Figure 5 is the step seen
when the length of the message increases from 7 to 8
bytes or from 14 to 15 bytes. This happens because the
API needs to use one packet more to send the message.
In this example, the packet can carry on 7 bytes. This
cost is related to the fragmentation/re-assembly
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Latency (ms)
Figure 5: APICOM benchmark latencies
The memory footprint was also evaluated to detect
the impact of communication API. Table 2 shows the
amount of memory, in bytes, used by the consumer-
producer benchmark. To offer a reference the
benchmark was implemented using a local object to
pass messages from consumer to producer. Thus, the
line “with API” in Table 2 indicates the total memory
used by the benchmark using APICOM to exchange
messages, while the line “w/o API” shows the total
memory without the API. ROM memory is the code
(object methods), while RAM is used for variables
(object attributes). It is important to note that
FemtoJava tools, eliminating all un-referenced
methods and attributes, automatically customize the
final code.
Table 2: Memory usage
Transmission Reception
w/o API 1200 Bytes 492 Bytes 1110 Bytes 632 Bytes
with API 4561 Bytes 2276 Bytes 3724 Bytes 2104 Bytes
5.2. Case study results
The experiment with the steer-by-wire system uses
a set of real-time threads running over a fixed priority
scheduler provided by our RTSJ API [10]. Five
FemtoJava processors, connected by a network, as
shown is Figure 4, were simulated. The processors’
clock rate of 20 MHz, used in the benchmark, could be
kept because it fulfills application time requirements.
Costs of main parts of the communication protocol
are shown in Table 3. “Diag send” is the time used by
processors 0, 1, and 2 to send a diagnostic message (2
bytes) to the Main Controller, while “Data recv” is the
time they use to receive the set point value (4 bytes).
“Data send” is the time used by processor 3 to send the
set point value (4 bytes), and processor 0 to send the
steering wheel position, while “Diag recv” is the time
used by the Main Controller to receive one diagnostic
message. Finally, “CAN-bus” indicates the physical
layer delay for one packet. CAN-bus latency is for a
packet of 11+64 bits (bus contentions were not
considered). All values are expressed in milliseconds.
Table 3: Communication latencies
Diag recv Data send Diag send Data recv CAN-bus
0.168 ms 0.179 ms 0.159 ms 0.231 ms 0.075 ms
Figure 6, Figure 7, and Figure 8 show the time line
for a control cycle of a road wheel processor, steering
wheel processor, and main processor, respectively. The
“Scheduler” cost indicates the time consumed by a
fixed priority scheduler. Its WCET (Worst Case
Execution Time) is 0,7 ms. The part of the time lines
involving diagnostic and car-speed message do not
occur in a period of 8 ms, as explained in Section 4. It
is shown in the figures to illustrate a worst-case cycle.
After all activities within a period have been executed,
the processor stays waiting for the next activation time
(message arrived or control thread).
Data Recv (from Main)
0 ms 8 ms
Control algorithm
. . .
. .
Diag Send
. .
Figure 6: Time line of a cycle for processors 1 and 2
0 ms
8 ms
Control algorithm
. . .
. .
Data Send
Diag Send
Data Recv (car-speed)
. .
Figure 7: Time line of a cycle for processor 0
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
0 ms
8 ms
. . .
Diag Recv
. .
. .
Data Send (Roads)
car-speed measurement
Data Send (car-speed)
Data Recv (Steering)
. .. .
Figure 8: Time line of a cycle for processor 3
To evaluate occupation level of the processors the
working time for a cycle was calculated. Let’s take a
Road Wheel processor as an example. The cycle that
has the higher occupation level is the one that have a
control thread and a diagnostic thread, as shown in
Figure 6. This total cost, t
, is indicated in the
equation (1), shown below. t
is the time used to
process an arrived message (Previously called “Data
Recv”) that brings the desired position, sent by the
Main Controller. t
express the WCET of the thread
that performs the PID control algorithm, while t
the WCET related to the thread that collect and
manage diagnostic data. The time used to send a
diagnostic message is indicated by t
called “Diag Send”). Finally, the time to process
scheduling algorithm, given by t
, should be added
three times, since the scheduler is activated before and
after every thread that runs. For the scheduler, the
WCET was used even though it is known that its
execution time depends on the thread that ran before it
and the one that should be scheduled to run after.
tttttt ⋅++++= 3
Since the control period imposed by the road wheel
system to the PID algorithm is 8ms, there is even a
space to reduce the clock frequency of the processors 1
and 2.
A similar approach was used to verify the feasibility
of control cycle of the other processors. Equations 2
and 3 show the results for Steering Control and Main
control respectively.
As one can see, the communication costs cannot be
neglected. The algorithms described in Section 3.3
justify the origin of these costs. In order to use a high-
level programming language and the object-oriented
paradigm, developers pay in memory usage and
performance. However, they look for a short time-to-
market, provided by reusability and flexibility. These
are the advantages provided by APICOM in this
6. Conclusions and future work
This paper described a mechanism to design
multithread object-oriented distributed real-time and
embedded applications. The developer can abstract
network-dependent aspect and use facilities provided
by an RSTJ API. Latencies and memory overhead for
communication resources were evaluated.
A platform that implements RTSJ, natively
executing Java bytecodes, was used in a case study of
a distributed servomechanism control for automotive
automation, and latencies and costs were measured.
The main contribution of the paper is to provide a
framework to handle communication channels between
real-time applications, extending RTSJ to cover
embedded distributed applications. Moreover, the
target platform is a very restrictive one with low
performance and small memory. To ensure time
correctness, a time-out mechanism was implemented.
The communication control (presentation and
session layers) is currently being implemented inside
the code of the final application. As a future work, this
control will be moved to a middleware, and the cost of
this alternative will be evaluated.
In order to broaden design space exploration
alternatives, a hardware implementation of the
APICOM is being developed. The flexibility in
choosing a hardware or software implementation is
going to be addressed by the use of the object-oriented
approach, encapsulating hardware features in a
surrogate class.
Thanks are given to the Brazilian funding agency
CNPq, which is the project sponsor. The authors also
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006
thank the other SEEP-project researchers for the
valuable discussions.
[1] G. Bollella, J. Gosling, and B. Brosgol, “The Real-
Time Specification for Java”,
V1.0.pdf, 2001.
[2] The JSR-50 Home Page –
[3] A. Wellings, R. Clark, D. Jensen, and D. Wells. “A
Framework for Integrating the Real-Time Specification
for Java and Java's Remote Method Invocation”, In:
Proceedings of the 5th IEEE International Symposium
on Object-Oriented Real-Time Distributed Computing,
Crystal City, USA, 2002, pp. 13-22.
[4] A. Borg and A. Wellings, “A Real-Time RMI
Framework for the RTSJ”, In: Proceedings of the 15th
Euromicro Conference on Real-time Systems, Porto,
Portugal, July, 2003.
[5] S.A. Ito, L. Carro, and R.P. Jacobi. “Making Java
Work for Microcontroller Applications”, IEEE Design
& Test of Computers, Sept/Oct. 2001, vol. 18, n. 5, pp.
[6] M.A. Wehrmeister, C.E. Pereira, and L.B. Becker,
“Optimizing the Generation of Object-Oriented Real-
Time Embedded Applications Based on the Real-Time
Specification for Java”, To appear in Design,
Automation and Test in Europe, Munich, Germany,
March, 2006.
[7] L.S. Rosa Jr. et al. “Scheduling Policy Costs on a Java
Microcontroller”, In: Proceedings of the workshop on
Java Technologies for Real-Time and Embedded
Systems (JTRES), Catania, Italy, 2003, pp. 520-533.
[8] A.C.S. Beck Filho, and L. Carro. “Low Power Java
Processor for Embedded Applications”, In: IFIP VLSI-
SoC’2003, Darmstadt, 2003, pp. 239-244.
[9] E.T. Silva Jr., L. Carro, F.R. Wagner, and C.E. Pereira.
“Development of Multithread Real-Time Applications
Using a Hardware Scheduler”, In: IFIP VLSI-
SoC'2005, Perth, Australia, October, 2005.
[10] M.A. Wehmeister, L.B. Becker, and C.E. Pereira,
“Optimizing Real-Time Embedded Systems
Development Using a RTSJ-based API”, Workshop on
Java Technologies for Real-Time and Embedded
Systems - JTRES 2004, In: Proceedings Springer
LNCS, Agia Napa, Cyprus, October, 2004, pp. 292-
[11] S.H. Jang, T.J. Park, and C.S. Han, “A control of
vehicle using Steer-by-Wire system with Hardware-In-
the-Loop-Simulation system”, In: Proceedings of the
2003 IEEE/ASME International Conference on
Advanced Mechatronics, 2003.
[12] S. Amberkar, F. Bolourchi, J. Demerly and S. Millsap,
“A Control System Methodology for Steer-by-Wire
Systems”, SAE World Congress, Detroit, 2004.
[13] A.C.S. Beck Filho, J. Mattos, F.R. Wagner, and L.
Carro. “CACO-PS: A General-Purpose Cycle-Accurate
Configurable Power Simulator”, In: Proceedings of the
16th Symposium on Integrated Circuits and Systems
Design, IEEE Computer Society Press, São Paulo,
Brazil, September 2003.
[14] Bosch. CAN Specification. Ver. 2.0. Robert Bosch
GmbH Stuttgart, 1991.
Proceedings of the Ninth IEEE International Symposium on Object and Component-Oriented Real-Time Distributed Computing
0-7695-2561-X/06 $20.00 © 2006