Cyber-Physical Event Processing Chao Wang CSE 520S References - - PowerPoint PPT Presentation

Cyber-Physical Event Processing

Chao Wang

CSE 520S

References

Ø Core material of this lecture:

Wang, C., Gill, C. and Lu, C., 2017, June. Real-time middleware for cyber-physical event processing. In Quality of Service (IWQoS), 2017 IEEE/ACM 25th International Symposium on (pp. 1-6). IEEE.

Ø Other references:

q The specifications of CORBA, Real-time CORBA, and DDS

http://www.omg.org/technology/documents/vault.htm

q TAO http://www.cs.wustl.edu/~schmidt/TAO.html q Vasisht, D., Kapetanovic, Z., Won, J., Jin, X., Chandra, R., Sinha, S.N., Kapoor, A., Sudarshan,

• M. and Stratman, S., 2017, March. FarmBeats: An IoT Platform for Data-Driven
• Agriculture. In NSDI(pp. 515-529).

q Khandeparkar, K., Ramamritham, K. and Gupta, R., 2017. QoS-Driven Data Processing

Algorithms for Smart Electric Grids. ACM Transactions on Cyber-Physical Systems, 1(3), p.14.

q Cugola, G. and Margara, A., 2012. Complex event processing with T

• REX. Journal of

Systems and Software, 85(8), pp.1709-1728.

q Mayer, R., Mayer, C., Tariq, M.A. and Rothermel, K., 2016, June. GraphCEP: Real-time data

analytics using parallel complex event and graph processing. In Proceedings of the 10th ACM International Conference on Distributed and Event-based Systems(pp. 309-316). ACM.

10/9/17

1

Industrial Internet of Things (IIoT)

Ø IIoT = Cyber-physical systems + Cloud Ø Cyber-physical event processing as a service

10/9/17

2

Application example

Ø Monitoring a farm of 100 acres (i.e., 75+ football fields)

10/9/17

3

Vasisht, D., Kapetanovic, Z., Won, J., Jin, X., Chandra, R., Sinha, S.N., Kapoor, A., Sudarshan, M. and Stratman, S., 2017, March. FarmBeats: An IoT Platform for Data-Driven Agriculture. In NSDI(pp. 515-529).

Application example

Ø Monitoring a nation-wide smart electric grid

10/9/17

4

Khandeparkar, K., Ramamritham, K. and Gupta, R., 2017. QoS-Driven Data Processing Algorithms for Smart Electric Grids. ACM Transactions on Cyber-Physical Systems, 1(3), p.14.

Application example

Ø Predictive maintenance

q The prediction of and response to component failures q E.g., maintaining wind turbines in a wind farm

10/9/17

5

Cyber-physical event processing as a service

Ø Example: sensor fusion

q Reducing data noises q Creating frequency domain representation q Concatenating results for a broaden spectrum assessment

10/9/17

6 EKF: Extended Kalman Filter FFT: Fast Fourier Transform CAT: Concatenation

Research challenges

Ø Enforcing QoS policies

q e.g., latency differentiation

Ø Computationally intensive

q e.g., sensor fusion q High input rate

Ø Event temporal validity

q Freshness of sensing data

10/9/17

7

Related work

Ø Real-time event services

q Real-time CORBA

• Industry standard
• Predictable and interoperable data exchanges between systems

q TAO

• CORBA-compliant C++ open-source implementation
• Widely used in industry in the past two decades

q Data Distribution Service (DDS)

• Industry standard
• Data-centric paradigm; publish-subscribe pattern

Ø Complex event processing (CEP)

q T

• REX
• Efficient implementation for a rich set of event processing rules

q GraphCEP

• Social network analysis (e.g., Facebook post ranking)

10/9/17

8

Related work

Ø Real-time event services

q Real-time CORBA

• Industry standard
• Predictable and interoperable data exchanges between systems

q TAO

• CORBA-compliant C++ open-source implementation
• Widely used in industry in the past two decades

q Data Distribution Service (DDS)

• Industry standard
• Data-centric paradigm; publish-subscribe pattern

Ø Complex event processing (CEP)

q T

• REX
• Efficient implementation for a rich set of event processing rules

q GraphCEP

• Social network analysis (e.g., Facebook post ranking)

10/9/17

9

Do not perform complex event processing …

Related work

Ø Real-time event services

q Real-time CORBA

• Industry standard
• Predictable and interoperable data exchanges between systems

Ø Complex event processing (CEP)

q T

• REX
• Efficient implementation for a rich set of event processing rules

q GraphCEP

• Social network analysis (e.g., Facebook post ranking)

10/9/17

10

Do not perform complex event processing … Do not differentiate application QoS … Do not support real-time at the millisecond scale …

Cyber-Physical Event Processing (CPEP)

Real-time CPEP middleware

✓ Processing prioritization ✓ Processing sharing ✓ Enforcing temporal validity

10/9/17

11

Wang, C., Gill, C. and Lu, C., 2017, June. Real-time middleware for cyber-physical event processing. In Quality of Service (IWQoS), 2017 IEEE/ACM 25th International Symposium on (pp. 1-6). IEEE.

Ø Constructing the graph of event streams from a file

q Operator arrangement q Consumer priority levels

Ø Prioritizing operators

q Priority propagation from consumers

System configuration

10/9/17

12

• 7
• 6
• 5
• 4
• 3
• 2
• 1

s1 s2 s3 s5 s4

High priority Middle priority Low priority

c1 c2 c3 c4

Low priority

Processing prioritization

10/9/17

13

Schedule the threads using a fixed-priority preemptive scheduling policy

Processing sharing

10/9/17

14

Save the time for lower-priority processing! Movers don’t interfere with higher- priority processing!

Enforcing temporal validity

Absolute validity intervals:

10/9/17

15

The interval for S2 The interval for S1 The interval for S3

An event for C2 is temporally valid

• nly before timepoint t4

Enforcing temporal validity

Absolute validity intervals:

10/9/17

16

The interval for S2 The interval for S1 The interval for S3

An event for C2 is temporally valid

• nly before timepoint t4

CPEP sheds invalid events for both correctness and efficiency

CPEP implementation

Ø Implemented based on RT

• CORBA event service

q TAO real-time event channel (version 2.3.0)

Ø Efficient memory management

q Zero-copy of same event for downstream operators

Ø Replacing older events by new instances

q Data freshness for physical states

10/9/17

17

Empirical evaluation

Ø Experimentation setup Ø Evaluating performance on

q Prioritization q Sharing q Shedding

Ø Workload: sensor fusion emulation, etc.

q Consumers of multiple priority levels q Suppliers of multiple sending rates q Sharing of operators

10/9/17

18

Machine 2 (“The middle box”)

CPEP

Machine 1 Event suppliers Machine 3 Event consumers

Graphs of event streams #1

Ø Consumers of multiple priority levels Ø Suppliers of multiple sending rates

10/9/17

19

Graphs of event streams #2

Ø Sharing of operators Ø Non-sharing

10/9/17

20

High priority Middle priority Low priority EKF3 FFT3 CAT2 AES2 FFT4 c2 s5 s6 EKF4 s7 s8 EKF1 FFT1 CAT1 AES1 FFT2 c1 s1 s2 EKF2 s3 s4 CAT3 AES4 c3 AES3 100 Hz 100 Hz

High priority Middle priority Low priority EKF4 FFT4 CAT2 AES2 FFT5 c2 s5 s6 EKF5 s7 s8 EKF1 FFT1 CAT1 AES1 FFT2 c1 s1 s2 EKF2 s3 s4 CAT3 AES4 c3 AES3 100 Hz 100 Hz EKF6 FFT6 FFT7 EKF7 FFT3 EKF3

CPEP prioritization: results

10/9/17

21

Protect the latency of high-priority events against contention

CPEP sharing: results

10/9/17

22

Sharing reduced the latency of lower-priority processing

Without sharing, no low-priority events were produced!

CPEP shedding: results

10/9/17

23

Valid throughput = temporally valid events / second Shedding can improve temporally valid throughput

CPEP shedding: results

10/9/17

24

Valid throughput = temporally valid events / second Shedding can improve temporally valid throughput

Summary

Ø Cyber-physical event processing is vital for Industrial IoT Ø Three main contributions of the CPEP middleware

✓ Processing prioritization ✓ Processing sharing ✓ Enforcing temporal validity

10/9/17

25