Peningkatan Skalabilitas Dan Availabilitas Sifars Dengan Microservice Dan Horizontal Scaling

Timur, Tahta Dari (2019) Peningkatan Skalabilitas Dan Availabilitas Sifars Dengan Microservice Dan Horizontal Scaling. Masters thesis, Institut Teknologi Sepuluh Nopember.

Text 07111750050004-Master_Thesis.pdf Restricted to Repository staff only Download (4MB)

Abstract

Deep learning menjadi salah satu metode pilihan dalam menyelesaikan masalah-masalah di bidang kecerdasan buatan. Beberapa masalah yang biasa dikaji adalah natural language processing, pengenalan suara, dan pengenalan wajah. Pengembangan sistem pengenalan wajah berbasis deep learning dapat dilakukan dengan adanya dataset dan arsitektur CNN yang tepat. Perangkat GPU umumnya lebih dipilih dibandingkan CPU karena dapat mempercepat proses training. Akan tetapi pada tahap deployment, sistem berbasis deep learning ini pada umumnya membutuhkan perangkat komputasi yang lebih besar, sehingga sistem monolithic tidak mampu menjalankannya pada skala yang lebih besar. Container orchestration system (COS) memberikan tawaran solusi peluncuran cluster multihost (scalable) yang dapat mengabstraksikan perangkat keras sehingga setiap host bisa saling berbagi pemakaian sumber daya komputasi. Dalam hal ini, sistem berbasis deep learning membutuhkan perangkat GPU agar bisa berjalan secara optimal. Dari hasil eksperimen, peluncuran sistem monolithic dan multihost tidak memiliki selisih performa yang signifikan. Hal ini menunjukkan bahwa solusi peluncuran sistem pengenalan wajah berbasis deep learning di atas COS dapat dilakukan tanpa kelebihan biaya komputasi yang signifikan.
==================================================================================================================================
Deep learning is one of the most widely used approaches for solving problems in the field of artificial intelligence. Common applications include natural language processing, speech recognition, and face recognition. The development of deep learning-based face recognition systems relies on the availability of large datasets and suitable convolutional neural network (CNN) architectures. Graphics Processing Units (GPUs) are generally preferred over Central Processing Units (CPUs) because they significantly accelerate the training process. However, during deployment, deep learning-based systems often require substantial computing resources, making monolithic architectures difficult to scale efficiently. Container Orchestration Systems (COS) provide a solution by enabling scalable multi-host clusters that abstract the underlying hardware and allow multiple hosts to share computing resources efficiently. Since deep learning applications perform optimally with GPU acceleration, integrating GPU resources into COS is essential. Experimental results show that the performance of face recognition systems deployed on monolithic and multi-host architectures does not differ significantly. These findings demonstrate that deploying deep learning-based face recognition systems on top of a Container Orchestration System can be achieved without introducing significant computational overhead.

Item Type:	Thesis (Masters)
Uncontrolled Keywords:	deep learning, pengenalan wajah, aplikasi terdistribusi, container orchestration system
Subjects:	Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science) Q Science > QA Mathematics > QA76.9.F38 Fault-tolerant computing
Divisions:	Faculty of Electrical Technology > Electrical Engineering > 20101-(S2) Master Thesis
Depositing User:	Tahta Dari Timur
Date Deposited:	05 Aug 2026 08:34
Last Modified:	05 Aug 2026 08:34
URI:	http://repository.its.ac.id/id/eprint/66426

Actions (login required)

View Item