Model- based design of energy- efficient applications for IoT systems Alexios Lekidis, Panagiotis Katsaros Department of Informatics, Aristotle University of Thessaloniki 1st International Workshop on Methods and Tools for Rigorous System Design (MeTRiD) Thessaloniki, Greece 15 April, 2018 A.U.Th. 1
Outline 1) Challenges towards energy estimation in the IoT ecosystem 2) Model- based characterization of energy consumption through the Contiki OS Rigorous system design method based on the BIP framework Accurate energy profiling through powertrace 3) Case study: Energy- aware building management system Application of the proposed method Requirement verification 4) Conclusion and ongoing work A.U.Th. 2
Outline 1) Challenges towards energy estimation in the IoT ecosystem 2) Model- based characterization of energy consumption through the Contiki OS Rigorous system design method based on the BIP framework Accurate energy profiling through powertrace 3) Case study: Energy- aware building management system Application of the proposed method Requirement verification 4) Conclusion and ongoing work A.U.Th. 3
IoT applications: Constraints Resource limitations (e.g. memory, CPU, battery) System heterogeneity Sensors, actuators Operating systems (e.g. Android, ios, Contiki OS, TinyOS) Web service interaction patterns (e.g. REST) Connectivity (e.g. WiFi, ZigBee, Bluetooth, NFC) Measurement units (e.g. Celsius, Fahrenheit) Overall code complexity A.U.Th. 4
Main challenges towards IoT adoption privacy storage implementation security IoT connectivity Energy management standardization A.U.Th. 5
Main challenges towards IoT adoption privacy storage implementation security IoT connectivity Energy management standardization A.U.Th. 6
Why energy is important? A.U.Th. 7
IoT devices Usually battery supply to widen the applicable deployment possibilities A.U.Th. 8
Existing approaches Special purpose tools to provide feedback about overall energy consumption by simulation or after the deployment ü fine-grained analysis of the energy consumption at the network-level Direct interaction with device hardware (not always supported) Device manufacturer characteristics, are not always accurate when compared with real measurements A.U.Th.
Solution: Energy characterization Ø Method allowing the proper characterization of all the parameters and scenarios that are impacting the energy consumption on a system-level Energy characterization through distribution fitting Energy evolution estimation over time Average power consumption of the device (Source: Borja Martinez, Marius Monton, Ignasi Vilajosana & Joan Daniel Prades (2015): The power of models: Modeling power consumption for IoT devices. IEEE Sensors Journal 15(10), pp. 5777 5789) A.U.Th. 10
Outline 1) Challenges towards energy estimation in the IoT ecosystem 2) Model- based characterization of energy consumption through the Contiki OS Rigorous system design method based on the BIP framework Accurate energy profiling through powertrace 3) Case study: Energy- aware building management system Application of the proposed method Requirement verification 4) Conclusion and ongoing work A.U.Th. 11
Introduction to Contiki IoT systems Modular: layered system construction Full support from application development libraries to integration of IoT platforms Native simulation environment (i.e. Cooja) Loosely coupled REST web services for IoT application development A.U.Th. 12
Energy parameter categories Ø Analysis remark: The energy consumed when a device is in transmitting/receiving mode is up to 5 times greater than in any other state Parameters influencing transmit/receive functionalities derive in their majority from the network stack Grouping according to the layers of the Contiki stack they belong MAC layer Application layer Physical layer A.U.Th. 13
Energy parameter categories Ø Analysis remark: The energy consumed when a device is in transmitting/receiving mode is 5 times greater than in any other state Parameters influencing transmit/receive functionalities derive in their majority from the network stack Grouping according to the layers of the Contiki stack they belong MAC layer Application layer Physical layer A.U.Th. 14
Energy parameter categories MAC Radio duty cycling mechanism A.U.Th. 15
Energy parameter categories CoAP vs MQTT usage in IoT applications Application Protocol choice up to the application needs Performance (i.e. CoAP) vs reliability (i.e. MQTT) Header should contain all the contextual info for packet identification In scenarios as packet forwarding compression/decompression is very energy demanding A.U.Th. 16
Energy parameter categories Ø Definition: Interference is defined in the form of additive noise from simultaneous transmissions with the same radio frequency from proximity networks Increased packet collision Nodes remain in Tx for longer time durations CommMedium A.U.Th. 17
Proposed method A.U.Th. 18
Proposed method A.U.Th. 19
Proposed method A.U.Th. 20
Proposed method A.U.Th. 21
Proposed method A.U.Th. 22
Modeling Contiki IoT systems in BIP [Wiley SPE, 2018] BIP models for every level of the IoT architecture with two layers: RESTful Application Model (REST module allocated to every node) Contiki Kernel Model (Contiki OS, protocol stack) A.U.Th.
Proposed method A.U.Th. 24
Powertrace Contiki library for monitoring the energy flow in IoT devices Monitoring in distinct operating modes: Low Power (LPM): idle device waiting for events CPU: used for calculations/data processing Radio transmission (Tx): data transmission Radio reception (Rx): data reception Duty cycle: percentage of time that a device remains in one operating mode Lifetime: total time duration for autonomous operation A.U.Th. 25
Energy model Injects energy- oriented behavior and characteristics to the model for every operating mode of each device: Calibrated by probabilistic distributions Obtained from the analysis of debugging traces from the Contiki simulation environment as well as the powertrace module λ Rx λ Tx ρ 2 ρ 1 A.U.Th.
Outline 1) Challenges towards energy estimation in the IoT ecosystem 2) Model- based characterization of energy consumption through the Contiki OS Rigorous system design method based on the BIP framework Accurate energy profiling through powertrace 3) Case study: Energy- aware building management system Application of the proposed method Requirement verification 4) Conclusion and ongoing work A.U.Th. 27
Building Management System topology Aim: Energy management through remote control of buildings by a WAN network that consists of multiple WPAN networks, one for each building floor WAN network WPAN network WPAN network A.U.Th. 28
BMS network architecture Network switch B- RTR SimpleLink Sensortag controller B- ASC Floor 4 CoAP / MQTT B- RTR B- RTR OpenMote controller Sky mote controller B- ASC B- ASC Floor 3 Floor 2 B- RTR Zolertia Z1 controller B- ASC Floor 1 A.U.Th. 29
Verification of example requirements Concern the IoT device lifetime, as well as the IoT device dutycycle in different operating modes Requirement 1. Device lifetime should be at least 1 week. Requirement 2. The duty-cycle in the LPM mode should remain higher than 90% during working hours. Requirement 3. The duty-cycle in the Rx mode should not exceed 20% during working hours. A.U.Th. 30
Verification of example requirements Concern the IoT device lifetime, as well as the IoT device dutycycle in different operating modes Requirement 1. Device lifetime should be at least 1 week. Requirement 2. The duty-cycle in the LPM mode should remain higher than 90% during working hours. Requirement 3. The duty-cycle in the Rx mode should not exceed 20% during working hours. A.U.Th. 31
Energy parameter impact in device lifetime φ " = lf 168 P(φ " ) = 0.9 for: 1) fixed default parameter values 2) parameter values within the allowed tolerance A.U.Th. 32
Verification of example requirements Concern the IoT device lifetime, as well as the IoT device dutycycle in different operating modes Requirement 1. Device lifetime should be at least 1 week. Requirement 2. The duty-cycle in the LPM mode should remain higher than 90% during working hours. Requirement 3. The duty-cycle in the Rx mode should not exceed 20% during working hours. A.U.Th. 33
Duty cycle during working hours φ D = D FG 20% P(φ D ) = 0.8 with the Low Power Probing duty cycle protocol P(φ D ) = 0 without a duty cycle protocol A.U.Th. 34
Conclusions Νovel method for characterizing the energy consumption in IoT applications and the individual IoT devices Energy- aware parameter configuration RESTful service- based applications over Contiki OS nodes Validating requirements related to energy characteristics Building Management System consisting of various devices (e.g. Zolertia Z1, Sky mote, OpenMote, SimpleLink Sensortag) System requirements concerning device lifetime and duty cycle A.U.Th. 35
Perspectives Energy optimization techniques for the IoT applications Large- scale testbed to demonstrate the scalability of the proposed method Impact of remote control in the overall energy consumption of the building Smarter logic actions in the Building Management Controller (e.g. shutting down the heating and lighting system in the absence of motion) A.U.Th. 36
ARISTOTLE UNIVERSITY OF THESSALONIKI Thank you for your attention. Questions? Further info: alekidis@auth.gr, katsaros@csd.auth.gr