Design of Wireless Sensor Networks

Transcription

1 Energy-Efficiency Efficiency Issues in Cross-Layer Design of Wireless Sensor Networks Sergio Palazzo Sensor Networks: : a small world Main Features small in size small transmission ranges low-powered cheap cost-effective Energy Efficiency is a major issue! Design Guidelines (1/3) Production Costs: considering that sensor networks are composed of a large number of sensor devices, reducing the cost of a single node is very important Hardware Constraints: a sensor node is made up of four basic components Sensing Unit: it performes SENSING UNIT sensing tasks and analog to digital conversion Sensor ADC Processing Unit: it is composed of a processor and of a small storage unit Transceiver Unit: it is a radio frequency modem that uses more than one coding technique to connect the node to the network Power Unit: it may be supported by solar cells GPS Unit: optionally a GPS unit may be required PROCESSING UNIT Processor Storage POWER UNIT TRANSCEIVER UNIT POWER GENERATION Design Guidelines (2/3) Network Topology: changes in sensor network s topology can happen during pre-deployment and deployment phase, post-deployment phase and re-deployment phase. They are caused by misfunctioning, low available energy, etc, Then,, in a network with a high number of nodes that can fail, it is necessary to implement special routing protocols that re-route route packets and re-organize the network Transmission Media: wireless connection between sensors can be performed through radio links infrared links optical links Much of the current hardware for sensor nodes is based upon RF circuit design. Infrared mode is license-free and robust to interference from electrical devices. Optical transmission is under development. Nonetheless, the transmitted information must be protected through robust coding and modulation schemes Design Guidelines (3/3) Operating Environment: a biologically or chemically contaminated field,, the bottom of an ocean,, a river moving,, fast moving vehicles, animals body Power Consumption: sensors lifetime is strongly dependent on battery lifetime.. A wireless sensor node is typically equipped with a limited power source (<.5 Ah, 1.2 V) to maintain low the size and weight. Power conservation and power management are of primary importance: researchers are currently focusing on the design of power-aware protocols and algorithms for sensor networks Scalability: nodes density depends on the particular application (the number of sensor nodes can vary from hundreds to thousands devices Fault tolerance and Reliability: the fault tolerance level depends on the application. Infact, if the environment, where the sensor nodes are deployed, is subject to interference or is hostile, then the protocols must be implemented with a high fault tolerance Energy-Efficiency Efficiency related issues Application requirements Sensing accuracy Reliability Coverage Wireless Technology Network Connectivity MAC Protocols Routing and Forwarding schemes Positioning and Localization Algorithms Congestion control Data aggregation In-network network Processing 1

2 Energy Conserving in the layered framework In principle, energy conserving strategies can be effectively introduced at all individual layers of the traditional layered architecture framework At the physical layer energy can be saved by proper selection of batteries,, hardware (for( example,, power amplifiers), antennas (diversity, beamsteering,, MIMO techniques), coding (turbocodes,, turbo trellis coded modulation), transmission power control (waterfilling( strategies), multiuser detection. In the following, only strategies that coarsely refer to MAC and network layers will be considered. Energy Conserving Approaches To meet energy conserving requirements in ad hoc networks, three approaches have been mostly used so far: 1. Power save protocols attack the problem of high idle state energy consumption by maximizing the amount of time nodes spend in the sleep state 2. Power control techniques increase network capacity and reduce energy consumption by allowing nodes to determine the minimum transmit power level required to maintain network connectivity and forward traffic with least energy cost 3. Maximum lifetime routing selects paths that maximize network lifetime by balancing energy consumption across the nodes of the network Operating Modes of a Sensor Node (1/2) The energy consumed by an interface depends on its operating mode Sleep Mode: an interface can neither transmit nor receive very low energy consumption Idle Mode: an interface can transmit or receive data at any time (this requires time and energy) it consumes more energy than it does in the sleep state Receive Mode and Transmit Mode: the energy consumption is of the same order of magnitude than idle state. Transmitting requires more energy than receiving, but the difference is generally less than a factor of two Pow er (m W) TX Operating Modes of a Sensor Node (2/2) Power consumption of node subsystems Sensing Unit Processing Unit TX RX IDLE SLEEP E E E >> E RX IDLE SLEEP Transceiver Unit Power-Save Switch-off Approaches MAC layer Power-Save mechanisms To save energy the nodes must spend more time in the sleep mode The unavailability of sleeping nodes may interrupt the flow of traffic through a multihop ad hoc network Power-save protocols have been introduced at MAC and network layers The nodes periodically wake up to listen the announcements of pending traffic and, if it is necessary, they remain awake to receive and d exchange traffic It is important to determine time windows to maximize energy saving while minimizing impact on throughput and latency In this approach the nodes must also maintain a globally synchronized nized sleep-wakeup cycle Practical Problem to establish the phase synchronization in a dynamic multi-hop ad hoc network 2

3 Signalling-based Power-Save Protocols MACA-based protocols exploit the media access control process to find intervals during which the network interface does not need to be awake While a packet is being transmitted, nearby nodes, whose transmissions might interfere with the ongoing transmission, must remain silent Therefore these nodes can sleep with little or no impact on throughput PAMAS PAMAS (Power Aware Multi-Access with Signalling for Ad Hoc Networks) uses an RTS/CTS mechanism as the IEEE MAC It relies on a separate control signaling channel (busy tone) A node that is going to transmit or is in the process of receiving a transmission causes other nodes to go to sleep by generating a busy tone When a sleeping node wakes up, it has no information about the state of channel, then it transmits a sequence of probe messages and awaits a response on the control channel Power-Save Based on Network Topology (1/2) A subset of nodes that are topologically representative of the full f network is selected This covering set must be chosen so as to maintain the effective capacity of the network and minimize the impact of the power save protocol on o throughput and latency A few selected nodes in the covering set remain in the idle state e and are responsible to forward traffic in the network Other nodes spend most of their time sleeping, consuming much less energy. They wake up periodically to participate in subset election or to receive pending traffic in the sleep state The rotation between the two roles among nodes in the network is necessary to maximize the network lifetime Power-Save Based on Network Topology (2/2) Selecting the optimal covering set is a non-trivial problem, especially when it is based on localized computation with minimal overhead and when it must be recomputed in response to nodes failure and mobility The protocols that use this approach are inherently asynchronous,, but they may rely on synchronized mechanisms for buffering traffic for f sleeping nodes. In the asynchronous protocols Connectivity information can be used to determine the covering set: s the nodes that are not currently part of the covering set must exchange traffic to determine their connectivity. It is also possible to select a covering set indirectly, for example, by using position information Topology-Based Power-Save Protocols: Dominating Sets (1/3) In a cluster-based algorithm - named CEDAR - each node selects a core-node as dominator, which will give access to the services offered by the core Each core node has a domain of non core- nodes A subset of nodes,, S V, is a dominating set for V if each node in V is included in S or it is distant 1 hop from S. The dominating set with the minimum number of nodes is called minimum dominating set (MDS) Topology-Based Power-Save Protocols: Dominating Sets (2/3) Heavy computation is required to find a minimal-size dominating set: to this purpose distributed algorithms are necessary Most of these algorithms use a two phase approach In the first phase nodes exchange neighbor information and any node that has two unconnected neighbors marks itself as part of the connected dominating set The second phase eliminates any marked node that is redundant (its one-hop neighborhood is contained within the one-hop neighborhood of an adiacent marked node or is contained within the union of two such one-hop neighborhoods) 3

4 Topology-Based Power-Save Protocols: Dominating Sets (3/3) Whenever possible,, the node with the lower energy supplies is preferentially removed from the dominating set: this allows to increase network lifetime The incremental cost of a dominating node is specified as a function of the routing overhead (proportional to the number of nodes) ) and the forwarding overhead (inversely proportional to the number of forwarders): the relative magnitude of these factors depends on the average node degree, path length, path lifetime (i.e. mobility) ) and the relative costs of transmitting and receiving PERFORMANCE Simulation results show that after the elimination phase 3-4% of the nodes are in the dominating set and the network lifetime is longer When no using power-save protocols, for all values of routing overhead and for all mobility levels,, network lifetime decreases asymptotically as the node density increase; instead, fixed energy consumption models yield a linear increase in network lifetime as a function of density (SPAN and GAF protocol) Topology-Based Power-Save Protocols: SPAN (1/3) The covering set in SPAN is a connected dominating set, whose nodes are called coordinators Coordinators are in the idle state. They act as low latency routing backbone nodes and buffer traffic for sleeping destinations (using mechanisms derived from IEEE 82.11) Non-coordinator nodes wake up periodically to exchange traffic with the coordinator nodes and participate in coordinator election Topology-Based Power-Save Protocols: SPAN (2/3) The coordinator election algorithm A. The nodes periodically exchange HELLO messages to discover their two-hop neighborhood B. A node marks itself as eligible coordinator if it discovers that two neighbors cannot comunicate directly or via other coordinators C. Each marked node schedules a backoff interval (it has both random and adaptive elements), ), during which it listens for announcements from other nodes D. If after this interval the node is still eligible, it sends its own coordinator announcement.. The nodes with greater connectivity and higher energy reserves announce themselves as coordinators more quickly E. After spending some time as a coordinator,, a node withdraws The rotation of the coordinator role tends to balance nodes energy reserves, even in the case of initially unequal distribution Topology-Based Power-Save Protocols: SPAN (3/3) SPAN is a synchronous power save protocol for two reasons 1. the buffering and announcements are based on the synchronous IEEE power save mechanism, also some form of asynchronous polling is a possible alternative 2. the nodes must be awake simultaneously to exchange traffic to determine their connectivity and participate in coordinator election PERFORMANCE Simulations and experimental energy measurements corroborate that SPAN provides about 5% energy saving in dense networks, with only minimal impact on throughput and packet loss Network lifetime increases roughly linearly with network density Also latency increases with node density Rotation of the roles allows a 5-1% increase of the network lifetime Topology-Based Power-Save Protocols: Geographic Adaptive Fidelity - GAF (1/2) GAF is a power save protocol that selects its representative nodes based on position (by GPS receiver, for example), rather than membership in a dominating set GAF partitions the network using a geographic grid The grid size of R/ 5, where R is the nodes transmission range, guarantees that each node in a grid square is within the transmission range of every node in each adjacent grid square All nodes in a grid square are regarded as equivalent with respect to their ability of forwarding packets One non-sleeping node in each grid square is sufficient to maintain the connectivity of the original network Topology-Based Power-Save Protocols: Geographic Adaptive Fidelity - GAF (2/2) Each node passes among three states: sleep, discovery and active Sleeping nodes in the grid periodically wake up and go to the discovery state Discovery (A): in this state a node send a State A discovery message containing its grid E positon ID and energy status, and B C listens for other discovery messages. If Sleep Active the node hears no higher ranking D State State announcements, it passes to the active state (B), otherwise it goes back to the sleep state (C) If an active node hears a discovery message from an active node of higher rank, it immediately goes to the sleep state (D). After spending some time in the active state, a node passes to the discovery state (E), allowing the active role s rotation between the nodes in the grid square 4

5 Power Control Techniques With these techniques, nodes modify their transmit power to reduce energy consumption and increase network capacity Low power transmissions reduce contention and increase network capacity, while at the same time consuming less energy This implies that a route with a larger number of low power hops may be more energy efficient than one with fewer high power hops Minimum Energy Routing Problem The problem is to minimize the energy consumed in forwarding a packet p from source to destination Minimum energy routing can exploit exponential path loss by forwarding traffic using a sequence of low power transmissions rather than a single direct transmission In a basic path loss model, received signal strength decreases exponentially with distance: therefore the minimum transmit power required to transmit from node i to node j must be Pminmin ij d α i,j with 2 α 4.. Besides, the signal to noise ratio (SNR) at the receiver must be greater than some threshold, which depends on the target bit error rate In general, because of the effects of terrain and other obstacles, s, a node can not determine the transmission power level required for nodes s to communicate given their position Topology Control Problem This problem consists of assigning per-node transmit powers that minimize the total transmit power, while still maintaining network connectivity Most strategies are focused on increasing throughput by reducing interference, with the associated reduction in energy consumption as a beneficial side effect Examples of Topology Control Strategies 1. Link state topology information is used to maintain a connected topology. When a route update indicates a link failure,, the appropriate nodes increase their transmission power (using slotted backoff) ) until the network is connected. In the absence of topology information,, each node increases its transmission power until its degree is sufficiently large, based on estimated node density 2. Each node modifies its transmission power until it has discovered d at least one neighbor in every direction with cones of angle α 2π/3: these neighbors combined wireless coverage provides connectivity (this guarantees to each node to be connected to the network). Besides,, thanks to topology of this neighbor set,, nodes can further reduce their transmission power by eliminating redundant nodes from the set 3. Given a broadcast source node,, the minimum energy broadcast problem is to select a set of re-broadcasters and transmission powers, such that the message is distributed to all nodes with minimum total energy cost. This type of strategies is not very used because minimum energy broadcast is an NP-complete problem Maximum Lifetime Routing Saving energy in the whole network is so much important as doing it at the individual nodes. This may be done balancing energy consumption across the nodes of the network Problem of Maximum Lifetime Routing three main Route Metrics used in selecting energy aware routes have been proposed alternative approaches have also been proposed Minimum Energy Routing Route Selection Metrics (1/2) This metric minimizes the total energy consumed as a packet is forwarded on a route It does not maximize network lifetime Nodes residual energy is not taken into account, therefore the nodes will suffer early failure due to their heavy forwarding load Max-Min Min Routing It selects the route which maximizes the minimum residual energy of any node on the route itself The routes selected may be longer or have higher total energy consumption than the minimum energy route This makes neither this metric succeeds in maximizing the network lifetime, though it generally performs much better than Minimum Energy Routing 5

6 Minimum Cost Routing Route Selection Metrics (2/2) It minimizes the total cost of forwarding the packet at each node, selecting the route that minimizes the sum of the link costs c ij The cost function controls the presence of a high-cost node on a route which may deflect traffic from that route In general, the cost function is increasing, so reflecting a node s increasing reluctance to forward traffic, as its residual energy decreases Alternatively, a capacity cost function incorporates both the link energy cost and the residual energy at a node: so it has ability to balance to maximize energy reserves and to minimize forwarding cost MACRO: a MAC/ROuting cross-layer protocol for WSNs MACRO is a cross-layer protocol developed at the WiNe-Lab (Wireless Network Laboratory) of the University of Catania. Unlike former solutions s (e.g. GeRaF) ) it doesn t t involve any location information exchanging. Each node only needs eds to know its own position in the coverage area. The metric to establish the t next hop is: Advantages of using different power levels Weighted Progress Factor (WPF) Set of power levels and coverage range R R S 1( R ') S 2 ( R') S ( R 3 ') D M=4 P 1 =.11W R 1 =62.5m P 2 =.176W R 2 =125m P 3 =.892W R 3 =182.5m P 4 =.2818W R 4 =25m M=3 P 1 =.35W R 1 =83.4m P 2 =.556W R 2 =166.7m P 3 =.2818W R 3 =25m M=2 P 1 =.176W R 1 =125m P 2 =.2818W R 2 =25m M=1 P 1 =.2818w R 1 =25m MACRO MAC Functionalities To select the next relay node R R triggers a competition Let R R be the winner of the competition in the set S 1 (R ) ) and G R R1 R1 its WPF If R R estimates that a higher WPF can be obtained increasing P, a new competition is triggered in the set S 2 (R ) The procedure is repeated until no better relay nodes can be found R R D S ( R 1 ') Nodes periodically switch ON and OFF to reduce energy consumption Synchronization is not needed A wake up phase is required for R R to identify the best relay node in S i (R ) To this purpose R R transmits several short WAKE UP messages for a period T Cycle Then R R sends a GO MESSAGE which triggers competition among nodes in S i (R ) A node in S i (R ) ) hearing the WAKE UP messages calculates its WPF and stays awake waiting for the GO MESSAGE T ON T Cycle T T 2T In WU ON TX WU Then, upon hearing the GO MESSAGE, a node sends randomly to R R its WPF so that R R performs the choice of the best relay in S i (R ) 6

7 MACRO Analytic Framework What is the probability that outside the coverage area obtained using P i, exists at least one node whose WPF is higher than g,, provided that G (M) i = G i? Once the probability is known, when is it worth enlarging the coverage area? Where f ) ( g G ) G ( M ) ( 1 G M i i+ i using the previous results as is the probability density function which can be evaluated Ri g Pi +1 Ri > g Pi +1 Where z is the nodes density and a(g G i ) is the area where a node B must be located in order to belong to S i+1 (R ) and have a WPF higher than g. Performance Evaluation GPSR (with STEM-B B MAC protocol) B. Karp and H.T. Hung, GPSR: Greedy Perimeter Stateless Routing for Wireless Networks. Proc. of ACM Mobicom 2. C. Schurgers,, V. Tsiatis,, G. Ganeriwal,, and M. Srivastava.. Optimizing Sensor Networks in the Energy-Latency Latency-Density Design Space. IEEE Transactions on Mobile Computing.. Vol. 1, No. 1, January-March 22. GeRaF (same MAC as MACRO) M. Zorzi and R.R. Rao.. Geographic Random Forwarding (GeRaF( GeRaF) ) for Ad Hoc and Sensor Networks: Multihop Performance. IEEE Transactions on Mobile Computing.. Vol. 2, No. 4. October-December 23. M. Zorzi and R.R. Rao.. Geographic Random Forwarding (GeRaF( GeRaF) ) for Ad Hoc and Sensor Networks: Energy and Latency Performance. IEEE Transactions on Mobile Computing.. Vol. 2, No. 4. October-December 23. Area 1 x 1 m 2 Number of nodes in the range [5,25] Four traffic sources; packet rate r in the range [.2,2]; packet size 512 bytes Duty cycle time.1 s, T ON three times Wake UP message transmission time Signal propagation model two-ray ground model (accurate in outdoor environment) Two nodes in the radio coverage if the received signal is higher r than.365e-6 6 W Four possible transmission power levels Impact of node density Impact of node density GPSR+STEM MACRO GPSR+STEM Average power consumption vs nodes nominal radio area GeRaF GeRaF MACRO GPSR+STEM MACRO GeRaF Ninra=NπR 2 /Ξ 2 Average power consumption Average number of hops 7

8 Number of packets delivered at destination Data Aggregation MACRO GeRaF MACRO GeRaF GPSR+STEM GPSR+STEM Scenario Applications: : Sensor Networks for phenomenon monitoring Sensor nodes notify the sink about the phenomenon they sense The sink uses this data to create a map of the phenomenon (spatial al propagation, temporal evolution, etc.) Data is funneled towards the sink and causes congestion Aggregation Due to sensors density and the kind of applications, redundant and highly correlated data (both( in space and time) are often propagated throughout the network. This can lead to energy waste which reduces network lifetime Mechanisms for combining data at forwarding intermediate nodes so as to propagate only useful and not redundant information should be considered Data aggregation is aimed at aggregating data coming from multiple sensors into useful non redundant fused data Data aggregation = gathering + fusion Relevant Metrics Metrics to be considered when performing aggregation are Energy efficiency: fairness in energy consumption at each aggregation round is to be searched. Network lifetime: the number of rounds until the first sensor is depleted of energy should be maximized Fidelity/distortion distortion rate: difference between the data collected at the sink when no aggregation is performed and when it is, should be evaluated Latency: aggregation implies additional processing of data. The delay introduced by these operations should be considered and should not have excessive impact on the timeliness of the system Two-phase pull mechanism: Direct Diffusion Phase 1: the sink pulls for information from sources using the propagation of interests for specific information. Sources answer using multiple paths Phase 2: the sink initiates reinforcement,, i.e. asks some specific nodes (based on considerations related to delay, quality, etc.) to increase their event sending rate Data arriving at intermediate nodes and coming from multiple sources can be aggregated when referring to the same interest Reinforcement can be incorporated into Phase 1, so the two phase process can become one phase Direct Diffusion can be costly because all nodes are required to send their data to the sink 8

9 LEACH To reduce the overhead of each node transmitting data directly to the sink, clusterhead solutions have been proposed. These allow to drastically reduce energy consumption In LEACH sensor nodes organize themsleves into clusters.. In each one, a node, denoted as clusterhead (CH), takes care of fusing data coming from other nodes in the cluster and sends the fused data to the sink. LEACH is useful for constant monitoring and periodic data reporting applications 2 phases: Set up: nodes organize themselves into clusters and CHs are randomly selected Steady state: CHs perform data aggregation Limitations: Nodes assumed to have the same energy capabilities so that random rotation of leadership can be performed Consequently all nodes are thought to be CHs and are assumed to be able to reach the sink in one hop transmission CONCERT: focus on congestion control Congestion typically occurs in nodes close to the sink and causes waste of bandwidth and energy resources Congestion leads to losses and thus causes inaccuracy in phenomenon monitoring New approach to solve the congestion problem in sensor networks using adaptive data-aggregation aggregation Previous works on congestion control used back-pressure to regulate sources transmission rate -> decrease in sources throughput and increase in signaling Data aggregation has been used for increasing network lifetime Positive effects on the distribution of the load in the network have been observed Data-aggregation aggregation used for counteracting congestion problem and guaranteing fidelity in the network Aggregation for congestion control purposes should be done while preserving information entropy To this purpose we focus on loss less aggregation which exploits spatial correlation of data We study the impact of aggregator nodes positioning on network performance in terms of loss probability and network overload We estimate the impact of data aggregation on energy consumption both at aggregator nodes and overall in the network Aggregator nodes positioning: Example 25 sink The load due to control packets forwarding for aggregator nodes control and management cannot be neglected Aggregating one hop away from the funneled nodes (i.e. node 9) performs well especially for high values of the aggregation function (i.e. ) There is not great adavntage in aggregating at all network nodes with respect to aggregating only at few nodes. It also results cheaper!,16,16 PLoss,14,12,1,8,6,4,2 Nodes P Loss,14,12,1,8,6,4,2 Node s Energy Cost Aggr=9 Aggr=3 Aggr=4 Aggr=7 Aggr=3, 9 Aggr=3, 4 Aggr=3, 7 Aggr=4, 9 Aggr=4, 7 Aggr=7, 9 Aggr=3, 4, 7 Aggr=3, 7, 9 Aggr=4, 7, 9 Aggr=3, 4, 7, 9,16,16.6,14,14,12,12 PLoss,1,8,6 PLoss,1,8,6 Cost [W].5,4,4.4,2,2.3 Nodes Nodes.2 PLoss,2,18,16,14,12,1,8,6,4,2 Nodes PLoss,2,18,16,14,12,1,8,6,4,2 Nodes.1 1 MAX α j Energy cost depending on aggregation function value and aggregator nodes positioning 9

10 Some slippery arguments and some remarks about layer integration 1. Energy is consumed not only through transmission, but through processing as well (so, for example, minimum energy routing subject to throughput constraints only can lead to select a multi-hop low-transmit transmit-power route that might be not the best from the viewpoint of the overall processing energy dissipated in too many nodes) 2. Modeling the optimization problem treating energy as a cost function and treating it as a hard constraint are two different things (the former f is appropriate in personal nomadic communications, where we may want t the batteries deliver as much functionality as possible before they are recharged or replaced, the latter is appropriate in sensor networks where the energy is not renewable at all) 3. In finite-energy energy cases, the relative importance of performance objectives being selected can be definitely argued (which strategy should be preferred between one that privileges throughput and another that privileges es network longevity?) 1. So far research efforts have mainly targeted isolated strategies,, thereby ignoring layer interdependencies, which actually exist at conceptual level and sometimes lead to antagonistic effects. 2. Example 1. The most efficient routing protocol in an ad hoc network may suggest to use a temporarily centrally-located located node to forward packets to other nodes. However, the battery of that node will be quickly exhausted. e 3. Example 2. According to the Shannon theory (the energy required to communicate one bit of information decrease as the bit time increases) energy can be conserved by transmitting a bit over a longer period of time. However, this will clearly impact the MAC protocol efficiency. 4. As a general indication, to support an adaptive cross-layer design, information should be exchanged across all layers in the protocol l stack and all strategies should be jointly optimized with respect to the energy e constraints of the global system. Nevertheless 1. the different timescales of the variations that may affect each layer should be considered anyway. 2. Example.. Variations in wireless link SINR are typically very fast (order r of microseconds), in network topology are slower (order of seconds),, in user traffic are even slower (order of tens to hundreds of seconds). So, by the time the information about weak link connectivity is relayed to a higher layer of the protocol stack (i.e., the network layer for rerouting or the application layer for reduced-rate rate compression), the link SINR will most likely have changed again. 3. Therefore, as a further indication, each layer should attempt to compensate for variation at that layer first. If the local adaptation is insufficient, the performance metrics at the next layer of the protocol stack will degrade as a result. Adaptation at this next layer will then try to mitigate the problem that could not be fixed through local adaptation, and so forth. Conclusions Energy efficient communication is a multi-faceted problem Focusing on only one aspect of the problem, or optimizing a single element of the protocol stack, can lead to sub-optimal performance with respect to maximizing node and battery lifetime It is necessary to find an appropriate tradeoff between the various approaches when applied to the same network to obtain good performance for every context Much work is still needed to understand how the various intra-layer mechanisms can effectively interact in a cross-layer approach 4. The OSI original sin still makes a sense! 1