3.3 Energy Efficient Network Layer
3.3.2 QoS-oriented Energy Aware Routing
With the increasing popularity of multimedia applications, the high requirements on quality of service provided to users are another challenge faced by routing protocols, espe- cially energy aware solutions which normally compromise the delivery performance.
Real-time Power-Aware Routing (RPAR) [158] is a protocol that guarantees the min- imum required quality of service levels while decreasing energy consumption. Neighbour discovery in RPAR is not initiated until a packet transmission is required and there is no establishing route in the routing table. Therefore it achieves energy saving as periodical beacon is avoided and only neighbours that meet certain requirements such as high veloc- ity are asked to reply to join routing table. It also dynamically adopts the transmission power using multiplicative increase and linear decrease to achieve maximum energy effi- ciency. When the required velocity of a packet is met, RPAR will decrease transmission power to improve energy efficiency.
Both energy and QoS constraints are evaluated in the Sequential Assignment Rout- ing (SAR) [157] as part of the route selection process. SAR was designed for Wireless Sensor Networks (WSN) with low mobility. The main idea is to establish trees at a node which contains multiple paths to other hosts with the quality of service and energy con- straint maintained. Each path is evaluated in response of the maximum number of packet forwarding before battery depletion, packet delay, and packet priority and the results are compared in order to select the optimal route.
Bandwidth allocation efficiency has been exploited in many energy aware routing pro- tocols such as [159] [160]. In [159], hosts are required to send bandwidth availability query to neighbours and estimate the required bandwidth for effective packet transmission.[160] attempts to address the problem of varied bandwidth in heterogeneous networks. It adopts cross-layer solution where application layer provides information of device mobility pat- tern and estimation of next access point to connect, and physical layer assists detection of overlapped networks and possible hand-over decision. Transport layer creates two connec- tions and sends dummy packets for the purpose of estimating the bandwidth of the new
connection.
Besides bandwidth, other metrics are used to perform routing such as route quality. For example, Energy Efficient and Reliable Routing Scheme (EARS) [163] adopts cooper- ation between network layer with MAC layer, which gathers information of data rate and frame error rate, to choose the path with better quality in order to provide better quality in terms of reliability and energy efficiency.
Energy and Delay Efficient State Dependent Routing (EDEAR) [161] adapts its routing table to the network environment in order to satisfy QoS requirements. It applies reinforcement learning which collects information of environment parameters such as en- ergy consumption of each link, remaining energy level of hosts, based on which the cost of each path and packet delay is calculated. Reinforcement learning enables adaptive routing and uses explore agent to learn from past experience in order to adjust the optimal path constantly.
Delay Guaranteed Routing and MAC (DGRAM) [162], as a TDMA-base solution, reuses slots to decrease packet delay. Hosts are required to be uniformly distributed and location information is exchange within networks to build a topology tree. The topology is established with different tiers so that packets are guaranteed to be transmitted from the inner tier to outer tier towards their destinations. According to mathematical analysis, DGRAM is supposes to provide delay guaranteed routing with energy efficiency.
Instead of traditional address-centric routing where routing is performed in an end-to- end manner according to IP address, some studies allow data-centric routing where routing is done based on the content of data packets. In Directed Diffusion [165], pairs of attribute- value are treated as basic units of data and hosts spread their interests for certain kind of data. Events as reply to the queries are generated and flow towards the originator through multiple paths, one or a small number of which are reinforced by the network. Directed Diffusion introduces the novel feature of in-network data aggregation and caching which means data is cached in intermediate nodes for further queries so that the amount of end-to- end traffic is decreased. Sensor Protocols for Information via Negotiation (SPIN) [164]
Figure 3.7 The implosion problem. A broadcasts data to B and C, and two copies data is sent from B and C to D eventually
is another data-centric protocol which allows broadcasting of advertising data before real data transmission to solve the implosion problem and the overlap problem. Implosion, as illustration in Figure 3.7 is a common issue in flooding algorithms and refers to the redun- dant data receiving caused by a node sending data to its neighbours regardless of whether or not the neighbour has already received the same copy of data from other sources. Over- lap happens when two nodes cover an overlapped geographical area and send overlapping data to the same neighbours. Figure 3.8 illustrates what happens when two nodes send overlapped data to the same recipient. Both problems of implosion and overlap waste en- ergy and bandwidth. Under these circumstances, SPIN instructs sensors to broadcast data advertisements before the actual data transmission, allowing neighbor hosts to check if the advertised data has been requested in order to conserve energy by avoiding unnecessary transmissions. It further saves energy by letting nodes check energy availability before participating in data transmission and only hosts with enough energy are able to request data.
The idea of Content Delivery Network (CDN) is derived from replicated server sys- tems and cooperative caching/streaming systems due to increasing request of multimedia content. It consists of distributed servers caching parts of data. Queries are made to the clients nearby CDN server and the server check if it has the content. Content is sent back
Figure 3.8 The overlap problem. C receives data from A containing information about area g and d, and C received data from B containing information about area f and d, two duplicate information about d are transmitted
to client if it is available or content routing is performed to locate and deliver the content in the network. Hierarchical Content Routing [167] is a CDN-based protocol and cate- gorizes servers containing the same content into clusters to form a hierarchical structure. Both intra-cluster-wise and inter-cluster-wise routing are performed for content routing. For intra-cluster routing, each router maintains a hashing table for content cached by other servers to locate the nearby server when a piece of data content is required. For inter- cluster based routing, requests are made by cluster head to neighbour clusters for data needed. Wide area content based routing [168] uses tags to save energy for content- based routing. It constructs virtual content network between clients and servers. The first query from the client is tagged and all the following packets within the same traffic flow are then assigned the same tag. This tag is used for routing purpose so energy efficiency is achieved by saving the efforts of routing of packets within the same flow. Miura et al. [166] applies probabilistic server selection in case client requests concentrate at a server. It further reduces response time by measuring the RTT information at a router. The shorter RTT reflects quicker response as server latency and network delay are dominant elements of user response time.
Figure 3.9 State diagram of WNIC