Weave: An architecture for tailoring urban sensing applications across multiple sensor fabrics

1
Weave An architecture for tailoring urban
sensing applications across multiple sensor
fabrics

• V. Kulathumani, M. Sridharan, R. Ramnath, A.
  Arora
• Dept. of Computer Science and Engineering
• The Ohio State University

2
Motivating scenario campus surveillance

• Track suspicious persons across campus outdoors
  and indoors

• Outdoor surveillance by building mounted
  camera network

• Monitors suspicious activity such as stopped car

• Indoor surveillance by motion triggered camera
  network

• Should only monitor suspicious persons
• Should be cued to operate by outdoor network

3
Motivating scenario locality queries

• Variety of networks deployed to monitor local
  environment

Users ask locality specific queries Where is
Anish? Where is elevator?
4
Challenges

• Developing applications on already existing
  networks
• Example programming detector for stopped car
  near entrances
• Users who write applications may be different
  from network designers
• Composing applications across independent, varied
  networks
• Example indoor network cued by outdoor network
• Rapid prototyping despite retrieving semantically
  different data
• Events of interest not known a priori or change
  quickly
• Numerous networks, each sensing different data
• Need unified way to retrieve

5
Contributions

• Architecture for urban sensing systems (Weave)
• Individual networks viewed as programmable
  fabrics
• Fabrics are independently managed
• Fabrics need not share data
• Fabrics conform to APIs
• Generic (Standard across fabrics)
• Domain specific
• Domain specific API for class of Search
  applications
• Illustrating use of Weave in composing
• Campus surveillance application
• Locality query applications

6
System components
Client
Applications that access fabrics via
services Granted access via access manager
Directory
APIs
Communicate via standardized APIs
APIs
Access Manager
Independent decoupled networks Sense, communicate
physical phenomena Need not share state with
other fabrics
Maintain directory of services offered Grants
access rights for fabric
7
System components Specific to Search applications

• Object types
• Defined by predicates on physical phenomenon
• Cooperative / non-cooperative
• Attributes such as location and velocity
• Objects
• Instances of object types
• Detector
• Associated with each object type
• Implemented by
• Fabric designer
• Client

8
Generic APIs in Weave
Register
FabricSession Register(Fabric F, Client C)

• Handles authentication
• FabricSession used in future references
• Contains expiration

9
Generic APIs in Weave
Instantiate
Boolean Instantiate(FabricSession, Detector,
Shareable)

• Program clients detector on fabric
• User should be aware of internal fabric APIs
• Made available to other Clients using Shareable
  
• Security
• Run in isolation
• Restrict hardware APIs

10
Vertical APIs for Search class
Search
SearchSession Search (FabricSession, ObjectType,
Object, Locations, Parameters, Persistence,
Duration, Periodicity, SearchListener)

• Operates in following modes
• Find if object exists
• Search temporally
• Search spatially
• SearchListener is call back provided by Client

11
Locality specific querying
Where is Anish? Search CSE PeopleNet fabric and
Recreation Center fabric
Register with fabrics CSE peopleNet and RPAC
PeopleNet
FabricSession_cse Register(CSE_peopleNet,
Client) FabricSession_rec Register(REC_peopleNet,
Client)
12
Locality-specific querying
Search PeopleNet CSE for Anish
Result_cse Search (FabricSession_cse, ADVERTISED,
ANISH, LOCATION_ALL, Parameters, ONCE, NOW, NULL,
SearchListener)
CSE building PeopleNet tracking
13
Campus surveillance scenario
Register with fabrics Outdoor camera, Indoor
motion triggered camera
FabricSession_m Register(OutdoorFabric, Client)
FabricSession_ca Register(IndoorCamera, Client)
Instantiate detector for stopped car near
entrance on outdoor fabric
Result_m Instantiate(FabricSession_m, stoppedCar,
YES)
14
Campus surveillance scenario
Search for StoppedCar object type
Result_m Search (FabricSession_m, StoppedCar,
NULL, LOCATIONALL, Parameters, PERSISTENT, 0,
FOREVER, SearchListener)
Calls SearchListener whenever new object is
detected
Stopped car detected This will cue indoor fabric
15
Campus surveillance scenario
Outdoor fabric cues indoor camera using time and
location
Result_ca Search (FabricSession_ca, ANY, NULL,
LOCATIONS, Parameters, PERSISTENT, 120, 1,
SearchListener)
LOCATIONS set to near entrance of building
ObjectType is ANY Duration - next 120 seconds
Periodicity - return results at 1 second interval
Indoor camera takes picture
16
Related work

• Simplifying application design across
  heterogeneous networks
• Senserweb
• Push based model
• GENI
• All services in or architecture compatible with
  GENI user services
• We do not force isolation
• Viewing network as database of queryable object
• TinyDB
• Queries on raw data
• We operate on semantic values exposed by fabric
• Focus on optimizing querying in a single network

17
Current work

• Implement low level services necessary for fabric
  model
• Power management
• Secure scheduling services
• Security
• Jamming detection
• Restricted set of hardware APIs accessible to
  client