Algoritma Deterministik Paralel Untuk Proses Discovery Real-Time Pada Event-Log Terdistribusi

Hermawan, Hermawan (2024) Algoritma Deterministik Paralel Untuk Proses Discovery Real-Time Pada Event-Log Terdistribusi. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

	Text Buku Disertasi Final Upload.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (6MB) | Request a copy
	Text 05111660010004-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (6MB) | Request a copy

Abstract

Proses discovery membutuhkan peningkatan kualitas dan kapasitasnya untuk dapat diakses secara online dan real-time terutama pada completeness dan correctness. Untuk peningkatan correctness dan completeness, disertasi ini memberikan kontribusi penggunaan pola control-flow pada algoritma AlphaT dan AlphaT++ untuk discovery event-log yang memiliki kandungan uncertainty dan short-loop kompleks yang menjadi kendala utama deterministik-miner. dimana Alpha dispesifikasikan untuk uncertainty sedangkan peningkatannya pada AlphaT++ untuk short-loop. Dari hasil pengujian kedua algoritma telah mampu meningkatkan completeness>97% dan correctness>95%, sehingga workflow hasil discovery terhindar dari dead-task dan dead-lock. Adapun untuk optimalisasi kapasitas proses discovery pada sistem paralel terdistribusi dilakukan melalui paralelisasi discovery yang dieksekusi secara online dan real time. Pada tahapan pertama optimalisasi untuk online dan realtime, dengan berbasis komputasi paralel multi-thread CPU digunakan strategi MIMD (Multiple Instruction Multiple Data) yang mampu mengeksekusi multi kernel keseluruhan tahapan discovery didalam thread lokal secara sekuensial independen dari banyak sumber event-log secara paralel. Adapun optimalisasi yang kedua digunakan kombinasi komputasi paralel CPU-GPU, yang terdiri dari eksekusi MIMD-CPU untuk pembacaan streaming menggunakan lokal thread, eksekusi MISD-CPU (Multiple Instruction Single Data) untuk membentuk Big-Footprint, dan berikutnya mengeksekusi discovery pada reduksi pararel dan penentuan jalur maksimum secara SIMD didalam GPU. Dari hasil pengujian, kombinasi ketiga strategi tersebut telah terbukti mampu meningkatkan performansi kecepatan eksekusi antara 10 hingga 30 kali tergantung jumlah task aktifitas pada proses discovery paralel didalam sistem terdistribusi, adapun hasil optimal didapatkan pada jumlah thread>100 dan jumlah aktifitas>45. Dari hasil penerapan metode pada disertasi ini dapat dipakai sebagai perkakas Business Process Monitoring (BPM) untuk audit dan peningkatan proses bisnis.
==================================================================================================================================
The discovery process requires enhancements in quality and capacity to be accessible online and in real time with maintaining high standards of completeness and correctness. To address this, the dissertation introduces control-flow patterns in the AlphaT and AlphaT++ algorithms for discovering event logs containing uncertainty and complex short-loops, which are significant challenges for deterministic miners. Alpha is designed to handle uncertainty, while its enhancement, AlphaT++, targets in short loops. Testing results show that algorithms achieved over 97% completeness and over 95% correctness, ensuring a workflow of discovery results free from dead tasks and deadlocks. To optimize the capacity of the discovery process in distributed parallel systems, the dissertation implements parallelization of discovery online and in realtime. The first optimization stage for online and real-time processes employs the MIMD (Multiple Instruction Multiple Data) strategy based on multi-thread CPU parallel computing. This strategy executes multiple kernels across all discovery stages within local threads, sequentially and independently, from multiple event log sources in parallel. The second optimization leverages a combination of CPU-GPU parallel computing. This includes MIMD-CPU execution for streaming reading using local threads, MISD-CPU (Multiple Instruction Single Data) execution to form a Big-Footprint, and subsequent discovery execution on parallel reduction and maximum path determination using SIMD within the GPU. Testing demonstrated that these strategies significantly increased execution speed-up by 10 to 30 times, depending on the number of active tasks in the parallel discovery process within the distributed system. Optimal results were obtained with more than 100 threads and over 45 activities. The methods applied in this dissertation can be used as tools for Business Process Monitoring for auditing and enhancement business processes.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Proses Discovery, Sistem Terdistribusi, Komputasi Paralel, MIMD (Multi Instruction, Multi Data), SIMD (Single Instruction, Multi Data), Big-Eventlog, Computer Processor Unit (CPU), Graphics Processing Unit (GPU), Business Process Monitoring (BPM).
Subjects:	T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK5105.546 Computer algorithms
Divisions:	Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Informatics Engineering > 55001-(S3) PhD Thesis (Comp Science)
Depositing User:	Hermawan Hermawan
Date Deposited:	17 Oct 2024 04:01
Last Modified:	17 Oct 2024 04:01
URI:	http://repository.its.ac.id/id/eprint/115751

Actions (login required)

View Item