Page 145 - IJOCTA-15-2

E. Karabulut T¨urkseven, E. Gen¸c, and I. Safak / Vol.15, No.2, pp.330-342 (2025)
number of tasks allowed to run in parallel {10, to reduce the duration of the EOD process by op-
12, 15, 18, 20} are shown in the columns, and timizing the resources allocated to each task and
the number of threads {150, 180, 200, 250} are their order of execution. We model this prob-
shown in the rows. Tables 7 and 9 reveal that lem as a multi-mode resource-constrained project
Stages 1 and 4 are not utilizing the full capacity scheduling problem, and present a flow-based for-
of 200 threads and 15 parallel tasks. The same mulation of MRCPSP. We also provide a SA
makespans can be achieved with values as low as heuristic for implementation purposes and verify
150 threads and 10 parallel tasks. Stage 2 how- that SA reaches the optimal solution for this par-
ever, as shown in Table 8, significantly benefits ticular problem setting. The motivating purpose
from increasing the thread count up to 200, with of the proposed scheduler is to determine the exe-
a slower rate of improvement from 200 to 250 cution order of EOD tasks, along with the number
threads; and as the other stages it is not impacted of threads to be used during execution to mini-
by the change in the number of parallel tasks in mize the makespan of the EOD process.
the range of 10-20. It is also shown that the scheduler can be
used for analytical purposes. Considering various
Table 7. Estimated Stage 1 durations in minutes
methods such as code optimization, paralleliza-
with varying resource availability
tion, data partitioning and load balancing that
can be used to optimize the EOD process, the
10 12 15 18 20
150 27.0 27.0 27.0 27.0 27.0 scheduler in this paper can be used to detect
180 27.0 27.0 27.0 27.0 27.0 the importance of scheduling and crashing tasks,
200 27.0 27.0 27.0 27.0 27.0 along with the sufficiency of the infrastructure for
250 27.0 27.0 27.0 27.0 27.0
optimizing the EOD process. Our analyses have
Table 8. Estimated Stage 2 durations in minutes effectively identified three distinct structures and
with varying resource availability
requirements for the three most critical stages
10 12 15 18 20 of the EOD process. The existing infrastructure
150 167.0 167.0 167.0 167.0 167.0 is shown to be sufficient for two of the stages,
180 150.7 150.7 150.7 150.7 150.5 whereas the third stage benefits from increas-
200 112.2 112.2 112.2 112.2 112.2 ing the available threads. For one of the stages,
250 110.8 110.8 110.8 110.8 110.8
merely solving for the optimal schedule and not
Table 9. Estimated Stage 4 durations in minutes
changing the thread allocations can give shorter
with varying resource availability
makespans. All three stages benefit from using
10 12 15 18 20 multiple modes. In other words, the flexibility
150 75.3 75.3 75.3 75.3 75.3 of changing the thread allocations and changing
180 75.3 75.3 75.3 75.3 75.3 task durations lead to shorter makespans in all
200 75.3 75.3 75.3 75.3 75.3
250 75.3 75.3 75.3 75.3 75.3 three stages.
The work in this research can be further de-
veloped to deal with Fixed Tasks in smarter ways;
6. Conclusion i.e. dynamic resource availabilities. Further-
more, the flow-based MRCPSP can be tested for
Banks use EOD processes to generate and deliver
larger instances, and for problem sizes that ex-
reports to stakeholders for regulatory monitoring,
ceed the capacity of commercial solvers, smarter
business users, and downstream stream systems.
algorithms can be devised.
EOD processes enable the company to meet its
objectives, such as minimizing the time required
to complete all tasks, maximizing customer sat- Acknowledgments
isfaction, maximizing profit, or minimizing costs. The authors would like to thank Aykut Y¨uksek
The EOD process may last longer and more chal- and Yal¸cın Yılmaz from Fibabanka for their con-
lenging to coordinate in corporate systems when tributions.
applications are extensive, distributed, and uti-
lize diverse technologies, as seen in microservices Funding
architectures. Consequently, conventional batch
This research has been funded by Fibabanka, with
processing techniques may prove inadequate for
grant number E.D.VR-21-02416.
managing EOD batch processes in these systems,
necessitating the development of novel EOD al- Conflict of interest
gorithms.
This paper presents a case study of an EOD The authors declare that they have no conflict of
process in a Turkish bank, where the objective is interest regarding the publication of this article.
340