https://www.mdu.se/

mdu.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Offloading Accelerator-intensive Workloads in CPU-GPU Heterogeneous Processors
Mälardalen University, School of Innovation, Design and Engineering, Embedded Systems.ORCID iD: 0000-0001-8096-3891
Mälardalen University, School of Innovation, Design and Engineering, Embedded Systems.ORCID iD: 0000-0003-3242-6113
Mälardalen University, School of Innovation, Design and Engineering, Embedded Systems. Unibap AB, Sweden.ORCID iD: 0000-0002-8785-5380
Mälardalen University, School of Innovation, Design and Engineering, Embedded Systems.ORCID iD: 0000-0002-1687-930X
Show others and affiliations
2021 (English)In: 26th IEEE International Conference on Emerging Technologies and Factory Automation, ETFA 2021, 2021Conference paper, Published paper (Refereed)
Abstract [en]

Autonomous vehicular systems require computer vision and intelligent on-board decision making functionalities that include a mix of sequential and parallel workloads. The execution times of the workloads and power consumption in these functionalities can be lowered by utilizing the accelerators (e.g., GPU) instead of running the workloads entirely on the host processing units (CPU). However, allocating all the parallelizable workload to accelerators can create a computation bottleneck in the accelerators that, in turn, can have an adverse effect on schedulability of the systems. This paper presents a novel framework that can allocate the accelerate-intensive workloads to the accelerators as well as to the non-accelerated host processing units. Within the context of this framework, the paper introduces five offloading techniques to mitigate the accelerator-intensive workloads by utilizing excess capacity of non-accelerated processing units under dynamic scheduling in CPU-GPU heterogeneous processors. The proposed techniques are evaluated using simulation experiments. The evaluation results indicate that one of the proposed techniques can achieve up to 16% improvement in schedulability of the task sets compared to the traditional non-offloading technique.

Place, publisher, year, edition, pages
2021.
National Category
Computer Systems
Research subject
Computer Science
Identifiers
URN: urn:nbn:se:mdh:diva-56081DOI: 10.1109/ETFA45728.2021.9613666ISI: 000766992600230Scopus ID: 2-s2.0-85122955086ISBN: 9781728129891 (print)OAI: oai:DiVA.org:mdh-56081DiVA, id: diva2:1599791
Conference
26th IEEE International Conference on Emerging Technologies and Factory Automation, ETFA 2021; Virtual, Vasteras 7 September 2021 through 10 September 2021
Available from: 2021-10-01 Created: 2021-10-01 Last updated: 2022-11-08Bibliographically approved
In thesis
1. Space Computing using COTS Heterogeneous Platforms: Intelligent On-Board Data Processing in Space Systems
Open this publication in new window or tab >>Space Computing using COTS Heterogeneous Platforms: Intelligent On-Board Data Processing in Space Systems
2021 (English)Doctoral thesis, comprehensive summary (Other academic)
Abstract [en]

Space computing enriches space activities such as deep-space explorations and in-orbit intelligent decision making. The awareness of space computing is growing due to the technological advances of high-performance commercial off-the-shelf (COTS) computing platforms. Space offers a complex, constrained and challengeable environment to the developers, researchers, as well as human beings. The challenges are size, weight and power (SWaP) constraints, real-time requirements, communication limitations as well as radiation effects. The research conducted in this thesis aims at investigating and supporting intelligent on-board data processing using COTS heterogeneous computing platforms in space systems. These platforms embed at least one Central Processing Unit (CPU) and one Graphics Processing Unit (GPU) on the same chip. 

The main goal of the research presented in this thesis is twofold. First, to investigate the heterogeneous computing platforms for the purpose of proposing a solution to tackle the above-mentioned challenges in space systems. Second, to complement the proposed solution with novel scheduling techniques for real-time applications that run on COTS heterogeneous platforms under harsh environments like space.

The proposed techniques are based on the system model that considers the use of alternative executions of parallel segments of tasks. Although offloading a parallel segment to a parallel computation unit (such as GPU) improves the best-case execution times of most applications, it can increase the response times of tasks in some applications due to the overuse of GPU. Hence, the use of the proposed task model can be a key to decrease the response times of tasks and improve schedulability of the system. The server-based proposed scheduling techniques support the proposed task model by guaranteeing the execution slot for parallel segments on CPU(s). 

The experimental evaluation conducted in this thesis shows that the proposed task model can improve the schedulability of the real-time systems up to 90% with the static allocation of applications. Moreover, the dynamic allocation method using the server-based scheduling with the proposed task model can improve the schedulability up to 16%. Finally, the thesis presents a simulation tool that simulates real-time applications using the proposed task model while considering the different levels of radiation tolerance to different processing units.

Abstract [sv]

Rymddata berikar rymdaktiviteter som utforskningar i djupa rymden och intelligent beslutsfattande i omloppsbana. Medvetenheten om rymddatorn ökar på grund av de tekniska framstegen inom högpresterande commercial off-the-shelf (COTS). Utrymme erbjuder utvecklare, forskare och människor en komplex, begränsad och utmanande miljö. Utmaningarna är storlek, vikt och effekt (SWaP), realtidskrav, kommunikationsbegränsningar samt strålningseffekter. Forskningen som bedrivs i denna avhandling syftar till att undersöka och stödja intelligent omborddatabehandling med hjälp av COTS heterogena datorplattformar i rymdsystem. Dessa plattformar bäddar in minst en Central Processing Unit (CPU) och en Graphics Processing Unit (GPU) på samma chip.

Huvudmålet för den forskning som presenteras i denna avhandling är tvåfaldigt. För det första att undersöka de heterogena dataplattformarna i syfte att föreslå en lösning för att hantera ovan nämnda utmaningar i rymdsystem. För det andra, för att komplettera den föreslagna lösningen med nya schemaläggningstekniker för realtidsapplikationer som körs på COTS heterogena plattformar under tuffa miljöer som rymden.

De föreslagna teknikerna är baserade på systemmodellen som överväger användningen av alternativa utföranden av parallella segment av uppgifter. Även om avlastning av ett parallellt segment till en parallell beräkningsenhet (t.ex. GPU) förbättrar de bästa tillämpningstiderna för de flesta applikationer, kan det öka svarstiderna för uppgifter i vissa applikationer på grund av överanvändning av GPU. Därför kan användningen av den föreslagna uppgiftsmodellen vara en nyckel för att minska responstiderna för uppgifter och förbättra systemets schemaläggning. De serverbaserade föreslagna schemaläggningsteknikerna stöder den föreslagna uppgiftsmodellen genom att garantera exekveringsplatsen för parallella segment på CPU (er).

Den experimentella utvärderingen som utförs i denna avhandling visar att den föreslagna uppgiftsmodellen kan förbättra schemaläggningen för realtidssystem upp till 90% med statisk tilldelning av applikationer. Dessutom kan den dynamiska tilldelningsmetoden som använder den serverbaserade schemaläggningen med den föreslagna uppgiftsmodellen förbättra schemaläggningen med upp till 16%. Slutligen presenterar avhandlingen ett simuleringsverktyg som simulerar applikationer i realtid med hjälp av den föreslagna uppgiftsmodellen samtidigt som man beaktar de olika nivåerna av strålningstolerans för olika behandlingsenheter.

Place, publisher, year, edition, pages
Västerås: Mälardalen University, 2021
Series
Mälardalen University Press Dissertations, ISSN 1651-4238 ; 347
Keywords
space computing, CPU-GPU heterogeneous computing, intelligent on-board data processing
National Category
Computer Systems
Research subject
Computer Science
Identifiers
urn:nbn:se:mdh:diva-56086 (URN)978-91-7485-528-9 (ISBN)
Public defence
2021-11-18, Alfa, Mälardalens högskola, Västerås, 13:15 (English)
Opponent
Supervisors
Available from: 2021-10-08 Created: 2021-10-01 Last updated: 2021-10-28Bibliographically approved

Open Access in DiVA

No full text in DiVA

Other links

Publisher's full textScopus

Authority records

Tsog, NandinbaatarMubeen, SaadBruhn, FredrikBehnam, MorisSjödin, Mikael

Search in DiVA

By author/editor
Tsog, NandinbaatarMubeen, SaadBruhn, FredrikBehnam, MorisSjödin, Mikael
By organisation
Embedded SystemsEmbedded Systems
Computer Systems

Search outside of DiVA

GoogleGoogle Scholar

doi
isbn
urn-nbn

Altmetric score

doi
isbn
urn-nbn
Total: 69 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf