Download presentation
Presentation is loading. Please wait.
Published bySidney Grice Modified over 10 years ago
1
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 1 Content-aware and context- aware resource adaptation Juan Hamers (06/06/2006)
2
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 2 Content providers offer multiple versions : which to choose Media stream low medium high small medium large 56k 100k 300k
3
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 3 2 Major issues for client when selecting media Decoder powerful enough –No skipping distortion Sufficient battery life Battery Empty
4
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 4 Impact of content itself on decoding time / Energy consumption Platform capacity
5
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 5 Exploiting this variance to adapt resources for energy efficiency 4080120 P(W) t (ms) 5 DVS (P ~ fV², V ~ f) frequency power
6
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 6 Oracle needed for accurate timing information Missed deadline Less Energy conserved 4080120 P(W) t (ms) 5
7
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 7 Content-provider has to add scenario information Platform & decoder independent Adequate Resources
8
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 8 Looking into content structure …… Decoder MB type Scaling
9
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 9 A Macroblock profile summarizes the structure of a GOP/frame Histogram
10
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 10 Macroblock profile relates to decode time MB- profile Decode time
11
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 11 Provider determines scenarios based upon macroblock profiles Content Provider
12
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 12 Client trains Scenario Adaptation Table with representative samples freqVEn (J)dtime (ms) 3001.26100 5331.3515200 4001.22510140 6661.620300 Client Side Content Provider
13
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 13 What about new content ???
14
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 14 Determining resource needs during handshake Energy / Time
15
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 15 Client uses scenario IDs and its SAT to scale frequency & voltage freqV 3001.2 5331.35 4001.225 6661.6 Decoder … …
16
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 16 Experimental setup 12 Video sequences (300 frames) Macroblock profiles per GOP/frame Train with 11, test other Cluster centroid for training target platform (+ Additional offset) 12/32 scenarios performed best
17
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 17 Predicting decode complexity
18
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 18 Decoder Adaption : 3 Cases considered
19
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 19 We realized an average energy reduction of 46%
20
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 20 Results energy prediction
21
Content-aware and context-aware resource adaptation Juan Hamers – 2006-06-06 Faculteit Ingenieurswetenschappen – Vakgroep Elektronica en Informatiesystemen pag. 21 Conclusion Macroblock profiles correlate well with decoding time/Energy usage. Recognizing scenarios derived from macroblock profiles allows for –efficient resource adaption –Predicting required resources –In a platform independent way Average energy reduction of 46% Used for resource prediction < 2 % error
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.