Reliable Dynamic Embedded Data Processing Systems Twan Basten ‘Knowing is not understanding.’ Charles Kettering Twan Basten Joint work with Marc Geilen, AmirHossein Ghamarian, Hamid Shojaei, Sander Stuijk, Bart Theelen, Yang Yang and others Funding: NWO PROMES EC FP6 Betsy, FP7 MNEMEE SenterNovem Octopus Embedded Data Processing Systems 2 sony 2 philips asml océ thales sony apple 3 Embedded Data Processing Systems sony sony 4 Embedded Data Processing Systems
11
Embed
Reliable Dynamic Embedded Data Processing Systemstbasten/presentations/ipa20091126.pdf · 2009-11-29 · Reliable Dynamic Embedded Data Processing Systems Twan Basten ‘Knowing is
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Reliable Dynamic Embedded Data Processing Systems
Twan Basten
‘Knowing is not understanding.’
Charles Kettering
Twan Basten
Joint work with
Marc Geilen, AmirHossein Ghamarian, Hamid Shojaei, Sander Stuijk, Bart Theelen, Yang Yang
and others
Funding:
NWO PROMES
EC FP6 Betsy, FP7 MNEMEE
SenterNovem Octopus
Embedded Data Processing Systems2
sony
22
philips
asml
océ
thales
sony
apple
3 Embedded Data Processing Systems
sonysony
4 Embedded Data Processing Systems
Trend: Dynamic Behavior5
• Multiprocessor systems / networking / concurrency
• Application convergence / application evolution
• Intra-application dynamism
• Inter-application dynamism
What is the optimal configuration?
- given applications starting/stopping over time
- given data-dependent execution times
- given varying communication bandwidth
- given battery status
- …
6
Intra-application dynamism
Dataflow analysis
7 Dataflow Analysis: Challenges
predictability and efficiency
… are contradictory
8 Synchronous Data Flow Graphs
A,1 B,2 C,22 3 1 2
2 3 1 2
1 1 1 1 1 1
4 2
channelratetoken
execution time
actor
(SDFGs [Lee 1986])
1 1
1
1 1
1
1 1
1
9 Scenario-aware MPEG-4 SP Decoder Model
1
1VLD IDCT
RCMCFD1
1
1
1
11
1 e
ddd
a
bc c
Kernel
Parameterized
rate
Control channelData
channel
Fixed rate
Execution time
VLDP0 0
others 40
IDCTP0 0
others 17
I, P0 0
P30 90
P40 145
P 190
(FSM-based Scenario-Aware DataFlow (SADF))
RCMCFD
3
1 11
Detector Tokens
I P99 P99
P0
...
1
Rate I Px P0
a 0 1 0
b 0 x 0
c 99 x 1
d 1 1 0
e 99 x 0
MCP50 190
P60 235
P70 265
P80 310
P99 390
RC
I 350
P0 0
P30, P40, P50 250
P60 300
P70, P80, P99 320
FD All 0
x={30,40,50,60,70,80,99}
MEMOCODE 2006
10
Throughput analysis for SDF
ACSD 2006
11 Synchronous Data Flow Graphs
A,1 B,2 C,22 3 1 2
2 3 1 2
1 1 1 1 1 1
4 2
channelratetoken
execution time
actor
(SDFGs [Lee 1986])
1 1
1
1 1
1
1 1
1
Throughput: average number of actor firings over time
Iteration: smallest non-empty set of actor firings that does not change
the token distribution (example: A:3, B:2, C:1)
12 SDF Throughput Analysis
A
A, C
A, CB
B
A
Periodic PhaseTransient Phase
Self-timed execution:
A
BB
Throughput can be calculated from the periodic phase
Efficient implementationConsider one designated firing per iteration only to detect a recurrent state
13 Throughput Analysis: Our Result
State
Space
Dasdan
Gupta
Howard Young
Tarjan
Orlin
MP3 dec. 1. 10-3 1. 10-3 1. 10-3 1. 10-3
Modem 1.10-3 82. 10-3 81. 10-3 81. 10-3
Traditional methods
Modem 1.10 82. 10 81. 10 81. 10
Sample Rate 2. 10-3 >1800 >1800 >1800
Satellite 56. 10-3 >1800 >1800 >1800
H.263 decoder 10. 10-3 >1800 >1800 >1800
Runtimes of various throughput analysis methods (in seconds)
increase buffers with n times the minimal step size
22 Trade-off Analysis
• Start small, recursively resolve bottlenecks
• Fast in general, approximation possible
• Generalizable to other resources, computational models
23
Scenarios
24 Scenario-aware Throughput Analysis
• SDF throughput analysis considers worst-case actor execution times
• Scenario-aware throughput analysis considers worst-case execution times per scenario, but it must consider scenario transitionstimes per scenario, but it must consider scenario transitions
• Analyzing all scenario transitions separately can be avoided
• Separate scenario transitions with an invariant reference schedule