
operation. To each subtransaction is associated a
normal mode and a survival mode.
The participant site is composed of various
modules, implementing the Overload Manager, the
EDF Scheduler and the Data Manager. All these
modules are defined like threads and work
concurrently.
The application developed to evaluate the
performance of our platform under various working
conditions is composed of three modules:
- the Execution Module : that gets the
transactions to be executed from a file and sends
them simultaneously, in the form of HTTP requests,
to the master site and waits for the execution results.
- the Configuration Module that enables the user
to change the execution parameters of the
distributed
system. These parameters allow the user to
determine the conditions under which the
transactions would be executed. For example, the
user can determine whether the importance value of
transactions should be considered during the
stabilization process or not, etc.
- the Statistics Module has the purpose of
analysing the series of execution grouped according
to the type of configuration under which they have
been executed and to display graphically the results
of the analysis.
7.2 Simulation analysis
During the simulation phase, several series of
execution have been used to measure the
performance of the system under different working
conditions. The series vary in terms of the number of
transactions (50, 100, 150, 200, 250, 300) in order
to allow one to evaluate the behavior of each
simulation configuration vis-à-vis a linear increase
of the number of requests. The series have also been
designed in a way to ensure several conflicting data
access between read and write operations and few
conflicts between write operations. All the tests have
been carried out on a platform consisting of
participant sites, each having a local database of 30
records. Also, each distributed transaction contains
three subtransactions, one for each participant site,
and each sub-transaction has at most 3 data
operations. The tests presented in Figure 3 have been
carried out to evaluate the system under various
configurations, comprising the multimode concept
and that of the importance value.
The results of Figure 3 show that the system
achieves the best performance when the multimode
concept is applied and when the importance value of
transactions is considered during the stabilization
process.
8 CONCLUSION
In this paper, we have focused on the design of a
model which defines a transactional behaviour
adapted to the context of real time. The behavioural
and the structural specifications of this model
involving several execution modes for real-time
transactions is an efficient solution to manage
overload situations. A simulation platform based on
a commit processing protocol that manages
transient-overload situations of the distributed
system has been developed. When an overload is
detected within a participant site, the transactions
that are important for the application are favoured.
The less important ones are switched into degraded
modes or discarded if the degraded modes are not
sufficient to resorb overload. The simulation
platform is based on Java technology and makes use
of MySQL databases. Transactions are sent via
HTTP requests to the master which is implemented
as a Java Servlet on TomCat server. Each participant
site implements an overload manager, an EDF
scheduler and a data manager. This platform
integrates a graphical interface to submit
transactions, a configuration module to fix a certain
number of parameters and a statistical module that
displays the simulation tests in a graphical way. The
simulation results show good performances under
overload and multimode execution.
REFERENCES
(Atlas et al. 1998) A. Atlas and A. Bestavros, "Statistical
Rate Monotonic Scheduling", in proc. of IEEE Real-
Time Systems Symposium, Madrid, dec. 1998.
(Bernstein et al. 1987) P. Bernstein, V. Hadzilacos and N.
Goodman, "Concurrency Control and Recovery in
Database Systems”, Addison Wesley, 1987.
(Baker 1991) T. P. Baker, "Stack-based Scheduling of
Real-Time Processes", in Real-Time Systems Journal,
3(1), pp. 67-99, march 1991.
(Bernstein et al., 1987) P. Bernstein, V. Hadzilacos and N.
Goodman, "Concurrency Control and Recovery in
Database Systems”, Addison Wesley, 1987.
(Bestavros et al., 1996) A. Bestavros and S. Nagy,"Value-
cognizant Admission Control for RTDB systems",
Proc. of the 17th Real-Time Systems Symp., pp. 230-
239, IEEE Computer Society, dec. 1996.
(Buttazo et al., 1998) G. C. Buttazo, G. Lipari and L.
Abeni, "Elastic Task Model for Adaptive Rate
Control", in Proc. of IEEE Real-Time Systems
Symposium, Madrid, dec. 1998.
ICEIS 2004 - DATABASES AND INFORMATION SYSTEMS INTEGRATION
334