Vous êtes sur la page 1sur 9

Code offloading in accessors

Atul Sandur

29th June, 2016

Atul Sandur

Code offloading in accessors

29th June, 2016

1/9

Motivation

Accessors used to compose heterogeneous devices/services


Could be running on smart sensors/actuators, smartphone, swarm
boxes, etc.
Resource/energy constrained, hardware limitations
Can we move an accessor or parts of its execution to other devices?
Scenarios:
Object recognition in video input from smartphones or real time video
surveillance systems
Spike in other application activity, low battery power or network
bandwidth limited for data transfer
Spare sensors which activate in case of primary sensor failures

Atul Sandur

Code offloading in accessors

29th June, 2016

2/9

High level architecture

Figure: Architecture for offloading framework

Atul Sandur

Code offloading in accessors

29th June, 2016

3/9

Prototype using smartphones


Android accessor host
Apache Cordova with JXCore plugin
Allows using Javascript for cross platform mobile development
Platform specific wrappers provide access to device capabilities like
sensors
Comes prebuilt with APIs for range of mobile sensors
Has a plugin for running Node.JS code in Android
Enables reusing node host implementation along with its modules,
minimizes maintenance burden

Monitoring accessor
Implemented for node host, one per host
Runs periodically to collect statistics for execution time of accessor
react functions

Figure: Mobile-swarmlet-cloud
Atul Sandur

Code offloading in accessors

29th June, 2016

4/9

Demo

Monitoring accessor
Cordova deployment
Cordova on mobile

Atul Sandur

Code offloading in accessors

29th June, 2016

5/9

Next steps

Implement accessor offloading within Cordova host, for a specific


scenario like object recognition in video
Design coordinator/decision making model for offloading on each host
Implement coordinator along with offloading decision making
Design energy model for monitoring energy characteristics of an
application

Atul Sandur

Code offloading in accessors

29th June, 2016

6/9

Extra slides: Types of offloading

Offload parts of computation in an accessor to another device


Specified ports can be handled by the cloud instead of device
Historical queries on a certain time range of sensor data vs. point in
time queries
Javascript level function offloading

Accessors are just sensor data collection points


All interactions with accessor handled by cloud endpoint instead of
actual device
Accessor that provides ML and optimization algorithms to handle noisy
sensor data and make actuation decisions

Complete accessor offloading


Accessors are migrated to a hot spare to handle component faults
Proximity sensors in self driving cars

Atul Sandur

Code offloading in accessors

29th June, 2016

7/9

Prototype using smartphones


Have a rich set of sensors to leverage for lighting up interesting
scenarios
Good candidate for offloading, with complex applications and
resource constraints
Offloading across heterogeneous platforms on mobile, cloud and
intermediate swarm box

Figure: Mobile-swarmlet-cloud
Atul Sandur

Code offloading in accessors

29th June, 2016

8/9

Update

Working implementation of Android host for accessors based on


Apache Cordova with JXCore plugin for Node.JS code
Monitoring accessor for node host, collecting react execution time
statistics
Progress on understanding/design of accessor offloading framework

Atul Sandur

Code offloading in accessors

29th June, 2016

9/9

Vous aimerez peut-être aussi