Embedded Software for SoC - Grupo de Mecatrônica EESC/USP

462 Chapter 33
Table 33-7. Results on adaptive checkpointing for five jobs: variation of P with for (a) C =
10, and k = 5. (b) C = 20, and k = 5.
Fault arrival rate
Probability of timely completion of tasks, P
Poisson-arrival
k-fault-tolerant
ADT_MUL
(a)
6
10
14
18
0.996
0.967
0.870
0.647
1.000
0.983
0.899
0.610
1.000
0.987
0.914
0.696
(b)
6
10
14
18
0.884
0.487
0.149
0.018
0.815
0.370
0.086
0.013
0.930
0.542
0.153
0.028
approach provides fault tolerance and facilitates dynamic power management.
The proposed energy-aware adaptive checkpointing scheme uses a dynamic
voltage scaling criterion that is based not only on the slack in task execution
but also on the occurrences of faults during task execution. We have presented
simulation results to show that the proposed approach significantly reduces
power consumption and increases the probability of tasks completing correctly
on time despite the occurrences of faults. We have also extended the adaptive
checkpointing scheme to a set of multiple tasks. A linear-programming model
is employed in an off-line manner to obtain the relevant parameters that are
used by the adaptive checkpointing procedure. Simulation results show that
the adaptive scheme is also capable of providing a high probability of timely
completion in the presence of faults for a set of multiple tasks.
We are currently investigating checkpointing for distributed systems with
multiple processing elements, where data dependencies and inter-node communication
have a significant impact on the checkpointing strategy. We are
examining ways to relax the restrictions of zero rollback and state restoration
costs, as well as the assumption of no fault occurrence during checkpointing
and rollback recovery. It is quite straightforward to model nonzero
rollback and state restoration cost, but it appears that the assumption of no
faults during checkpointing is difficult to relax.
REFERENCES
1.
2.
3.
P. Pop, P. Eles and Z. Peng. “Schedulability Analysis for Systems with Data and Control
Dependencies.” Proceedings of Euromicro RTS, pp. 201–208, June 2000.
T. Ishihara and H. Yasuura. “Voltage Scheduling Problem for Dynamically Variable Voltage
Processors.” Proceedings of International Symposium on Low Power Electronics and
Design, August 1998.
Y. Shin, K. Choi, and T. Sakurai. “Power Optimization of Real-Time Embedded Systems