Page 1
HINC – Harmonizing Diverse Resource
Information Across
IoT, Network Functions and Clouds
Duc-HungLe, Nanjangud Narendra, Hong-Linh Truong
Distributed Systems Group, TU Wien
[email protected]
http://dsg.tuwien.ac.at/staff/truong
FiCloud 2016, Vienna, 24th August 2016 1
Page 2
Outline
Background and motivation
HINC framework
Distributed resource information model
Architecture and implementation
Testbed and experiments
FiCloud 2016, Vienna, 24th August 2016 2
Page 3
Background - Systems
3
Analysis &
managementHot deploy
Control
Re-route
FiCloud 2016, Vienna, 24th August 2016
Page 4
Background – elastic service models
Cloud service models
Networks
Network function
virtualization
Pay-per-use IoT
communication
IoT
Fixed IoT infrastructures
On-demand IoT
Human participation
(sensing and analytics)
FiCloud 2016, Vienna, 24th August 2016 4https://arrayofthings.github.io/
http://www.sktelecom.com/en/press/detail.do?idx=1172
Page 5
Background - application scenarios
Emergency responses, on-demand crowd sensing, Geo
Sports monitoring, cyber-physical systems testing, etc.
FiCloud 2016, Vienna, 24th August 2016 5
Need to have an end-to-end provisioning of resources
E.g., sensors, network function services, storage, virtual machines
Short, crucial and heavily workload; elasticity and uncertainties.
Geo Sports: Picture courtesy
Future Position X, Sweden Indian Overfly collapses
figure source: http://timesofindia.indiatimes.com
Page 6
Motivation – End-to-End resource
slice provision
6
Emergency
response
Hospital &
traffic
Emergency
response
service
Early
treatment
protocol
Best route
to the
hospital
Most
suitable
hospital
- Wearables
- Mobie medical
equipment
- First aid info.
- Vehicle capability
- Location
- Hospital capability
- Traffic status
Victims
Distributed
resource
management
IoT resource
provisioning
Dedicate
sub-network
Coordinate
operations
FiCloud 2016, Vienna, 24th August 2016
Page 7
Motivation – End-to-End resource
slice provisioning
FiCloud 2016, Vienna, 24th August 2016 7
End-to end
Resource slice
CPS Applications/Virtual
infrastructures
http://sincconcept.github.io/
This paper: harmonize resource information from
IoT, network functions and cloud providers for
resource slice provisioning
Page 8
Examples of existing
providers/modelsProvider Category APIs Information models
FIWare Orion IoT RESTful (NGSI10), one-time
query or subscription
High level attributes on
data and context
FIWare IDAS IoT RESTful for read/write custom
models and assets
Low level resource
model catalogs
IoTivity IoT REST-like OIC protocol, support
C++, Java and JavaScript
Multiple OIC model
OpenHAB IoT RESTful for query and control
IoT resources
Low level resource
model catalogs
OpenDayLight Network Dynamic REST generated from
Yang model (model-driven)
Low level resource
model catalogs
OpenBaton Network RESTful for network service
description
ETSI MANO v1.1.1 data
model
OpenStack Cloud RESTful, multiple language via
SDK, OCCI, CIMI
OpenStack model,
OCCI, CIMI
8FiCloud 2016, Vienna, 24th August 2016
Page 9
Approach
9FiCloud 2016, Vienna, 24th August 2016
• Avoid top-down
• Design a “super” model to manage the world.
• Focus and suitable for single-purpose solution.
• Bottom-up
• Let providers use their own models.
• Integration and link diverse types of information.
• Adaptor: to interface with providers’ APIs.
• Transformer: integrate our distributed resource model.
• Focus on resource relationships across IoT, Network
functions and clouds.
Page 10
Information model
Physical: Sensor/actualtor/devices in providers’ models
Virtual IoT: SD-Gateway and capabilities.
Network functions: edge-to-edge, edge-to-cloud network.
Clouds: VM, data services, data analytics.
10FiCloud 2016, Vienna, 24th August 2016
Page 11
Resource information integration
• The model aims to be extensible to cater
multiple underlying devices and services.
• To cope with the rapidly increasing of systems.
• A process to interface with resource providers.
11FiCloud 2016, Vienna, 24th August 2016
Page 12
Examples
FiCloud 2016, Vienna, 24th August 2016 12
"data": {"DeviceProps": {
"commandURL": "http://...OpenIoT/..","lastIP": "195.97.103.225","commands": true },
"asset": {"name": "00:3b:B6:BodyTemperature","description": "asset model protocol" },
"model": "SENSOR_TEMP","registrationTime": "2015-04-16T15:39:58Z","status": "Active","sensorMetaData": [
{"ms": {"dataType": "BodyTemperature","unit": "Celsius","rate": "10" }
}]}
{"type": "LocationItem","link": "http://..../rest/items/DemoLocation"
}
Resource from OpenHAB
- Simple data format.- A link for more information.- Information is static.
- Complex data format.- Have control capability.- More meta data.
A resource from OpenIOT
Page 13
Examples
FiCloud 2016, Vienna, 24th August 2016 13
"SoftwareDefinedGateway":{
"uuid": "5a60...",
"name": "gateway1",
“datapoints": [
{ “name": “Temp1",
"datatype": "BodyTemerature",
"measurementUnit": "Celsius",
"resourceID":
"00:3b:B6:BodyTemperature",
"extra": [
<imported List 1 and List 2> ...}
], },
“controlpoints”: [
{ "name": "changeRate",
"resourceID": "00:3b:B6:BodyTemperature",
"description": "change sensor rate",
"reference":"http://.../OpenIoT/assets/..",
} ], }
Virtual IoT resource information
- Software-Defined Gateway
wraps a set of capabilities.
- Data Point extracts a set of
interesting attributes for
higher level management.
- Control Point contains a
reference to the provider
API for controlling
resources.
Page 14
Architecture
14
Global management
service
- Run on users’ site.
- Coordinate Local
Management Service.
- Manage relationships.
Local management
service
- Deployed on gateway or
network station.
- Interface with provider.
- Transform information.
FiCloud 2016, Vienna, 24th August 2016
Page 15
Prototype
15FiCloud 2016, Vienna, 24th August 2016
http://sincconcept.github.io/HINC/
Page 16
Testbed
16FiCloud 2016, Vienna, 24th August 2016
Page 17
Testbed
17FiCloud 2016, Vienna, 24th August 2016
In-lab testbed:
- Server: 8 CPU-i7, 3.60GHz, 32GB RAM
- Edge: 100 dockers with emulated sensors + gateways.
- Network: virtual routers (https://www.weave.works/)
- Cloud: event processing.
Page 18
Testbed
18FiCloud 2016, Vienna, 24th August 2016
Distributed testbed
- Edge: physical/virtual machines on different cities.
- Communication: CloudAMQP.
- Cloud: AmazonEC.
Page 19
Reducing complexity in accessing
and control resources
19FiCloud 2016, Vienna, 24th August 2016
1. Query data points
2. Control the
resource
3. Query network
functions and clouds
Page 20
Query time by number of gateways
Distributed sites
testbed
FiCloud 2016, Vienna, 24th August 2016 20
In-lab testbed
Page 21
Gateway’s response time variability
FiCloud 2016, Vienna, 24th August 2016 21
Distributed sites
testbed
In-lab testbed
Page 22
Number of sensors and locations
FiCloud 2016, Vienna, 24th August 2016 22
Query time from
distributed sites
Query time by
number of sensors,
distributed sites
Page 23
Conclusion and future work
Harmonizing information in 3 dimensions:
High-level view of low level resources
End-to-end view of IoT, network functions and clouds
Large-scale view of highly distributed sites
Future work:
Information-centric resource provisioning
Dynamic IoT infrastructure configuration
End-to-end resource optimization
23FiCloud 2016, Vienna, 24th August 2016
Page 24
Thanks for your
attention!
Questions?
Hong-Linh Truong
Distributed Systems GroupTU Wien
dsg.tuwien.ac.at/staff/truong
FiCloud 2016, Vienna, 24th August 2016 24