Lower Layers PART 3: RPL layer in Contiki with the Z1

Transcription

1 Slide ower ayers PRT 3: RP layer in ontiki with the Z Kris Steenhaut - Jacques Tiberghien Remark: all numerical data refer to the parameters defined in for 32. Kbytes/s transmission in the 2.4 z frequency band.

2 Slide 2 Routing in Wireless Sensor and ctuator etworks lmost all WSs use multi-hop communication, thus requiring routing of messages through the network. Power grows at least as the square of the distance, but more often as the 4 th power. d P ~ (2d) 2 = 4d 2 d P ~ d 2 + d 2 = 2d 2 2

3 Slide 3 Routing in WSs The nternet ateway (sink node) Sensors Radio links ommunication needs:. ollect : 2P From sensors to gateway To collect field data 2. istribute : P2 From gateway to sensors To download commands, parameters and software. 3. Sensor to Sensor : P2P From one sensor to an other To run distributed data processing algorithms. The collect function is by far the most important, as most of the Wireless Sensor etworks are mainly intended for collecting data from the field. Power economy is often essential for extending lifetime and/or reducing maintenance costs. atency can be an issue when the sensors can generate alarms. n most Wireless Sensor etworks the distribute protocol is only used for (exceptional) maintenance functions. Sensor to sensor communication is gaining importance because, as nodes become more powerful, local data pre-processing and data clustering can reduce the volume of data to send to the gateway. n Wireless Sensor and ctuator networks, the istribute function is used to give commands to local actuators. atency and reliability can be important issues. Some references: Philip evis, ric Brewer, avid uller, avid ay, Sam adden, eil Patel, Joe Polastre, Scott Shenker, Robert Szewczyk, lec Woo: The mergence of a etworking Primitive in Wireless Sensor etworks, ommunications of the, Volume, umber 7, July 2008.

4 Slide 4 ontiki ommunication stacks onceptual view Rime UP6 2P P2P + BRO 2P / P2 P2P Services TP/UP OT P+RP 6OWP 6 bit addresses 64 bit addresses R Physical ayer The et subsystem of ontiki implements the communications facilities. This slide shows the simplified layered structure. Two different communication stacks are available: rime and UP6. The rime stack offers simple, ontiki specific, communication facilities, including single-hop unicast (P2P) or broadcast (BRO) and multi-hop multipoint to point (2P) functionality. The UP6 stack offers, multi-hop multipoint to point (2P), point to multipoint (P2) and point to point (P2P) functionality via the RP routing protocol. By bypassing the transport and network layers, single hop unicast (P2P) and broadcast (BRO) functionality is also available. The, R and physical layers are common to both stacks, but their code contains many if(pv6) and if(!pv6) statements. The RP routing protocol uses additional information about links gathered by the and R layers. The most common example of this is the stimated number of Transmissions needed to cross a specific link which is commonly used as a cost function for links.

5 Slide Pv6 on ow Power and ossy etworks? pplications TP/UP Pv6 &PY pplications The nternet B TP/UP RP 6owPan F J K The ow Power network is just one subnet The sink node plays the role of a router t is attractive to integrate low power and lossy networks in the internet, as it makes their access from anywhere and the development of applications much simpler. s these networks will often have a large number of nodes, only the Pv6 internet protocol has provisions for enough addresses. owever, as low power networks have limitations, such as very limited memory capacity in the nodes, that other parts of the internet don t have, their integration will need a specific architecture and specific protocols. The low power network will be considered as a specific subnet, connected to the internet through its sink node(s), which behave(s) like routers between the subnet and the internet. Reference : Jeongil Ko, ndreas Terzis, Stephen awson-aggerty, avid. uller, Jonathan W. ui and Philip evis: onnecting ow-power and ossy etworks to the nternet, ommunications agazine, pril 20, pp.96-0.

6 Slide 6 TP/UP RP 6owPan owPan Why? Pv6 address : T res S 64 T Frame length : Pv6: <= 280 bytes; : <= 8 bytes. Frame fragmentation & defragmentation eader length : Pv6 header >= 40 bytes; TP header = 20 bytes, o version field, no flow label, One subnet : 64 instead 28 bit addresses, eader compression onnectivity inside the subnet : Single broadcast domain vs. multi-hop connectivity Route-over architecture (see next slide) 6 s specific protocols for low power and lossy networks using Pv6, there is first an adaptation layer between the network and the layer. This layer is needed for several reasons: - first there is the length of the frames: max 280 bytes according to Pv6, while in some configurations, frames can be as short as 8 bytes. 6owPan has to provide facilities to fragment and rebuild frames. - Second, there is the size of the Pv6 headers. n Pv6 header, without extensions requires 40 bytes. f one adds a TP header (20 bytes) to that, there are, in the shortest frames just 2 bytes left for data. Fortunately, in the context of low power and lossy networks, lots of redundant or useless information can be removed from the headers. The version field is not needed as 6owPan uses, by definition, v6. The flow label is also useless as no connection oriented, high-throughput data flows will be established in a low power and lossy network. s the entire low power network is considered as a single Pv6 subnet, the 64 most significant bits of the Pv6 addresses, which identify the subnet, are not needed for addressing inside the subnet. - Third, an Pv6 subnet generally constitutes a single broadcast domain, while a low power and lossy network offers, most often, multi-hop connectivity, with the need for routing frames through the subnet. The RP routing protocol has been proposed for this purpose and modified versions of the Pv6 eighbor iscovery and configuration protocols can be used with RP to integrate a low power and lossy network in an Pv6 internet.

7 Slide 7 Routing Protocol for ow Power and ossy etworks RP origin: The nternet of things. TF working group for Pv6 routing. one of the existing protocols is adequate! ew design combining interesting ideas of all. RP Basic echanisms Routing based upon one or more estination Oriented irected cyclic raphs (Os) Optimal routes between sink and all other nodes for both the collect and distribute data traffics. Redundant equivalent routes are kept for reliability in case of link or node failure. ultiple Os if different optimisation criteria. 7 Some references: nternet ngineering Task Force (TF), Request for omments: 60 - RP: Pv6 Routing Protocol for ow-power and ossy etworks, SS: , arch 202. nternet ngineering Task Force (TF), Request for omments: 6 - Routing etrics Used for Path alculation in ow-power and ossy etworks, SS: , arch 202. nternet ngineering Task Force (TF), Request for omments: 62 - Objective Function Zero for the Routing Protocol for ow-power and ossy etworks (RP), , arch 202.

8 Floating Slide 8 RP: the O floating has no connection with the sink..g : failure of link -B B rounded O O = estination Oriented, irected, cyclic raph.. J F K grounded O contains the sink node and offers routes from all its nodes to that sink node. 8 Routing in the RP protocol is based upon the O concept. t is a distributed data structure build-up and maintained during the WS usage. t is a kind of distributed routing table, build by the RP algorithm and used for routing frames through the network. O = estination Oriented irected cyclic raph estination Oriented = ll routes end in a Root irected: one way edges cyclic: no loops odes and links map into to those of the physical network Root = ode with no outgoing links Roots interconnect the low power and lossy network with external networks such as the nternet. O roots correspond to the sink nodes traditionally used in the WS literature.

9 Slide 9 O: neighbors and parents The neighbors of a node are all the nodes that can be reached via a singlehop radio link. The neighbors of are,, and rounded O The parents of a node are all the neighbors that are on a possible route from the node to the sink. Parents of are, and. The preferred parent of a node is the neighbor that is on the best route from the node to the sink. The preferred parent of is (will be explained in a later slide). 9 To understand the building of a O the concepts of neighbor, parent and preferred parent need to be introduced: - n a WS nodes communicate by radio. The neighbors of a node are all the nodes that can communicate via single-hop radio links with that node. (To simplify this introduction, all radio links are supposed symmetric). - O defines routes from all nodes to the root. The parents of a node are the neighbors of that node that belong to such a route. When a node has to send a message to the route, it suffices that it sends it to one of its parents. - The quality of a route can be defined according to criteria dependent on the application ( Routing etric ). node should select among its parents the one that ensures the best route to the root, according to the selected criterion. This parent is the preferred parent. - node will always use its preferred parent to send messages to the route, unless the preferred parent is unavailable. n this case the node can use its other parents as backup.

10 Slide 0 O: the Routing etric ach link and each node has an associated cost, defined in terms of the specific Routing etric. n this example nodes have no cost. Of course, no loops can be allowed in the routing scheme. The situation shown here could occur if suddenly the cost of the link increases ach node computes the cost of the best route to the root by adding to the cost of the preferred parent the cost to reach that preferred parent > For selecting the best route to the sink a Routing etric is defined. This metric is similar to the distance parameters used in algorithms like istance Vector. t gives the cost for reaching the sink from a specific node. The routing metric normally takes into account, depending on the application, not only the cost of traveling along a link, but also some node specific dynamic parameters such as the state of the battery. When a node has to select between several parents its preferred parent, that selection should minimize the routing metric. Of course, no loops may exist in the routes towards the root, because they would make the root unreachable from certain nodes.

11 Slide O: the RK of nodes ll nodes have a rank, which is some measure of the distance between node and root. n this example it is simply the hop-count. To prevent the creation of routing loops, all parents of a node must have a rank lower than the rank of that node. When a node joins the O, its rank is computed from the data provided in the O message. owever a node is authorized to increase its rank to expand the set of its parents. n this example has increased its rank from 2 to 3 in order to add to its parents. / 4/4 2 / 3 0/0 2/2 /2 2/3 To avoid creation of loops in the O, the notion of rank has been introduced. n a O, all nodes have a rank. This rank expresses in some way the distance from the node to the sink. This distance could be a simple hop-count or it could be a much more complex function, such as the average number of times a message needs to be retransmitted by radio to reach the root. The rank of a node is computed by an Objective Function (OF, not defined in the standard, because it should be specific for the application), that takes into account the properties of the nodes and the links along the route between the node and the root. The rank of a node should always be higher than the rank of its parents (this prevents the creation of loops) n older versions of the proposal nodes with the same rank were called siblings. node can voluntarily increase its rank in order to increase the set of its parents. n these introductory slides, rank is considered equal to the hop count to the root.

12 Slide 2 RP : Building the O () The root initiates the building of a new O by sending O messages to all its neighbors. odes, and join the O. and set their rank =. / 0/0 2 / / 4/4 3 2/2 2/3 2 Building a new O is initiated by the root node of that O. t assigns a unique O identity to the new O and sends O nformation Object (O) messages to all its neighbors. ach O messages carries the identity of the new O, the rank of its sender (0 by definition for the root), the cost to reach the root from its sender (0 by definition for the root itself) as well as data about the link between its sender and its destination. The nodes receiving these O messages join the new O. Their rank is, with the simple hop-count Objective Function used for this example, as they are at one hop from the root.

13 Slide 3 RP : Building the O (2) odes, and have only one parent,. t is automatically their preferred parent. odes, and set their cost function / 0/0 2 / /2 3 3 The neighbors of the root have each one parent, the root, and their cost to the root is determined by the cost of the link from the root. Obviously the root is also their preferred parent.

14 Slide 4 RP : Building the O (3) ode ignores the O messages because they come from neighbors with a higher rank. odes, and start sending O messages to their neighbors odes, and join the O. and set their rank. / 2/ 2 3 0/0 / 2/ /2 2/ 4 fter receiving O messages from the root and updating their state, nodes, and send O messages to their neighbors: sends a message to and to, to, and while sends one to and. ignores all messages it receives because they come from nodes with a rank higher than the rank of their destination. fter receiving O messages nodes, and join the new O and set their rank value to 2.

15 Slide RP : Building the O (4) odes, and set up their list of parents and select a preferred parent odes, and update their cost figures / 2/4 3 0/0 2 / 2/2 /2 2/3 t this point the parents of are and, with as preferred parent. and each have only one parent, respectively and, wich become automatically preferred parents., and can now update their cost figures by adding to the cost of their preferred parent the cost of the link towards that preferred parent.

16 Slide 6 RP : Building the O () odes, and ignore the O messages from, and, because they come from nodes with a higher rank odes, and send O messages to all their neighbors odes, and join the O and set their rank / 2/4 3/ 3 0/0 2 / 2/2 3/ /2 2/3 3/ 6 odes, and now start sending O messages to their neighbors. sends one to,, and. sends one to,, and. just sends one to and., and ignore the O messages they receive because the come from a node with a higher rank., and join the O and set their rank to 3.

17 Slide 7 RP : Building the O (6) odes discovers that using neighbor as a preferred parent would result in a better cost figure, but has the same rank as and therefore can not be a parent of. ode decides to increment its rank to be allowed to adopt as a parent. now has to send O messages to its neighbors, to inform them of the change in rank. / 4/ 3 0/0 2 / 2/2 3/ /2 2/3 3/ 7 can not become a parent of, nor a parent of because and have the same rank but discovers that the cost via (3) is lower than the cost via its preferred parent (4). Therefor increments its rank by one unit. ow can become a parent of and it even beciomes its preferred parent. This change in its rank obliges to send again O messages to its neighbors,, and., and ignore these messages because they come from a node with a higher rank. will increment its own rank to 4.

18 Slide 8 RP : Building the O (7) ow nodes, and can select their parents and their preferred parent. / 4/4 3 0/0 2 / 2/2 /2 2/3 8 t this stage nodes, and can set up their parents list and select their preferred parents. This completes the building of the O as now all nodes are included.

19 Floating O O Slide 9 RP: essages for routing rounded O B S F J K 9 O (estination nformation Object) Transmitted downward by root and nodes in the O for discovery, construction and maintenance. O (estination dvertisement Object) Propagates upwards destination information along the O to populate the routing tables of ancestor nodes in support of the distribute traffic. S (O nformation Solicitation) iscovers Os in the neighborhood for grounding a floating O. S message is answered by a O

20 Slide 20 RP: ollect, istribute and P2P routing ollect (2P) istribute (P2) P2P 20 The collect protocol is used to collect data from all the nodes and to send it to the sink node (). The routing simply consist in following the O. The distribute protocol is just the opposite: data from the sink is sent to specific nodes. When connected to the O, each node has sent to the sink a O message containing the name of its starting point. This message is augmented in each traversed node with the information necessary to retrace the itinerary in the opposite direction. These messages are kept in the sink and are used to establish the route of messages to be distributed. n the Peer to Peer mode, a message is first sent to the sink, and from there to its destination. f the nodes have enough memory to keep locally the information contained in the O messages, there is no need to go by the sink if the collect and distribute itineraries have common links.

21 Slide 2 RP : aintaining the O Basic mechanism: ach node sends O messages to its neighbors When they receive O messages from their parents When their local trickle timer expires These O messages inform children bout nodes leaving the O (rank = ) bout rank and/or cost changes These O messages can cause O repairs The trickle timer is a timer with variable period. Whenever a change occurs the period decreases f no changes the period increases 2 Os are maintained by means of O messages. These messages, issued by the root or by any node inform the children of each node of any changes that have occurred between that node and the root. This can be simple variations in the value of the cost function, it can be more important changes that cause a node to change its rank, possibly forcing other nodes to also change their rank (children need to have a rank higher than their parents). When a node wants to leave a O (for instance when its battery gets low), it broadcasts an infinite rank, forcing all children to choose other parents. The trickle timers ensure that the O structure is periodically checked by means of O messages. f links and nodes are stable the period of the trickle timers increases progressively up to a maximum value. When changes occur, that period returns to its minimum value, to ensure fast adaptations of the O to new conditions.

22 Slide 22 The TX cost function TX = stimated number of transmissions from x to y. = umber of times a frame needs to be transmitted over a link before being acknowledged. = (d xy x d yx ) - d xy = success rate from x to y. ach node broadcasts precisely once per second a short message, each receiver counts during 0 seconds the number of broadcasts it receives from each neighbor. d xy = (number of frames received from x by y) / 0 n the ontikimac R protocol, the number of times a frame is retransmitted is counted and averaged. This is used as TX. 22 The TX cost function has been introduced in 2003 and is considered a good cost function for multi-hop radio networks. t can easily be measured during network operation, it gives a realistic estimation of the amount of energy used for transmission. Reference: ouglas S. J. e outo, aniel guayo, John Bicket and Robert orris: igh- Throughput Path etric for ulti-op Wireless Routing, in the proceedings of obiom 03, September 4 9, 2003, San iego, alifornia, US.

23 Slide 23 ac ayer optimization for RP Synchronize receiver wake-ups to minimize latency along specific routes. Wake-up = t0 - δt Wake-up = t0-3δt Wake-up = t0 4δt / 4/4 3 0/0 2 / 2/2 /2 2/3 Wake-up = t0-2δt Wake-up = t0-3δt 23 f mostly collect or distribute traffic exists, most messages follow a predictable path determined by a O. This could be used to minimize message latency in a network using Radio uty ycling to optimize power usage. The protocol proposes to schedule the wake up periods of receivers along the paths taken by messages, so that on a multi-hop route, waiting times for receiver wake-ups should be minimized. The ontikimac protocol could be modified to schedule receiver wake-ups along O routes to minimize wake up waiting. For collect traffic, the wake-up times should follow the ranks of the nodes. Reference: ang u, Bhaskar Krishnamachari, auligi S. Raghavendra: n daptive nergy-fficient andow-atency for ata athering in Wireless Sensor etworks, Proceedings of the 8th nternational Parallel and istributed Processing Symposium (PPS 04), 2004.