Just-in-time video transcoding system on heterogeneous high performance computing architectures

doctoral thesis

• Title (croatian): Sustav za pravovremeno videotraskodiranje na raznorodnim arhitekturama za računarstvo visokih performaci
• Author: Igor Piljić
• Mentor: Mario Kovač (mentor)
• Committee member: Mario Kovač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; (Department of Control and Computer Engineering); Zagreb
• Defense date and country: 2019, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing; Architecture of Computer Systems
• Universal decimal classification (UDC): 621.3 - Electrical engineering

Abstract

Latest analysis show that 82% of global IP traffic will be video traffic by 2022. Handling this amount of data is a very challenging task for video content providers. Another factor that highlights this problem is the continuously growing number of different devices that can play video content. With such diversity of devices a single copy of the video cannot match requirements of all playback conditions. Just-in-Time (JiT) video transcoding has one of the key roles in resolving these issues. However, it is an extremely compute-intensive and resource-hungry process. This thesis presents a novel algorithm for reusing coding information from the input video stream. The main concept behind the proposed algorithm is to estimate the computational complexity of re-encoding each coding block based on the information retrieved from the decoded frame. The final goal is to achieve an optimal trade-off between video quality of the transcoded bitstream and coding efficiency while conforming to strict timing requirements. To achieve a more efficient solution, a hardware accelerator for inter prediction, as one of the key compute-intensive kernels, is designed and implemented. An integrated system composed of implemented algorithm and custom hardware accelerator is evaluated. Compared with JiT transcoder, the proposed solution increases video quality by 0.945 dB and reduces bitrate on average by 35.06%. Significant speedups of up to 4 times are achieved compared with transcoder without timing requirements but with average losses of 0.592 in PSNR and 21.74% in bitrate.

Abstract (croatian)

Statistike pokazuju da će 82% globalnog Internet Protokol (IP) prometa do 2022. godine činiti video promet. Rukovanje tolikom količinom podataka predstavlja iznimno zahtjevan zadatak za poslužitelje video sadržaja. Još jedan faktor koji naglašava složenost ovog problema je činjenica da broj različitih uređaja koji mogu prikazivati video sadržaj konstantno raste. Takvu raznovrsnost uređaja nemoguće je zadovoljiti jednom verzijom videa. Pravovremeno videotranskodiranje ima ključnu ulogu u rješavanju ovog problema, ali je računalno iznimno zahtjevno. Glavni cilj ove doktorske disertacije bilo je istražiti tehnike iskorištavanja informacija o kodiranju ulaznog video toka kodiranog HEVC standardom da bi se ubrzao proces ponovnog kodiranja, ali bez značajnog negativnog utjecaja na kvalitetu videa i/ili učinkovitost kodiranja. U cilju daljnjeg poboljšanja procesa videotranskodiranja, istražene su i učinkovite izvedbe pravovremenog videotranskodiranja na raznorodnim arhitekturama za računarstvo visokih performanci i razvijen je sklopovski ubrzivač za inter predikciju, kao jednu od ključnih jezgri u procesu videotranskodiranja. Integrirano rješenje, koje se sastoji od razvijenog algoritma i specijaliziranog sklopovskog ubrzivača, postiže bolju kvalitetu videa od 0.945 dB te bolju učinkovitost kodiranja od 35.06% u odnosu na druge pravovremene transkodere. U usporedbi s drugim transkoderima koji ne ostvaruju pravovremeno transkodiranje, postiže se ubrzanje do 4 puta, uz prosječne gubitke od 0.592 dB u kvaliteti te 21.74% u učinkovitosti kodiranja.

• Keywords: video transcoding; HEVC; reusing data; hardware accelerators; heterogeneous high-performance architectures
• Keywords (english): videotranskodiranje; HEVC; iskorištavanje informacija o kodiranju ulaznog video toka; sklopovske jezgre za ubrzanje; raznorodne arhitekture visokih performanci
• Language: english
• URN:NBN: urn:nbn:hr:168:782998
• Study programme: Title: Electrical Engineering and Computing; Study programme type: university; Study level: postgraduate; Academic / professional title: Doktor znanosti elektrotehnike i računarstva (Doktor znanosti elektrotehnike i računarstva)
• Type of resource: Text
• Extent: 131 str. ; 30 cm
• File origin: Born digital
• Access conditions: Open access
• Created on: 2020-01-21 12:04:25