Performance in Operating Systems and Middleware Frank Feinbube - - PowerPoint PPT Presentation

performance
SMART_READER_LITE
LIVE PREVIEW

Performance in Operating Systems and Middleware Frank Feinbube - - PowerPoint PPT Presentation

Performance in Operating Systems and Middleware Frank Feinbube Operating Systems and Middleware Performance To the limit of computation! What is possible, what is not? Law-abiding hardware and operating system trends (Moore, Dennard,


slide-1
SLIDE 1

Performance

in Operating Systems and Middleware

Frank Feinbube Operating Systems and Middleware

slide-2
SLIDE 2

■ To the limit of computation! What is possible, what is not? ■ Law-abiding hardware and operating system trends (Moore, Dennard, ...) ■ Integrating accelerators with the operating system ■ GPU Virtualization ■ Operating System Support for Non-Volatile Memory ■ ...

Performance

Frank Feinbube, 14.04.2016 Performance @OSM Chart 2

slide-3
SLIDE 3

■ Landauer’s principle ■ Margolus-Levitin theorem ■ g64 ■ Quantum computers ■ Singularity “The speed of software halves every 18 months.” – Bill Gates

Frank Feinbube, 14.04.2016 Performance @OSM Chart 3

To the limit of computation! What is possible, what is not?

1 talk 1 student

slide-4
SLIDE 4

■ Moore, Dennard, Koomey ■ Gates, Wirth, Zawinski ■ Power Wall ■ Memory Wall ■ ILP Wall ■ Multi-Core ■ Neural Chips “Every program attempts to expand until it can read mail.”

  • Jamie Zawinski

Frank Feinbube, 14.04.2016 Performance @OSM Chart 4

1 talk 1 student

  • r

+new data =2 students

Law-abiding hardware and

  • perating system trends
slide-5
SLIDE 5

■ User Space APIs are not enough ■ Accelerating OS functionality

  • vs. Kernel API vs. transparent usage

■ PTasks, GPUStore, Gdev framework, … ■ IBM CAPI, ARM, …

Frank Feinbube, 14.04.2016 Performance @OSM Chart 5

Integrating accelerators (with the operating system)

Up to 3 talks 1 student

  • r

+project =2 students

CPU CPU CPU CPU Memory I/O Interconnect

GPU Integration

slide-6
SLIDE 6

■ Performance, Fidelity, Multiplexing, Interposition ■ Direct Device Attachment, API Forwarding, Mediated Device Attachment / Full Virtualization ■ NVIDIA Volta

Frank Feinbube, 14.04.2016 Performance @OSM Chart 6

GPU Virtualization

1 talk 1 student

  • r

+project =2 students

slide-7
SLIDE 7

■ Memristor, NVRAM, Phase-Change Memory… ■ Performance, Reliability, Security

„Those who cannot remember the past are condemned to repeat it.“ – George Santayana

Frank Feinbube, 14.04.2016 Performance @OSM Chart 7

(Operating System) Support for Non-Volatile Memory

1 talk 1 student

  • r

+project =2 students AOL 1996

Windows 8

2012

slide-8
SLIDE 8

Frank Feinbube, 14.04.2016 Performance @OSM Chart 8

Rack-Scale Resource Management

■ Data-Center Trend: High density of SoCs incorporated with few HEs of rackspace ■ SoCs may contain both general purpose and special purpose units ■ General purpose units may vary in type (performance/power/…) ■ Using high-speed interconnects, logical machines can be formed in accordance to specific workloads

1 talk 1 student

slide-9
SLIDE 9

Frank Feinbube, 14.04.2016 Performance @OSM Chart 9

[Your ideas go here]

Surprise us