| Title |
Nulinės prastovos vartotojo saugyklos duomenų migracijos tyrimas |
| Translation of Title |
Zero-downtime user storage data migration study. |
| Authors |
Perkauskas, Linas |
| Full Text |
|
| Pages |
82 |
| Keywords [eng] |
artificial intelligence ; storage migration ; file access prediction |
| Abstract [eng] |
This paper presents a newly developed user storage / disk migration system that allows a user to simultaneously boot migrate an operating system from an old computer to a new empty one via network with minimal downtime. This is achieved by integrating the migration system into a file system mounted as rootfs using specially prepared initramfs. In the newly proposed solution newly accessed files are migrated on demand, while remaining files are transferred in the background to ensure full storage migration. The system viability and performance was validated using an Arch-Linux-based OS, emulating the new computer in the qemu emulator. The paper also proposes 3 graph-based AI models for file access prediction. 2 of which outperform the First-Successor, Last-Successor and Stable-Successor (Noah) traditional file access prediction models in terms of prediction accuracy and cache hit rate using the CMU-DFStrace dataset. The best proposed AI model was tested to predict file access order during storage migration. It was found that due to the overwhelming file system load at the beginning of OS boot, file access prediction in order to pre-migrate the next file that will be accessed in the near future is unusable. Additionally, it was found that using minimal pre-migration of metadata and postponed background file migration without prediction that starts after the downtime interval achieves the lowest downtime. The additional downtime during such migration reached only ~70 % of the standard system boot time. |
| Dissertation Institution |
Kauno technologijos universitetas. |
| Type |
Master thesis |
| Language |
Lithuanian |
| Publication date |
2025 |
