Arda Gumusalan CS788Term Project 2 1
2
Logical topology formation. Effective utilization of communication channels. Effective utilization of energy. 3
4
Exploits the tradeoff between CPU speed and time. Saves computation energy by simultaneously reducing CPU supply voltage and frequency however now computation takes longer period (direct concern in realtime systems). 5
CPU power consumption has two components: static and dynamic. Static power consumption is necessary to keep the circuit on. Dynamic power consumption is dissipated when CPU executes tasks. Dynamic power consumption typically dominates static but not always! Lower frequency -> Dynamic Static 6
There is a trade-off between the modulation levels used and the transmission time. Decreasing the modulation level implies reducing the number of bits in each symbol. This requires transmitting more symbols hence increasing transmission time. The energy consumed by the radio of a wireless device is made up of two components: Transmission energy. Depends on: Transmitter-receiver distance, Channel gain, Atmospheric noise. Electric circuit energy consumption. Depends on the circuit design of the transmitting and receiving device. A linear function of the time the transmitter and receiver circuit need to be on. 7
Transmission energy consumption increases with the distance and usually dominates circuitry energy consumption. At lower transmitter-receiver distance, continuously reducing the radio modulation level may lead to increased energy consumption. 8
Two reasons Deadline constraints, The lowest does not necessarily gives the lowest energy consumption (as shown in previous slides). 9
10
11
Predict possible energy harvest Maximum possible battery reserves Latency Battery levels Harvested energy Compute the feasibility of the current system Calculate the frequency and modulation levels Analyze Plan Monitor Execute Distribute this information to the sensors in the network 12
The first work to my best knowledge that mentions practical aspects of DMS and DVS in WSNs. Important because it shows that DMS and DVS competes for the available idle times and joint scheduling is not a trivial problem. Their solution is: Offline however can easily be converted to an online approach. Monitor: The communication and computation patterns of the nodes in the networks is logged for each. Analyze: Based on the logged information, analyze the patterns. Predict which computation and communication tasks Plan: Based on the prediction, compute the best frequency and modulation levels. Execute: Not a good execution plan specified. IMPORTANT: This paper has shown that DMS gives way more energy savings compared to DVS in WSNs. 13
Very well written paper. Mathematically very strong!! Analyzes a scenario where terrorists attack to water supplies that runs to our homes!! Assumes harvested energy from the flow of the water. In addition to previous paper, this needs to Monitor the latency of the system and battery levels of the nodes. Analyze Minimum and maximum possible battery reserves. Possible energy that can be harvested. 14
First paper to assume probabilistic workloads rather than deterministic. Aim the joint scheduling of DMS and DVS per GTS. Introduces speed scheduling concept for DMS: Planning phase is different: Start with lower levels and speed up as necessary if the deadline is getting closer. 15
Got the best paper award in ICESS 15. Assumes deterministic workloads. Includes energy harvesting. Aims to provide super-frame wide optimization of modulation levels. 16
17
Hopefully, will be published in EWSN 16 (under the shepherding process) DMS under probabilistic workloads. Minimize the energy consumption super-frame wide. 18
An energy efficient monitoring technology. 19
20
WSN are autonomic by nature. Application of DMS and DVS are not straight forward. Monitor: network conditions, other sensors communications, energy levels Analyze: Predict the harvested energy or the future workloads, and analyze the feasibility of the constraints. Plan: Compute the frequency and modulation levels. Execute: Distribute these results 21
[1] [2] [3] [4] 22
23