sensor network data dissemination based on the paper titled directed diffusion: a scalable and...

30
Sensor Network Data Dissemination Sensor Network Data Dissemination based on the paper titled based on the paper titled Directed Diffusion: A Scalable and Directed Diffusion: A Scalable and Robust Communication Paradigm for Robust Communication Paradigm for Sensor Networks Sensor Networks Presented by Baha’a Al-Deen Omar Presented by Baha’a Al-Deen Omar Student I.D: 0037224 Student I.D: 0037224

Upload: marjorie-houston

Post on 18-Jan-2018

219 views

Category:

Documents


0 download

DESCRIPTION

2- Distributed Sensor Networks Advances in processors, memory and radio technology will enable small and cheap nodes capable of wireless communication and significant computation. Advances in processors, memory and radio technology will enable small and cheap nodes capable of wireless communication and significant computation. Beside the sensing ability of these nodes, distributed microsensing can be achieved- a collection of nodes can coordinate to achieve larger sensing tasks, or collaborate to disambiguate a certain sensed phenomena. Beside the sensing ability of these nodes, distributed microsensing can be achieved- a collection of nodes can coordinate to achieve larger sensing tasks, or collaborate to disambiguate a certain sensed phenomena.

TRANSCRIPT

Page 1: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

Sensor Network Data DisseminationSensor Network Data Dissemination based on the based on the paper titled paper titled

Directed Diffusion: A Scalable and Robust Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor NetworksCommunication Paradigm for Sensor Networks

Presented by Baha’a Al-Deen OmarPresented by Baha’a Al-Deen OmarStudent I.D: 0037224Student I.D: 0037224

Page 2: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

11 - -IntroductionIntroductionWhat is data dissemination?What is data dissemination? It is the delivery of data between network nodes from source to It is the delivery of data between network nodes from source to

destination.destination.

What is Sensor network?What is Sensor network? It is a set of communication nodes planted, without previous It is a set of communication nodes planted, without previous

planning, in the environments in which we need to collect data planning, in the environments in which we need to collect data and provided with various types of sensors to collect the and provided with various types of sensors to collect the needed information. It may also contains processing unit to needed information. It may also contains processing unit to process data or used only to transmit time series of the sensed process data or used only to transmit time series of the sensed phenomena to other node.phenomena to other node.

Page 3: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

22 - -Distributed Sensor NetworksDistributed Sensor Networks Advances in processors, memory and radio technology will Advances in processors, memory and radio technology will

enable small and cheap nodes capable of wireless enable small and cheap nodes capable of wireless communication and significant computation.communication and significant computation.

Beside the sensing ability of these nodes, distributed Beside the sensing ability of these nodes, distributed microsensing can be achieved- a collection of nodes can microsensing can be achieved- a collection of nodes can coordinate to achieve larger sensing tasks, or collaborate to coordinate to achieve larger sensing tasks, or collaborate to disambiguate a certain sensed phenomena.disambiguate a certain sensed phenomena.

Page 4: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

22 - -Sensor Networks-- continueSensor Networks-- continue Due to the low cost of these nodes, tens of them can be Due to the low cost of these nodes, tens of them can be

deployed near the phenomena to be sensed, in unplanned deployed near the phenomena to be sensed, in unplanned fashion.fashion.

Despite of their low cost, these nodes can provide high SNR.Despite of their low cost, these nodes can provide high SNR.

Page 5: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

22 - -Sensor Deployment CategoriesSensor Deployment Categories

Large and complex sensor networks. These are usually Large and complex sensor networks. These are usually deployed far from the phenomena to be sensed, so they must deployed far from the phenomena to be sensed, so they must afford complex signal processing algorithms to separate the afford complex signal processing algorithms to separate the environmental noise from the targets.environmental noise from the targets.

Less complicated sensor networks. The node here only Less complicated sensor networks. The node here only transmits a time series of the sensed phenomena and do not transmits a time series of the sensed phenomena and do not posses any computational capability. These time series data is posses any computational capability. These time series data is sent to one or more nodes which performs the data reduction sent to one or more nodes which performs the data reduction and filtering.and filtering.

Page 6: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

22 - -Sensor Networks-- continueSensor Networks-- continue Because sensors are battery-powered, saving battery power is Because sensors are battery-powered, saving battery power is

a very important requirement. This means that using a short-a very important requirement. This means that using a short-term communication is preferred over long-term term communication is preferred over long-term communication. The short-term communication also helps in communication. The short-term communication also helps in communicating around obstacles.communicating around obstacles.

These power consideration and other communication These power consideration and other communication capabilities lead to a different sensor network organization. In capabilities lead to a different sensor network organization. In this paper the waveform from the target in converted into this paper the waveform from the target in converted into event description, and the request for a specific data is called event description, and the request for a specific data is called an interest.an interest.

Page 7: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

33 - -Directed DiffusionDirected Diffusion Data is named using attribute-value pairs.Data is named using attribute-value pairs. A sensing task is disseminated throughout the sensor network A sensing task is disseminated throughout the sensor network

as an interest (task description) for the named data.as an interest (task description) for the named data. This dissemination sets up gradient (paths) to draw events This dissemination sets up gradient (paths) to draw events

(data matching the interest).(data matching the interest). Events start to flow towards the originators of the interest, Events start to flow towards the originators of the interest,

called sink, along multiple paths.called sink, along multiple paths. The sink then reinforce, one or smaller number of these paths.The sink then reinforce, one or smaller number of these paths.

Page 8: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.13.1 NamingNaming Task description are named using a list of attribute-value pairs Task description are named using a list of attribute-value pairs

that describe a task.that describe a task.

The following is a request to find a four-legged animal in a The following is a request to find a four-legged animal in a certain region for the next 10s for every 10ms.certain region for the next 10s for every 10ms.

Page 9: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.13.1 Naming--continueNaming--continue When the node receive this interest, it uses it’s sensors to When the node receive this interest, it uses it’s sensors to

collect the desired information and are sent back to the sink collect the desired information and are sent back to the sink node.node.

Page 10: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.13.1 Naming--continueNaming--continue

It is obvious that selecting a naming scheme is the It is obvious that selecting a naming scheme is the first step in designing such networks. Here we used first step in designing such networks. Here we used attribute-value pairs. Other naming schemes such as attribute-value pairs. Other naming schemes such as hierarchical can be used.hierarchical can be used.

The naming scheme can affect the expressivity of The naming scheme can affect the expressivity of tasks and can affect the performance of a diffusion tasks and can affect the performance of a diffusion algorithms.algorithms.

Page 11: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an GradientsInterests an Gradients The interval specifies the data rate.The interval specifies the data rate. The sink node periodically broadcast the interest message to The sink node periodically broadcast the interest message to

it’s neighbors.it’s neighbors. An initial interest is transmitted with attributes exactly the An initial interest is transmitted with attributes exactly the

original message but with larger interval (lower data rate). original message but with larger interval (lower data rate). This is supposed to be exploratory; it tries to determine if there This is supposed to be exploratory; it tries to determine if there are any sensor node that detect a four-legged animal.are any sensor node that detect a four-legged animal.

Page 12: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients-- Interests an Gradients-- continuedcontinued

The sink periodically transmits the interest with monotonically The sink periodically transmits the interest with monotonically increasing timestamp. This is necessary because interest are increasing timestamp. This is necessary because interest are nor reliably transmitted through the network.nor reliably transmitted through the network.

Every node contains an interest cache. Each item in the cache Every node contains an interest cache. Each item in the cache refers to a distinct interest (all the attribute differs for each refers to a distinct interest (all the attribute differs for each interest).interest).

Two identical interest maybe represented in a single interest Two identical interest maybe represented in a single interest entry.entry.

Page 13: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

Each interest entry has a timestamp that indicated the last Each interest entry has a timestamp that indicated the last received matched interest.received matched interest.

Contains several gradients, up to one per neighbor. Each Contains several gradients, up to one per neighbor. Each gradient contains a data rate field requested by the specified gradient contains a data rate field requested by the specified neighbor, derived from interval attribute.neighbor, derived from interval attribute.

It also contains a duration field, derived from the timestamp It also contains a duration field, derived from the timestamp and the expiresAt attributes. This field indicates the life time and the expiresAt attributes. This field indicates the life time of the interest. of the interest.

Page 14: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

What happens upon receiving an interest?What happens upon receiving an interest? 1-1- If no matching entry exists, the node creates an interest If no matching entry exists, the node creates an interest

entry. The parameters of this entry are instantiated from the entry. The parameters of this entry are instantiated from the received interest. ( neighbors must be distinguished, 802.11 received interest. ( neighbors must be distinguished, 802.11 MAC addresses, Bluetooth cluster addresses.MAC addresses, Bluetooth cluster addresses.

2-2- If there exist an entry, but no gradient for the sender, If there exist an entry, but no gradient for the sender,

the node add a gradient of the specified value. It also updates the node add a gradient of the specified value. It also updates the timestamp and duration fieldsthe timestamp and duration fields

3-3- If there exist an entry and gradient, the node only If there exist an entry and gradient, the node only updates the timestamp and duration fields.updates the timestamp and duration fields.

Page 15: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

When a gradient expires, it is removed from the entry. Not all When a gradient expires, it is removed from the entry. Not all gradients will expire at the same time. (i.e. if two different gradients will expire at the same time. (i.e. if two different sinks posts indistinct interests with different expiration time, sinks posts indistinct interests with different expiration time, some node in the network may have gradients with different some node in the network may have gradients with different expiration time.expiration time.

When all gradients for an interest is removed, the interest When all gradients for an interest is removed, the interest entry itself is removed from the cache.entry itself is removed from the cache.

Why gradients are removed and how it is done is discussed in Why gradients are removed and how it is done is discussed in 3.3.3.3.

Page 16: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

Upon receiving an interest, the node may decide to re-send it Upon receiving an interest, the node may decide to re-send it to its neighbors. The interest appears to originate from the to its neighbors. The interest appears to originate from the sending node, not the original sink. This is called local sending node, not the original sink. This is called local interaction.interaction.

Not all received interest are re-sent. A node may no re-send a Not all received interest are re-sent. A node may no re-send a received interest if it is recently re-sent a matching interest.received interest if it is recently re-sent a matching interest.

Page 17: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

How to choose a neighbor?How to choose a neighbor? 1-1- Re-broadcasting the interest to all neighbor. This is the Re-broadcasting the interest to all neighbor. This is the

only choice in the absence of information about which node only choice in the absence of information about which node will satisfy the interest.will satisfy the interest.

2-2- It is possible to use geographic routing. This can limit It is possible to use geographic routing. This can limit the scope of interest diffusion. This yields to energy saving.the scope of interest diffusion. This yields to energy saving.

3-3- In immobile sensor networks, nodes may use data In immobile sensor networks, nodes may use data cache to direct interests. (i.e. If the node heard that neighbor A cache to direct interests. (i.e. If the node heard that neighbor A sent data in response to an earlier interest, the node may sent sent data in response to an earlier interest, the node may sent the interest to A rather than broadcasting to all neighbors.the interest to A rather than broadcasting to all neighbors.

Page 18: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

NoticeNotice thatthat inin figure2(afigure2(a), ), every pair of neighborsevery pair of neighbors establishes a establishes a gradients towards each others. gradients towards each others.

How local interaction is the reason for this?How local interaction is the reason for this?When a node receives an interest from its neighbor, it doesn’t When a node receives an interest from its neighbor, it doesn’t know whether this interest was in response to an interest it know whether this interest was in response to an interest it sent out earlier or identical interest from another sink I the sent out earlier or identical interest from another sink I the other side of that neighbor. This cause each node to receive other side of that neighbor. This cause each node to receive one copy of the low data rate events from its neighbors.one copy of the low data rate events from its neighbors.

This technique can enable fast recovery from failed paths or This technique can enable fast recovery from failed paths or reinforcement of better paths. Doesn't incur persistent loops.reinforcement of better paths. Doesn't incur persistent loops.

Page 19: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.23.2 Interests an Gradients--Interests an Gradients--continuecontinue

A gradient specifies both, a data rate and direction.A gradient specifies both, a data rate and direction.

More generally a gradients specifies a value and direction. The More generally a gradients specifies a value and direction. The designer have the freedom to attach different semantics to designer have the freedom to attach different semantics to gradients values.gradients values.

Page 20: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.33.3 Data propagationData propagation

Upon receiving an interest, the node tasks its sensors to start Upon receiving an interest, the node tasks its sensors to start collecting data.collecting data.

Comparison with pre-sampled waveforms.Comparison with pre-sampled waveforms.

When detecting a target, the node searches its interest cache When detecting a target, the node searches its interest cache for a matching interest (described later). When finding one, it for a matching interest (described later). When finding one, it starts to generate event at highest data rate among all of its starts to generate event at highest data rate among all of its gradients.gradients.

Page 21: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.33.3 Data propagation--continueData propagation--continue A node that receives a data attempts to find a matching entry A node that receives a data attempts to find a matching entry

(i.e. the same (i.e. the same rectrect and the same and the same typetype of the entry). of the entry).

Finding a matching interest entry for the received data;Finding a matching interest entry for the received data; 1-1- No match exists, the data is dropped.No match exists, the data is dropped. 2-2- A match exists, the node checks the data cache A match exists, the node checks the data cache

associated with the interest. (it tracks the recently seen data associated with the interest. (it tracks the recently seen data items for several purposes, one of them is loop prevention). If items for several purposes, one of them is loop prevention). If there exist a matching data in the data cache, the data is dropped.there exist a matching data in the data cache, the data is dropped.3-3- No matching data, the data is stored in the data cache and No matching data, the data is stored in the data cache and re-sent to the node’s neighbors.re-sent to the node’s neighbors.

Page 22: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.33.3 Data propagation--continueData propagation--continue

At what rate must the received data be re-sent?At what rate must the received data be re-sent?

1-1- If all gradients data rate is greater than or equal to the If all gradients data rate is greater than or equal to the rate of received events, the received data message will be re-rate of received events, the received data message will be re-sent to appropriate neighbors.sent to appropriate neighbors.

2-2- If some gradients have lower data rate than others If some gradients have lower data rate than others (caused by reinforcement, section 3.4), then the node may (caused by reinforcement, section 3.4), then the node may downstream to appropriate neighbor.downstream to appropriate neighbor.

Page 23: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.4.13.4.1 ReinforcementReinforcement When he source detects a target, it send the data along When he source detects a target, it send the data along

multiple paths. When the sink receives this low data rate multiple paths. When the sink receives this low data rate events, it reinforces one neighbor to “draw down” higher events, it reinforces one neighbor to “draw down” higher quality events (higher data rates).quality events (higher data rates).

The reinforcements is done according “data driven” local rules The reinforcements is done according “data driven” local rules (i.e. reinforce the neighbor from which it receives previously (i.e. reinforce the neighbor from which it receives previously unseen event.unseen event.

The reinforcement is done by The reinforcement is done by re-sending the original interest re-sending the original interest but with higher a smaller interval.but with higher a smaller interval.

Page 24: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.4.13.4.1 Reinforcement--continueReinforcement--continue When the neighboring node receives this smaller-interval interest. If the When the neighboring node receives this smaller-interval interest. If the

data rate is higher than that of any existing gradients, it must reinforce at data rate is higher than that of any existing gradients, it must reinforce at least one neighbor.least one neighbor.

How this reinforcement done?How this reinforcement done?The node uses its data cache. The same local rules applies. The node may The node uses its data cache. The same local rules applies. The node may choose a neighbor from whom it received the latest event matching the choose a neighbor from whom it received the latest event matching the interest. May also reinforce all neighbor form which new events (low data interest. May also reinforce all neighbor form which new events (low data rate events) were recently received.rate events) were recently received.

By this sequence of local interactions, a path of higher data rates is By this sequence of local interactions, a path of higher data rates is established from the source the sink.established from the source the sink.

Page 25: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.4.13.4.1 Reinforcement--continueReinforcement--continue

The mechanism described above is very reactive to changes in The mechanism described above is very reactive to changes in path quality. When one path delivers data faster than others, path quality. When one path delivers data faster than others, the sink attempts to reinforce this new path, which could be the sink attempts to reinforce this new path, which could be wasteful for resources.wasteful for resources.

To solve this problem, more sophisticated local rules are To solve this problem, more sophisticated local rules are possible. Such rules may be, reinforce the node that most possible. Such rules may be, reinforce the node that most events have been received from. Also, reinforce the node the events have been received from. Also, reinforce the node the deliver events before another nodes.deliver events before another nodes.

Page 26: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.4.23.4.2 Negative ReinforcementNegative Reinforcement The algorithm used above may result in more than one path. The algorithm used above may result in more than one path.

Consider that the sink reinforces A, then received a new event Consider that the sink reinforces A, then received a new event from B, it will reinforce B. If the path through B is better, the from B, it will reinforce B. If the path through B is better, the sink needs a mechanism to negatively reinforce the path sink needs a mechanism to negatively reinforce the path through A.through A.

One of the mechanisms of negative reinforcement is to time One of the mechanisms of negative reinforcement is to time out all high data rate gradients unless they are explicitly out all high data rate gradients unless they are explicitly reinforced. This need the sink to periodically reinforce the reinforced. This need the sink to periodically reinforce the neighbor B, and cease reinforcing neighbor A. This will neighbor B, and cease reinforcing neighbor A. This will degrade the path through A. Another approach is to explicitly degrade the path through A. Another approach is to explicitly degrade the path through A by re-sending interests at lower degrade the path through A by re-sending interests at lower data rates.data rates.

Page 27: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

3.4.23.4.2 Negative Reinforcement--Negative Reinforcement--continuecontinue

When A receives this lower data rate interest, it degrades its When A receives this lower data rate interest, it degrades its gradients toward the sink. When all its gradients are now low gradients toward the sink. When all its gradients are now low data rate, A negatively reinforce those neighbors that have data rate, A negatively reinforce those neighbors that have been sending data to it at higher data rates. The local been sending data to it at higher data rates. The local interaction ensures that the path through A is degraded rapidly, interaction ensures that the path through A is degraded rapidly, but the cost is in the increase in resource utilization.but the cost is in the increase in resource utilization.

Page 28: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

44 - -PerformancePerformance

Page 29: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

44 - -Performance--continuePerformance--continue

Page 30: Sensor Network Data Dissemination based on the paper titled Directed Diffusion: A Scalable and Robust Communication Paradigm for Sensor Networks Presented

44 - -Performance--continuePerformance--continue