Ad hoc and Sensor Networks Chapter 12: Data-centric and content-based networking
Holger Karl
Computer Networks Group Universitt Paderborn
Goal of this chapter
Apart from routing protocols that use a direct identifier of nodes (either unique id or position of a node), networking can talk place based directly on content
Content can be collected from network, processed in the network, and stored in the network
This chapter looks at such content-based networking and data aggregation mechanisms
Overview
Interaction patterns and programming model Data-centric routing Data aggregation Data storage
Desirable interaction paradigm properties
Standard networking interaction paradigms: Client/server, peer-to-peer
Explicit or implicit partners, explicit cause for communication
Desirable properties for WSN (and other applications)
Decoupling in space neither sender nor receiver need to know their partner Decoupling in time answer not necessarily directly triggered by question, asynchronous communication
Interaction paradigm: Publish/subscribe
Achieved by publish/subscribe paradigm
Idea: Entities can publish data under certain names Entities can subscribe to updates of such named data
Conceptually: Implemented by a software bus
Software bus stores subscriptions, published data; names used as filters; subscribers notified when values of named data changes
Variations
Topic-based P/S inflexible Content-based P/S use general predicates over named data
Publisher 1
Publisher 2
Software bus
Subscriber 1
Subscriber 2
Subscriber 3
Publish/subscribe implementation options
Central server mostly not applicable Topic-based P/S: group communication protocols Content-based networking does not directly map to multicast groups
Needs content-based routing/forwarding for efficient networking
t < 10 t > 20
t = 35!
t > 20 t < 10 or t > 20 t > 30
t < 10 or t > 20
t < 10
t > 20
6
Overview
Interaction patterns and programming model Data-centric routing Data aggregation Data storage
One-shot interactions with big data sets
Scenario
Large amount of data are to be communicated e.g., video picture Can be succinctly summarized/described
Idea: Only exchange characterization with neighbor, ask whether it is interested in data
Only transmit data when explicitly requested Nodes should know about interests of further away nodes
! Sensor Protocol for Information via Negotiation (SPIN)
SPIN example
(1) (2) (3)
(4)
(5)
(6)
Repeated interactions
More interesting: Subscribe once, events happen multiple times
Exploring the network topology might actually pay off But: unknown which node can provide data, multiple nodes might ask for data ! How to map this onto a routing problem?
Idea: Put enough information into the network so that publications and subscriptions can be mapped onto each other
But try to avoid using unique identifiers: might not be available, might require too big a state size in intermediate nodes
! Directed diffusion as one option for implementation
Try to rely only on local interactions for implementation
10
Directed diffusion Two-phase pull
Phase 1: nodes distribute interests in certain kinds of named data
Specified as attribute-value pairs (cp. Chapter 7)
Sink 1
Sink 2 Sink 3 X
Interests are flooded in the network
Apparently obvious solution: remember from where interests came, set up a convergecast tree Problem: Node X cannot distinguish, in absence of unique identifiers, between the two situations on the right set up only one or three convergecast trees?
Source
Sink
Source
11
Direction diffusion Gradients in two-phase pull
Option 1: Node X forwarding received data to all parents in a convergecast tree
Not attractive, many needless packet repetitions over multiple routes
Option 2: node X only forwards to one parent
Not acceptable, data sinks might miss events
Option 3: Only provisionally send data to all parents, but ask data sinks to help in selecting which paths are redundant, which are needed
Information from where an interest came is called gradient Forward all published data along all existing gradients
12
Gradient reinforcement
Gradients express not only a link in a tree, but a quantified strength of relationship
Initialized to low values Strength represents also rate with which data is to be sent
Intermediate nodes forward on all gradients
Can use a data cache to suppress needless duplicates
Second phase: Nodes that contribute new data (not found in cache) should be encouraged to send more data
Sending rate is increased, the gradient is reinforced Gradient reinforcement can start from the sink If requested rate is higher than available rate, gradient reinforcement propagates towards original data sources
Adapts to changes in data sources, topology, sinks
13
Directed diffusion extensions
Two-phase pull suffers from interest flooding problems
Can be ameliorated by combining with topology control, in particular, passive clustering
Geographic scoping & directed diffusion Push diffusion few senders, many receivers
Same interface/naming concept, but different routing protocol Here: do not flood interests, but flood the (relatively few) data Interested nodes will start reinforcing the gradients
Pull diffusion many senders, few receivers
Still flood interest messages, but directly set up a real tree
14
Overview
Interaction patterns and programming model Data-centric routing Data aggregation Data storage
15
Data aggregation
Any packet not transmitted does not need energy To still transmit data, packets need to combine their data into fewer packets ! aggregation is needed Depending on network, aggregation can be useful or pointless
16
Metrics for data aggregation
Accuracy: Difference between value(s) the sink obtains from aggregated packets and from the actual value (obtained in case no aggregation/no faults occur)
Completeness: Percentage of all readings included in computing the final aggregate at the sink
Latency
Message overhead
17
How to express aggregation request?
One option: Use database abstraction of WSN Aggregation is requested by appropriate SQL clauses
Agg(expr): actual aggregation function, e.g., AVG(temperature) WHERE: filter on value before entering aggregation process
Usually evaluated locally on an observing node
GROUP BY: partition into subsets, filtered by HAVING
GROUP BY floor HAVING floor > 5
18
Partial state records
Partial state records to represent intermediate results
E.g., to compute average, sum and number of previously aggregated values is required expressed as <sum,count> Update rule: Final result is simply s/c
19
Aggregation operations categories
Duplicate sensitive, e.g., median, sum, histograms; insensitive: maximum or minimum Summary or examplary Composable: for f aggregation function, there exist g such that f(W) = g(f(W 1), f(W2)) for W = W 1 [ W2 Behavior of partial state records
Distributive end results directly as partial state record, e.g., MIN Algebraic p.s.r. has constant size; end result easily derived Content-sensitive size and structure depend on measured values (e.g., histogram) Holistic all data need to be included, e.g., median
Monotonic
20
Placement of aggregation points
Convergecast trees provide natural aggregation points But: what are good aggregation points?
Ideally: choose tree structure such that the size of the aggregated dat to be communicated is minimized Figuratively: long trunks, bushy at the leaves In fact: again a Steiner tree problem in disguise
Good aggregation tree structure can be obtained by slightly modifying Takahashi-Matsuyama heuristic Alternative: look at parent selection rule in a simple flooding-based tree construction
E.g., first inviter as parent, random inviter, nearest inviter, Result: no simple rule guarantees an optimal aggregation structure
Can be regarded as optimization problem as well
21
Alternative: broadcasting an aggregated value
Goal is to distribute an aggregate of all nodes measurements to all nodes in turn
Setting up |V| convergecast trees not appropriate
Local measurement Current Estimate New estimate from neighbor
Module A: Local Fuse
Module B: Remote Fuse
New Estimate
Idea: Use gossiping combined with aggregation
When new information is obtained, locally or from neighbor, compute new estimate by aggregation Decide whether to gossip this new estimate, detect whether a change is significant
Transmit new estimate to neighbor
Module C: Decide on propagation
Do not transmit new estimate to neighbor
22
Overview
Interaction patterns and programming model Data-centric routing Data aggregation Data storage
23
Data-centric storage
Problem: Sometimes, data has to be stored for later retrieval difficult in absence of gateway nodes/servers Question: Where/on which node to put a certain datum?
Avoid a complex directory service
Idea: Let name of data describe which node is in charge
Data name is hashed to a geographic position Node closest to this position is in charge of holding data Akin to peer-to-peer networking/distributed hash tables Hence name of one approach: Geographic Hash Tables (GHT) Use geographic routing to store/retrieve data at this location (in fact, the node)
24
Geographic hash tables Some details
Good hash function design Nodes not available at the hashed location use nearest node as determined by a geographic routing protocol
E.g., the node where an initial packet started circulating the hole Other nodes around hole are informed about node taking charge
Key location
Timeout
Handling failing and new nodes
Failure detected by timeout, apply similar procedure as for initially storing data
Limited storage per node
Distribute data to other nodes on same face
New key location
25
Conclusion
Using data names or predicates over data to describe the destination of packets/data opens new options for networking Networking based on such data-centric addresses nicely supports an intuitive programming model publish/subscribe
Aggregation a key enabler for efficient networking Other options data storage, bradcasting aggregates also well supportable
26
