ensq/dv2524

DV2524 Vt14 Lp34 Examensarbete i datavetenskap för civilingenjör - Master's thesis in Computer Science at Blekinge Institute of Technology in collaboration with Intel Corporation. Tentative title: Paravirtualizing OpenGL ES in Simics; Author: Eric Nilsson; University advisor: Prof. Håkan Grahn, Research Dean of Faculty.

Paravirtualizing OpenGL ES in Simics

Master's Thesis in Computer Science

Author: Eric Nilsson ([email protected])

• Blekinge Institute of Technology
• Intel Corporation

Abstract

Full-system simulators provide benefits to developers in terms of a more rapid development cycle; since development may begin prior to that of next-generation hardware being available. However, there is a distinct lack of graphics virtualization in industry-grade virtual platforms, leading to performance issues that may obfuscate the benefits virtual platforms otherwise have over execution on actual hardware. This dissertation concerns the implementation of graphics acceleration by the means of paravirtualizing OpenGL ES 2.0 in the Simics full-system simulator. Furthermore, this study illustrates the benefits and drawbacks of paravirtualized methodology, in addition to performance analysis and comparison with the Android Emulator; which likewise utilize paravirtualization to accelerate simulated graphics.

In this study, we propose a solution for paravirtualized graphics using Magic Instructions; the implementation of which is subsequently described. Additionally, three benchmarks are devised to stress key points in the developed solution; comprising areas such as inter-system communication latency and bandwidth. Furthermore, the solution is evaluated based on computationally intensive applications. For the purpose of this study, elapsed frame times for respective benchmarks are collected and compared with four platforms; i.e. the hardware accelerated Host machine, the paravirtualized Android emulator, the software rasterized Simics- and the paravirtualized Simics platforms.

This thesis establishes paravirtualization as a feasible method to achieve accelerated graphics in virtual platforms. The study shows graphics acceleration of up to 34 times of that of its software rasterized counterparts. Furthermore, the study establishes magic instructions as the primary bottleneck of communication latency in the devised solution.

dv2524's Issues

Restructure question formulations under Research Questions

Research questions bör beskriva bl.a. metoder för att uppnå målen. Dessa bör betona generella lösning och analys för paravirtualisering av drivrutiner. Eric kommer sammanfatta frågeställningarna i respektive sektion under Research Questions och ställa cirka en frågeställning per område i början av kapitlet. Genomgående för hela dokumentet är att detta bör generaliseras mera.

Describe benchmarking and current situation

Citations that refer to no specific pages ought not be referring to bibliography

Refer to Further Reading-type section instead.

Expand upon Expected Outcomes

Delete branch dv2524_proposal

Formatting and structural section

In said section, in the preface of the thesis document, expand on the following:

• Bibliography styles
• Any formatting that may need explanation (how abbreviations are written, etc.)
• How citations are written
• Hyperlinks
• Structural section in accordance to book structure (wikipedia)
• Where to access later revisions of document
• Licensing of source code of document
• Copyright and licensing
• Contributing & contact
• Strongly advise reading document digitally (in regards to hyperlinks)
• Suggestions on how the user may benefit from reading certain attached section under Appendix beforehand.

Segment Web bibliography into different sections depending on source material

I.e., blog articles, technical docs, etc.

Commit specification in revision footer ought be a hyperlink to repository

Fewer bibliography categories

Write script to combine .bib source files into used bibs.

Inconsistencies in bibliographies

Bib-Digital Jazz, Man:

• Is Bergan_Devietti_Hunt_Ceze_2011 published in anything? Move to 'unpublished'-bib?
• Is Bartholomew:2006:QMM:1134160.1134163 really a scientific journal entry?
• Can more information be found concerning the Holzmann entry into the Dr. Dobb's journal?
• Rename Paper-references to Unpublished or similar?

Establish thesis title

Proposed Titles:

• Paravirtualization of OpenGL ES in Simics
• Paravirtualization of Graphics APIs in the Simics Full-system Simulator
• Achieving Graphics Acceleration by Paravirtualization in the Simics Full-system Simulator
• Paravirtualization of Graphics APIs in Virtual Platforms

Too many research questions?

Bring up with supervisors.

Acknowledgement

Inspiring quote

Expand upon Aims & Objectives

The Aim and Objectives of proposed study is not only implementation. Be sure to relate the purposes of developing methods and techniques for efficient communication and general execution in the Simics run-time environment. Additionally, comparisons with the Reference solution ought not only be related to performance.

Command to format numbers

Rename citation labels in accordance to type

Concatenate Proposal-commits into single commit

Terms command

Established terms, such as Checkpointing or Reverse Execution ought to be formatted with capitals the first time they're mentioned but formatted as plain text thereafter (e.g., checkpointing or reverse execution).
Create a command to support this.

Expand upon Risks

• Implementation takes longer than expected.
• Access to technical information and knowledge surrounding the Simics run-time environment.
• Support
• Relevant test-cases.

Expand upon Risks

• Implementation takes longer than expected.
• Access to technical information and knowledge surrounding the Simics run-time environment.
• Support

Implement minor changes as suggested by corporate supervisor

Conditional caponce-command does not function in section- or chapter headers

Add citation tooltips

http://tex.stackexchange.com/questions/15356/showing-the-bibliographic-entry-in-a-popup-when-you-hover-over-the-citation-key

Conditional caponce-command is incoherent with beginning of sentences

Terminology chapter

Compile a terminology chapter to describe concepts mentioned throughout the report (aka a glossary).

General & Abbreviations:

• API
• ABI
• CPU
• GPU
• GPGPU
• Rasterization
• Software Rasterization
• TTM - Time to Market
• SDK
• Hardware Acceleration
• Deterministic Algorithm
• Virtual Memory
• Paging
• Callback
• Texture
• State
• MIPS
• ISA
• R/W

APIs & Technologies:

• OpenGL
• OpenCL
• OpenGL ES
• EGL (XGL, AGL, and WGL)
• The TCP-, IP-, and MAC-protocols.
• PCI
• ARM
• X11

Simulatory:

• Paravirtualization
• Interpreters
• Just-in-Time Compilation
• Direct Virtualization
• Hardware Virtualization
• Hardware-assisted Virtualization
• Technologies such as KVM and whatever abbreviations Intel and AMD uses to support it.
• Emulation & Simulation and the difference thereof.
• Virtual Platform
• Timing

Organizations:

• Google (any special part of Google developed the Android Emulator?)
• Khronos Group
• QEMU Team
• Intel
• Virtutech
• Wind River

Products:

• Microsoft Windows
• Linux
• OS X
• Solaris
• CUDA

Conditional caponce-command does not function with separator symbol ~

Commands defined in dv2524commands.tex need a standardized prefix to distinguish them from package commands

Appendix section describing Simics

• Convey that simics may be run using networked machines to simulate large systems (becuase it's cool).
• Simics fundamentals (cha. 2) has some great info.

Eliminate unsupported claims or find viable sources

Don't specify page numbers in citations.

Alternative?

Re-write bit on Trans-simulation R/W

Nomenclature section

Nomenclature Section to describe code names for solutions. The section ought detail the differences in-between emulation and simulation, in addition to terms and concepts being used interchangeably.

General:

• Pre-Silicon Development

Simulatory:

• Host
• Target (in some other materials called Guest)
• Checkpointing
• Reverse Execution
• Deterministic Execution
• Magic Instruction
• Timing

Products:

• Simics FS

Concepts:

• Reference Implementation
• Reference Solution

Further Reading Section

• QEMU
• Simics
• Virgil3D (?)
• Android SDK Emulator

Web sources need be reviewed

Bibliography standard

The main bibliographies uses IEEE. Which standard ought the citations be using to differentiate them from peer-reviewed sources?

Move implementational details under Research Methodology to suitable chapter (.e.g, appendix)

Further Reading chapter

Further Reading Section to guide the reader to more material on the subject (e.g. non-valid references and technical documentation).

Possible risk: Virtual time and experiment methodology

Paragraph structure and indents

Scratch hypothesis

Specify motivation as to why we're not using an established OpenGL ES 2.0 benchmark such as Glmark2

Possible inherent flaw in OpenGL in terms of Determinism

Introduce under suitable section, probably under Appendix. Not a risk!

• Inconsistencies in OpenGL implementations may cause different resultsin-between host systems (e.g. minor color variations and rounding). Such errors, if calculated upon (e.g. zipping a screenshot in the target system) would cause in-determinism in Simics.

Command for abbreviations

Create command which outputs an entire phrase the first time it's mentioned, but the abbreviation otherwize.

Specify supported APIs

Command to format citations with specified pages

Double column bibliographies

Index chapter

Contribution and Expected Outcomes ought be merged into a new chapter

Said chapter need detail implementational details and technologies used. As such, corresponding sections from Research Methodology may be merged into the new chapter. In this way, the text must be extensive enough for others to do similar implementation based on the work described.

En god idé vore att inkludera identifiering av svårigheter och rekommendationer för framtida studier inom området i detta kapitel.

Page numbering formatting errors

Consider removing footer entirely if this is not solved in due time for the deadline.

Expand upon Research Methodology

Describe how the case study / literature study is to be performed along with what types of measurements are to be collected for the experiment.
Suggestion: describe possible benchmarking possibilities (i.e., number of GL-invocations, large data).