Keynote on Mobile Grid and Cloud Computing

한국해양과학기술진흥원
Mobile Grid and Cloud Computing
Opportunities and Challenges
2013.9.22
Sayed Chhattan Shah, PhD
Senior Researcher
Electronics and Telecommunications Research Institute, Korea
etri.re.kr | https://sites.google.com/site/chhattanshah/

Outline
 Background
 Mobile Grid and Cloud Computing
 Cloud Robotics
 Mobile Ad hoc Computational Grid and Cloud
 Opportunities
 Research Challenges
 Future Research Directions
 Conclusion

A collection of independent computers that appear to the
users of the system as a single computer
ATM Internet
Distributed System

Types of Distributed Systems
Cluster
Grid
Cloud

Overview: Clusters x GridsCluster - How can we use local networked resources
to achieve better performance for large scale
applications?
 High-speed LAN
 Centralized resource and task management
How can we put together geographically distributed
resources to achieve better performance?
 WAN
 Distributed resource and task management
Cluster and Grid Computing

Information
Generators
Information Distributed
Over the Grid
Customer
Access to
Information
Grid
 Computing power should be available on demand, for a fee
 Just like the electrical power grid
Basic Idea

Cloud Computing
Everything — from computing power to computing
infrastructure and applications are delivered as a service

Grid Computing
Computational Grids and Clusters have been extensively
deployed and widely used to solve complex and
challenging problems in science and engineering areas
such as drug design, earthquake simulation, and climate
modeling

Grid Computing
Due to recent advances in mobile computing and
communication technologies, it has become feasible to
use mobile nodes as a contributing entity to Grids and
Clouds

Grid Computing
Several approaches have been proposed to integrate
mobile nodes with Grid and Cloud computing systems
Mobile Grid and Cloud Computing Mobile Ad hoc Grid and Cloud Computing
Mobile Ad hoc
Network

Mobile Cloud Computing
 Data processing and data storage happen outside of mobile devices

Mobile Grid Computing
 Data processing and data storage happen outside of mobile devices

 Enabling Factors
 Wireless networks
• 3G networks: 14.4 Mbps
• 4G networks: 100~128 Mbps

Benefits
 Improved data storage capacity and processing power
 Apple’s iCloud enables users to store and synchronize data in the cloud
 Users can execute computationally and data-intensive applications
on mobile devices
 Image processing
 Natural language processing
 Video processing
 Extended battery life
 Improved reliability
 Data and application are stored and backed up on a number of computers

Cloud Robotics
 Robots rely on a cloud-computing infrastructure to access
vast amounts of processing power and data
 Robots can offload heavy tasks
 Image processing
 Voice recognition

Benefits
 Provides a shared knowledge database
 Organizes and unifies information about the world in a format usable
by robots
 Robot Goggles
 Upload images -> Download Semantic
• Object name
• 3D model, mass, materials, friction properties
• Usage instructions - function, how to grasp, operate
• Context and Domain knowledge

Benefits
Skill / Behavior Database
 Reusable library of “skills” or behaviors that map to perceived task
requirements / complex situations
 Matrix Movie Scene
 For humans, still science fiction
 For robots?

Benefits
Offloads heavy computing tasks to the cloud
 Cheaper, lighter, easier-to-maintain hardware
 Longer battery life
 Less need for software pushes/updates
 CPU hardware upgrades are invisible & hassle-free

Cloud Robotics Projects
 Researchers at Social Robotics Lab have built a cloud
computing infrastructure to generate 3-D models of
environments
 Allowing robots to perform simultaneous localization and
mapping much faster than by relying on their onboard
computers
• SLAM refers to a technique for a robot to build a map of the environment without
a priori knowledge, and to simultaneously localize itself in the unknown
environment

 At CNRS, researcher are creating object databases for
robots to simplify the planning of manipulation tasks like
opening a door
 The idea is to develop a software framework where
objects come with a "user manual" for the robot to
manipulate them

 Gostai, a French robotics firm, has built a cloud robotics
infrastructure called GostaiNet, which allows a robot to perform
speech recognition, face detection, and other tasks remotely
 Jazz telepresence robot uses the cloud for video recording and
voice synthesis

Cloud Robotics
 Same as:
 Remote computing?
 Mobile cloud computing?
 Mobile Grid Computing?

Computation Offloading
Migrating computation to more resourceful computers
Computation offloading = Surrogate computing = Remote execution

 Offloading decisions are usually made by analyzing several
parameters including
 Bandwidths
 Server speeds
 Available memory
 Server loads
 Amounts of data exchanged between servers and mobile systems

 Offloading approaches are classified based on various factors
including
 Why to offload
• Improve performance or save energy
 What mobile systems use offloading
• Smart phones, robots, sensors
 Infrastructures for offloading
• Cluster, Grid, Cloud
 Types of applications
• Multimedia, gaming, calculators, text editors

 Application partitioning
• Static vs. dynamic
 When to decide offloading
• Static vs. dynamic
 Offloading data-intensive interdependent tasks
 Offloading small tasks
• May not improve performance or reduce energy consumption

Mobile Ad hoc Computational Grid

Mobile Ad hoc computational Grid
The mobile Grid and Cloud computing systems are
restricted to infrastructure-based communication
systems such as cellular network, and therefore cannot
be used in mobile ad hoc environments

 A distributed computing infrastructure that allows mobile
nodes to share computing resources in mobile ad hoc
environments
Service Provider Node
Service Requesting Node
Service Requesting Node
Service Broker Node
Mobile Ad hoc Network

 Computational Grid
 allows distributed computing devices to share computing resources
to solve computationally-intensive problems
 Mobile ad hoc network
 a wireless network of mobile devices that communicate with each
other without pre-existing network infrastructure
COMPUTATIONAL GRID
MOBILE AD HOC NETWORK
MOBILE AD HOC COMPUTATIONAL GRID
APPLICATIONS
MOBILE NODES

Autonomous Threat Detection in Urban Environments
 A group of miniature autonomous mobile robots are
deployed in urban environments to detect and monitor a
range of military and non-military threats
 Use sophisticated image and video processing algorithms
 Vision-based navigation algorithms to navigate in the
environment
 Beyond capabilities of single miniature mobile node

Construction of 3D-Map and Identification of Targets within Map
A set of miniature unmanned aerial vehicles or mobile ro
bots can be deployed in a targeted area
 Broadcast live video streams
 Processed to construct map and indentify stationary and mobi
le targets
 Requires huge processing power

Contents
38

Video Data Mining
Fighting units need to know activities of target in the last
60 minutes from archived video content which requires
storing live video content
 To store content, a large amount of storage space is required
 Processing of stored video content according to user demand
also requires large amounts of processing power
 Nodes owned by soldiers or fighting units can form an ad hoc
data and computational Grid

Future Soldier
 In warfare soldiers may experience
physical and mental problems
 In such situations, various biomedical
devices can be used to continuously
monitor the soldiers' psychophysiological
health
 Data from devices can be used to assess physical
and mental health
 Soldiers also need to rely on various sensing,
processing and communication systems in the
vicinity to achieve situational awareness and understanding
of the battlefield
 Simultaneously executing computationally-intensive
models for deriving physiological parameters and
for acquiring battlefield awareness in real time
requires computing capabilities that go beyond
those of an individual sensing and processing devices

Mobile Ad hoc Computational Grid
 Mobile ad hoc computational Grid is attractive even
when network infrastructure is available
 Short-range wireless communication consumes less
energy and provides faster connectivity
 3G networks: 2~14.4 Mbps
 4G networks: 100~128 Mbps
 Wi-Fi LAN 400Mbps

Research Challenges and Future Research Directions
 Compared to traditional parallel and distributed
computing systems such as Grid and Cloud mobile ad
hoc computational Grid is characterized by
 Node mobility
 Limited battery power
 Low bandwidth and high latency
 Shared and unreliable communication medium
 Infrastructure-less network environment
• No one is in charge
• No one to provide standard service

Node mobility
Node
RESOURCE
ALLOCATION
Node
Task
Grid
Members
Task Queue
Task
NODE
SELECTION
DISPATCHER

Node mobility
 Global Node Mobility
 Task Failure
 Local Node Mobility
 Increased data transfer times
 Mobility of an Intermediate Node
 Increased data transfer times and may disconnect network
 Approaches:
 Task migration
 Task reallocation
 In both cases, delay due to reallocation or migration of task

Node mobility
 To improve performance and avoid task failure or migration,
nodes with long-term connectivity are required for the allocation
of tasks
 An effective and robust two-phase resource allocation scheme
 Exploit the history of user’s mobility patterns in order to select nodes
that provide long-term connectivity
 Location prediction schemes
 Use node’s direction and speed to predict future connectivity

Node mobility
 Makes it difficult to design an efficient and robust resource
discovery and monitoring system
 After reporting status a node may move across the coverage area
 Grid management system would assume that status is valid and would make decisions
accordingly
 To avoid this problem
• Proactive approach
 Resources can be monitored continuously or with minimum update interval
 In both cases, there will be a communication overhead
• Use reactive approach
 Reduces communication cost but introduces delay

 Power management
 Main sources of energy consumption are CPU processing,
memory, and data transmission in the network
 Key factors that contribute to transmission energy consumption
• transmission power required to transmit data and
• communication cost induced by data transfers between tasks
 Most of the schemes are focused on the conservation of
processing energy
 Saving energy in data transfers between tasks remains an open
problem
• becomes even more critical for data-intensive parallel applications

 Power management
 Energy efficient resource allocation scheme
• Aims to reduce transmission energy consumption and data transfer
cost
• Basic idea is to allocate tasks to nodes that are accessible at minimum
transmission power
Y
1TPL
X
1TPL 3TPL 4TPL2TPL

 Constrained communication environment due limited power,
shared medium and node mobility
 Suffers from low bandwidth, high latency and unstable
connectivity problems
 In such an environment, data transfer cost is very critical for
application and system performance
 To reduce data transfer costs, directional antennas, efficient medium
access control, channel switching, and multiple radios are a few
promising approaches
 Parallel applications usually consist of a range of tasks with varying
bandwidth, processing, and deadline constraints
 Work is needed to develop a Grid management system that should
exploit a diverse range of links, node capabilities, and application’s
characteristics

 Dynamic network performance
 Bandwidth at different network portions varies over the time
and different nodes often experience different connection
quality at the same time due to the traffic load and
communication constraints
 Grid management system that should consider network
dynamics particularly when data-intensive interdependent
tasks need to be allocated

 Task Migration
 To improve application performance and resource utilization,
and to avoid task failure and load imbalance
 Most common migration strategy is to estimate migration cost
and determine task completion time before and after the
migration of task
 However, estimation of migration cost particularly of data
intensive task is not straightforward due to dynamic
communication environment
 How to estimate data transfer time?
 In addition, this strategy works well when amount of data
transmitted or processed by a task is known in advance

 Parallel programming model
 Programming model provides an abstract view of computing
system
 The traditional parallel programming models do not deal well
with communication issues
• Therefore are not suitable for mobile ad hoc environments where
communication latencies and link failure and activation ratios are
too high
 Actor-based programming model could be the possible
candidate because it deals quite well with high latencies,
offers lightweight migration and can be easily adopted to deal
with node mobility

 Security risks
 Mobile ad hoc computational Grid may include heterogeneous
devices owned by various individuals, organizations and
groups
 can be used in various scenarios such as military, disaster
relief and urban surveillance where security is a primary
concern
 Compared to traditional wired and wireless networks, design
of an efficient security system for mobile ad hoc computational
Grid is a challenging task
• due infrastructure-less network environment, shared
communication medium, and node mobility

 Incentive mechanism
 Assume a scenario where an individual travelling with
strangers requires additional computing resources to perform
a computationally intensive task
• The problem is how to or what will motivate an individual to share
her resources with a stranger?
 To address this problem, a few solutions have been proposed
in the literature where either battery power or processing
cycles are traded
• Effective when both parties are in need of resources from each
other
 The design of an incentive mechanism for mobile ad hoc
computational Grids is difficult due to lack of central authority
and ad hoc system architecture

 Architecture for mobile ad hoc computational Grid
 Centralized
• Single point of failure and scalability
 Decentralised
• Group management
• Ineffective resource allocation
 Distributed
• Ineffective resource allocation
 Hybrid architecture

 Failure management
 Migrate the task or restart the task on another node
 estimation of task completion time with and without migration
cost?
 Quality of Service support
 application’s demands such as energy, bandwidth guarantees
and real-time services
 Standards for heterogeneous environments
 Wireless Communication Technologies

FARE-SHARE Project
 Aims to exploit collective capabilities of nearby devices
 To execute compute-intensive models for deriving physiological parameters
and for acquiring context awareness in real time

 Aims to develop a system for aerial surveillance to assist a rescue team in
case of a disaster situation
 Video data is submitted to an evaluation system via a high performance
communication network where a 3D virtual world is created in quasi real
time
Collaborative Drones

 Aims to develop a system for aerial surveillance to assist a rescue team in
case of a disaster situation
Master-Slave Collaborative UAV Surveillance System Architecture

 Troops frequently have to wait until they’re back at camp to download latest updates
 Mission opportunities may erode because the information needed at the tactical edge isn’t immediately
available
 CBMEN program aims to rapidly share up-to-date imagery, maps and other vital
information directly among front-line units
 Each squad member’s mobile device function as a server, so content is generated,
distributed and maintained at the tactical edge where it’s needed
 A key factor that enables CBMEN is the tremendous computing power available in
current mobile devices
 64 gigabytes of storage in a single smartphone
 A squad of nine troops could have more than half a terabyte (500 GB) of cloud storage
Content-Based Mobile Edge Networking Program

Conclusion
 Due to recent advances in mobile computing and
communication technologies it has become feasible to
design and develop next generation of distributed
applications through sharing of computing resources in
mobile and ad hoc environments
 Further investigation is required
 Resource Management
 Programming model
 Communication performance
 Mobility
 QoS support

Cloud Robotics and Networked Robots

Cloud Robotics and Networked Robots
 Peer-based Model
 Proxy-based Model
 Clone-based Model

Vision Understanding
 Attention Detection
 Body pose recognition
 Face detection
 Face pose recognition
 Eye detection
 Lip Motion Detection
 Face & eye tracking
 Mouth location & tracking
 Speaking recognition (spatial-temporal analysis)
 Facial Expression and Emotion
 Local feature analysis
 Global face pattern analysis
 Online face learning and recognition

Keynote on Mobile Grid and Cloud Computing

More Related Content

What's hot (18)

Similar to Keynote on Mobile Grid and Cloud Computing (20)

More from Sayed Chhattan Shah (13)

Recently uploaded (20)

Keynote on Mobile Grid and Cloud Computing