Cloud Expo 2014-11-05 - OnLive Presentation - No Video

14
Making Interactive Video Streams Truly Instant November 2014

Transcript of Cloud Expo 2014-11-05 - OnLive Presentation - No Video

Making Interactive Video

Streams Truly Instant

November 2014

US, EU, International Patents, Patents Pending

Cloud Apps

Page 1

• Cloud apps are not new • But they have been transactional• Cloud video is solved (buffered)• Cloud interactive Graphics

Intensive Applications – (GIAs) –much harder

US, EU, International Patents, Patents Pending

GIAs have been Tethered and Inflexible

Page 2

You want to interact with a powerful graphics app on BYOD or location independent. But:• Supercomputer interactivity has

required presence• Or you can watch it

o No real-time controlo Limited remote function

• You can use cloud benefits of resources, but no cloud benefits of computer/device-independence or use-anywhere

US, EU, International Patents, Patents Pending

0

500

1,000

1,500

2,000

2,500

3,000

3,500

4,000

0%

5%

10%

15%

20%

25%

30%

2004A 2005A 2006A 2007A 2008A 2009A 2010A 2011A 2012A 2013A 2014E

GIA

Clo

ck S

peed R

equirem

ents

2 (M

Hz)

% o

f D

evic

es C

apable

of

Handlin

g G

IAs

1Growing Number of Devices Unable to Meet

Increasing Processor Demands of GIAs

Page 3

Source: Gartner Reports

1. Calculated based on desktop and laptop shipments as a percentage of total device shipments, which include mobile phones, laptops, desktops, tablets and ultra mobiles

2. Based on system requirements of AutoCAD in respective years

Growing GIA Requirements vs. Decreasing Device Capability

US, EU, International Patents, Patents Pending

Two Myths (Among others)

Page 4

• Defies laws of physics

• Prohibitive cost ($89B !?!?)

US, EU, International Patents, Patents PendingPage 5

Simple Enough…Cloudifying GIAs requires:

• Encoding: quality and speed• Performance on user’s network • Reliable and fast input • Cloudify any as-yet-unclouded resources

• Motherhood:

Operational Reliability Global

Any End-User-Device Standard Equipment

Any Broadband Network Secure

US, EU, International Patents, Patents PendingPage 6

Display

DecodeNetwork (down)Encode

Application Pipeline

Input

180ms end-to-end

Latency is a Budget

MYTH: If I had zero-time encoder, I will be done TCP hurts; we use UDP – why?

Network (up)

US, EU, International Patents, Patents Pending

Reality Bites…

Page 7

• Networks are certainly imperfect. • Wired networks are lossy too.• DSL, cable, managed networks: all different.• Different vendors have different network behavior.• Wireless networks differ greatly.

• It’s not always obvious what matters, sometimes app-specific• Video harder, audio more immediately noticeable. Audio tricks don't

translate well for video.• Framerate vs. resolution vs. colorspace vs. smoothness vs. ….

• You can’t lose any input events.• Fix/anticipate/prevent/ignore errors.

• Some app virtualizations are hard.

• Your Mileage Will Vary

US, EU, International Patents, Patents Pending

Packet Loss is Real

MYTH: It’s all about the bandwidth. Note: LTE Low Latency profile is fast but “best effort”

Observed Sep/2014, consumer game service, all carriers, US only• 25% of sessions have ≥ 1.5% packet loss• 10% of sessions have ≥ 7.5% packet loss

Observed July-Aug 2014, sessions believed to involve WiFi• 35.9% of sessions have ≥ 1% loss peaks• 19.7% of sessions have ≥ 5% loss peaks• 14.1% of sessions have ≥ 10% loss peaks

DSL: impulse noise?Cable: neighborhood congestion?WiFi, cellular, & last 10 feet: everything?Fiber: ???Internet: routing and TCP practices

Page 9

US, EU, International Patents, Patents Pending

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

Impact of Packet Loss on QoSUsability across Bandwidth Ranges as a Function of Packet Loss

Source: Based on testing of publicly available sources

Packet Loss

Page 9

Packet Loss

Packet Loss Packet Loss

Packet Loss

App is usable and responsive to user input;

User experience is good / acceptableApp responsiveness / video quality is no

longer acceptable; App is not usable

App stops working altogether;

App session fails

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

Packet Loss

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

Packet Loss

0% 1% 2% 3% 4% 5% 6% 7% 8% 9% 10%

12 Mbps

7 Mbps

5 Mbps

3 Mbps

Packet Loss

Note: Hitching is present at all data-rates. Cells

marked in yellow are "distracting" levels of hitching.

C4

C1

C2

C3

C5

C6

C7

App Fails

Session Fails

Good/Acceptable

US, EU, International Patents, Patents Pending

0%

20%

40%

60%

80%

100%

7Mbps 5Mbps 4Mbps 2Mbps

OnLive Competitors

Bandwidth’s Impact on QoS

WiFi Wired LTE

% of Real World Connections Supported with Good Performance (30fps)

0%

20%

40%

60%

80%

100%

45Mbps 12Mbps 7Mbps 5Mbps 4Mbps 2Mbps

OnLive Competitors

0%

20%

40%

60%

80%

100%

45Mbps 12Mbps 7Mbps 5Mbps 4Mbps 2Mbps

OnLive Competitors

Source: Based on testing of publicly available sources

Others OthersOthers

Page 11

US, EU, International Patents, Patents Pending

Making Interactive Video Streams Truly Instant

Page 11

Musts:• Immediate error-free input• Immediate output (latency, not bandwidth) - audio & video• Fix/anticipate/prevent/ignore errors • Eliminate round trips• Be smart about network routing

Device agnostic, secure, accessible, etc…

MYTH: Scaling and tenancy

No Magic: App Needs * Count ≤ avail resources

US, EU, International Patents, Patents Pending

Making it Work

Page 12

• There’s no Silver Bullet

• Lots of potentially killer little problems

• It’s Possible!

Thank You