Artificial intelligence in cloud-based mobile radar computing
The introduction of Artificial Intelligence into mobile radar computing based on cloud resources has made it possible to combine radar resources at the stage of receiving, processing and presenting information. The radar system has become integral and flexible. The convergence of mobile applications...
Збережено в:
| Дата: | 2023 |
|---|---|
| Автори: | , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут програмних систем НАН України
2023
|
| Теми: | |
| Онлайн доступ: | https://pp.isofts.kiev.ua/index.php/ojs1/article/view/567 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Problems in programming |
| Завантажити файл: |  |
Репозитарії
Problems in programming| id |
pp_isofts_kiev_ua-article-567 |
|---|---|
| record_format |
ojs |
| resource_txt_mv |
ppisoftskievua/71/d67ead0f44de0dde5fab28d152f1e871.pdf |
| spelling |
pp_isofts_kiev_ua-article-5672024-04-26T21:28:48Z Artificial intelligence in cloud-based mobile radar computing Штучний інтелект у мобільних радарних обчисленнях на основі хмарних ресурсів Коsovets, M. Tovstenko, L. Service Oriented Architecture (SOA); Sense-Compute-Actuate (SCA); High Performance Computing (HPC); Message Passing Interface (MPI); virtual machine (VM), Machine Learning and Data Analysis (MLDM); Distributed File System (HDFS); Java Database Connectivit UDC 517.9:621.325.5:621.382.049.77 сервіс-орієнтована архітектура (SOA); збір даних-обробка-відповідна дія (SCA); високопродуктивні обчислення (HPC) УДК 517.9:621.325.5:621.382.049.77 The introduction of Artificial Intelligence into mobile radar computing based on cloud resources has made it possible to combine radar resources at the stage of receiving, processing and presenting information. The radar system has become integral and flexible. The convergence of mobile applications in portable devices with cloud computing is a revolution in the organization of distributed computing. Attention is paid to the architecture of the client part as a neurocomputer distributed in space with deep learning capabilities. Providing analysis of radar data in Real Time for a mobile platform is very important and its implementation with packet data transmission at different stages accelerate the analysis process.Problems in programming 2023; 2: 49-58 Впровадження Штучного інтелекту в мобільні радіолокаційні обчислення наоснові хмарних обчислень дозволило об’єднати радарні ресурси на етапах прийому, обробки і представлення інформації. Радіолокаційна система стала цілісною та гнучкою. Конвергенція мобільних додатків у портативних пристроях із хмарними обчисленнями — це революція в організації розподілених обчислень радарної інформації. Приділено увагу архітектурі клієнтської частини як розподіленого в просторі нейрокомп’ютера з можливостями глибокого навчання. Забезпечення аналізу радіолокаційних даних у режимі реального часу для мобільної платформи є дуже важливим, а реалізація пакетної передачі даних на різних етапах обробки прискорює процес аналізу.Problems in programming 2023; 2: 49-58 Інститут програмних систем НАН України 2023-08-04 Article Article application/pdf https://pp.isofts.kiev.ua/index.php/ojs1/article/view/567 10.15407/pp2023.02.049 PROBLEMS IN PROGRAMMING; No 2 (2023); 49-58 ПРОБЛЕМЫ ПРОГРАММИРОВАНИЯ; No 2 (2023); 49-58 ПРОБЛЕМИ ПРОГРАМУВАННЯ; No 2 (2023); 49-58 1727-4907 10.15407/pp2023.02 en https://pp.isofts.kiev.ua/index.php/ojs1/article/view/567/618 Copyright (c) 2023 PROBLEMS IN PROGRAMMING |
| institution |
Problems in programming |
| baseUrl_str |
https://pp.isofts.kiev.ua/index.php/ojs1/oai |
| datestamp_date |
2024-04-26T21:28:48Z |
| collection |
OJS |
| language |
English |
| topic |
Service Oriented Architecture (SOA); Sense-Compute-Actuate (SCA); High Performance Computing (HPC); Message Passing Interface (MPI); virtual machine (VM) Machine Learning and Data Analysis (MLDM); Distributed File System (HDFS); Java Database Connectivit UDC 517.9:621.325.5:621.382.049.77 |
| spellingShingle |
Service Oriented Architecture (SOA); Sense-Compute-Actuate (SCA); High Performance Computing (HPC); Message Passing Interface (MPI); virtual machine (VM) Machine Learning and Data Analysis (MLDM); Distributed File System (HDFS); Java Database Connectivit UDC 517.9:621.325.5:621.382.049.77 Коsovets, M. Tovstenko, L. Artificial intelligence in cloud-based mobile radar computing |
| topic_facet |
Service Oriented Architecture (SOA); Sense-Compute-Actuate (SCA); High Performance Computing (HPC); Message Passing Interface (MPI); virtual machine (VM) Machine Learning and Data Analysis (MLDM); Distributed File System (HDFS); Java Database Connectivit UDC 517.9:621.325.5:621.382.049.77 сервіс-орієнтована архітектура (SOA) збір даних-обробка-відповідна дія (SCA) високопродуктивні обчислення (HPC) УДК 517.9:621.325.5:621.382.049.77 |
| format |
Article |
| author |
Коsovets, M. Tovstenko, L. |
| author_facet |
Коsovets, M. Tovstenko, L. |
| author_sort |
Коsovets, M. |
| title |
Artificial intelligence in cloud-based mobile radar computing |
| title_short |
Artificial intelligence in cloud-based mobile radar computing |
| title_full |
Artificial intelligence in cloud-based mobile radar computing |
| title_fullStr |
Artificial intelligence in cloud-based mobile radar computing |
| title_full_unstemmed |
Artificial intelligence in cloud-based mobile radar computing |
| title_sort |
artificial intelligence in cloud-based mobile radar computing |
| title_alt |
Штучний інтелект у мобільних радарних обчисленнях на основі хмарних ресурсів |
| description |
The introduction of Artificial Intelligence into mobile radar computing based on cloud resources has made it possible to combine radar resources at the stage of receiving, processing and presenting information. The radar system has become integral and flexible. The convergence of mobile applications in portable devices with cloud computing is a revolution in the organization of distributed computing. Attention is paid to the architecture of the client part as a neurocomputer distributed in space with deep learning capabilities. Providing analysis of radar data in Real Time for a mobile platform is very important and its implementation with packet data transmission at different stages accelerate the analysis process.Problems in programming 2023; 2: 49-58 |
| publisher |
Інститут програмних систем НАН України |
| publishDate |
2023 |
| url |
https://pp.isofts.kiev.ua/index.php/ojs1/article/view/567 |
| work_keys_str_mv |
AT kosovetsm artificialintelligenceincloudbasedmobileradarcomputing AT tovstenkol artificialintelligenceincloudbasedmobileradarcomputing AT kosovetsm štučnijíntelektumobílʹnihradarnihobčislennâhnaosnovíhmarnihresursív AT tovstenkol štučnijíntelektumobílʹnihradarnihobčislennâhnaosnovíhmarnihresursív |
| first_indexed |
2024-09-21T04:05:13Z |
| last_indexed |
2024-09-21T04:05:13Z |
| _version_ |
1818527982986199040 |
| fulltext |
49
Моделі та методи машинного навчання
Introduction
There has been an intensive
development of radar technology over the
past fifty years in the military, medicine,
biology, security, agriculture and others. This
trend continues today at an ever faster pace.
The development of scientific approaches
to the acquisition and processing of radar
information is always determined by the
achievements of manufacturing technologies
for more sensitive radar sensors, which
are small in size, weight and low radiation
power, as well as the development of new
radio wave bands. Computing technologies
have received significant development: data
parallelization, streaming and neural network
calculations, and mathematical modeling.
The mechanism of access and data exchange
has improved.
Cloud computing makes data mobile
by giving users access to the cloud of their
choice from any Internet-enabled device
without requiring a detailed understanding
of the underlying computing technology. The
increasing ability of radars to sense the world
around them is leading to a growing need
for electronic database applications driven
by data under the control of workflows.
The main focus is on the collection, origin
of the data of these workflows, necessary
to validate the workflow and determine the
quality of the generated data. The challenge
is to capture and use a single origin metadata
that matches the needs of the domain,
minimizing the modification burden on the
services and the overhead of the workflow
engine and services. The framework is based
on the generation of discrete data related to
the origin, during the work cycle of the life
cycle execution can be aggregated to form
complex data and process origin graphs,
which can cover different workflows. The
implementation uses a loosely coupled
architecture where the capabilities of the
system meet the needs of detailed information
gathering.
This idea raised the level of
abstraction for the programmer. Klient needs
additional capacity based on demand, drives
hybrid cloud concept. That is, the use of a
public cloud as a continuation of the internal
infrastructure – a platform for automating
machine learning procedures, analyzing
streaming data; building your own cloud with
open source software and cloud security.
1. Client cloud environment for radar
information processing
The topic of data analytics in the
cloud is huge and booming in the new data-
driven open data and design paradigm [1].
Consider data from radar sensors on board
an autonomous vehicle. If the value of radar
data rapidly diminishes with time coming
in every second, and the volume is so large
that it cannot be stored, real-time processing
or data reduction is the only way to analyze
УДК 517.9:621.325.5:621.382.049.77 http://doi.org/10.15407/pp2023.02.049
M. Коsovets, L. Tovstenko
ARTIFICIAL INTELLIGENCE IN CLOUD-BASED
MOBILE RADAR COMPUTING
The introduction of Artificial Intelligence into mobile radar computing based on cloud resources has made
it possible to combine radar resources at the stage of receiving, processing and presenting information.
The radar system has become integral and flexible. The convergence of mobile applications in portable
devices with cloud computing is a revolution in the organization of distributed computing. Attention is
paid to the architecture of the client part as a neurocomputer distributed in space with deep learning
capabilities. Providing analysis of radar data in Real Time for a mobile platform is very important and its
implementation with packet data transmission at different stages accelerate the analysis process.
Keywords: Service Oriented Architecture (SOA), Sense-Compute-Actuate (SCA), High Performance
Computing (HPC), Message Passing Interface (MPI), virtual machine (VM), Machine Learning and Data
Analysis (MLDM), Distributed File System (HDFS), Java Database Connectivity (JDBC).
© М. Косовець, Л. Товстенко, 2023
ISSN 1727-4907. Проблеми програмування. 2023. №2
50
Моделі та методи машинного навчання
data coming from unlimited data streams,
which has created part of the intellectual
foundation of modern systems [2].
Machine learning has become central
to cloud computing programs. Although
machine learning is considered as a part of
the field of artificial intelligence, it has roots
in statistics and the theory and practice of
mathematical optimization. It has gained in
importance in recent years as a number of
critical applications have taken place. This
includes real-time radar image recognition.
Algorithmic progress and more powerful
computers make it possible to train deep
neural networks. It introduces some of the
core machine learning tools available in
public clouds, as well as toolkits that can be
installed in a private cloud.
Summarizing the above, it can be noted
that in recent years a significant contribution
has been made to the organization of the client
part – the operator’s workplace, improving
access to information databases [3]. They
began to introduce elements of artificial
intelligence, teaching search engines to find
information on the distinguished features of
the problem being solved. Also, attention
was paid to the architecture of the client part,
which can be explained by a small selection
of standard computing tools. Today, the
client part is a neurocomputer distributed in
space with deep learning capabilities. While
batch analysis of big data is important, real-
time data analysis is becoming increasingly
critical. For example, data from radars
that control complex systems for targeting
weapons to objects on board a vehicle [4].
To improve performance, we introduce
the idea of container programs, which
allow us to create an application container
that will run on any machine, and we do
not need to make changes to the machine
itself. This is a way to share ready-to-run
applications such as a mobile radar server or
a simulation application. There are various
ways to scale applications in the cloud for
greater parallelism. You can create an HPC
(High Performance Computing) style cluster
in the cloud to run MPI (Message Passing
Interface) applications, or based on Docker
on a distributed cluster. Docker is a software
platform for rapid development, testing and
deployment of applications. Docker packages
software into standardized building blocks
called containers. Each container includes
everything the application needs to run:
libraries, system tools, code, and runtime.
However, it was clear that the need for
reproducible container-based environments
was especially important for radar computing,
and would really help both users and cluster
administrators. As a rule, the centers provide
standard libraries that meet the needs of
most users. Containers allow you to install
software [5].
Large data sets from radar sensors,
represented by the Internet of Things (IoT),
can perceive information, exchange data,
calculate and receive data streams from radar
sensors and contribute to the emergence of
the big data paradigm. We will discuss the
new architecture of the Internet of Things
(IoT) radar system, large-scale additional
sensor networks, the integration of sensor
networks, radar sensor data and related
methods for collecting context, problems of
cloud management, storage, archiving and
processing of radar sensor data.
The world is filled with devices
including sensors and data processors. This
concentration of computing resources makes
it possible to detect, capture, collect and
process real-time data from connected radar
devices serving many different applications,
including environmental monitoring [6].
These developments have led us to the era
of the Internet of Things (IoT). However,
a sense of the environment and the objects
that inhabit this environment has become
synonymous with the introduction of
pervasive or ubiquitous computing. Sensor
networks are the main enabler of IoT. The
IoT has three unique features: Periodic
Probing, Regular Data Collection, and
Sense-Compute-Actuate (SCA) cycles. But
this will only be useful if the “terabytes” of
data it generates can be collected, analyzed
and interpreted [7].
There are a number of problems
associated with the processing and analysis
of data. Therefore, extensive data have
become popular [8] and are analyzed based
on some of its characteristics. There are three
characteristics that can be used to define big
51
Моделі та методи машинного навчання
data, also known as 3Vs: volume, variety,
and speed.
• Volume: data size (terabytes), etc.;
• Diversity: data types in the form of
big data received from radar sensors;
• Rate: The rate at which data is
generated.
The frequency of processing depends
on the user’s requirements for obtaining
the result. Generally, we can distinguish
three main categories: random, frequent and
real-time. There are a number of problems
associated with big data – collection, storage,
search, analysis and virtualization. Additional
technologies are identified, such as database
parallel processing (MPP), distributed file
systems, cloud computing technologies to
complement radar data management.
Methods for solving problems have
been developed: extraction, transformation,
integration, sorting and manipulation of
data. The main methods of obtaining data
consist of five stages: definition, search,
transformation, entity resolution, response to
a request. These conceptual steps are more
related to the traditional field of data science.
However, the technology behind these
conceptual steps will differ significantly due
to the unique characteristics of the radar
data. Radar systems provide data using very
complex and advanced sensors. They need
high computing power for processing and
storage.
Provision of a resource as an Internet
technology for external clients. Cloud
computing plays a significant role in the
IoT paradigm. Cloud storage and processing
capabilities are critical to delivering the
service vision; a model that has emerged
from cloud computing as a style of computing
where large-scale IT-related capabilities are
delivered «as a service» using IoT.
Cloud computing includes three main
layers or models: Infrastructure as a Service
(IaaS), Platform as a Service (PaaS) and
Software as a Service (SaaS). In addition to
the above core layers, some other layers are
also introduced and discussed in literature
such as Database as a Service (DBaaS), Data
as a Service (DaaS), Ethernet as a Service
(EaaS), Network as a Service (NaaS).
In the IoT domain, sensor data is not
the only information that will be stored. More
importantly, relevant contextual information
will always be added to the original sensor
data for further retrieval. For example,
Sensor-Cloud is a framework that aims to
manage physical sensors by connecting them
to the cloud. The Sensor-Cloud system uses
sensors to describe sensor metadata.
The problems associated with rich
data can be divided into two categories:
engineering and semantic. The challenge
in engineering is to efficiently perform
activities such as query and storage. The
semantic challenge is to extract information
from large amounts of unstructured raw data.
The main problems of big data
management are:
• High volume processing using low
power digital processing architecture.
• Use of adaptive machine learning
methods for real-time data analysis.
• Developping scalable data
warehouses that enable efficient data analysis.
Big data consists of useful information
mixed with «dirty» (noise, false and raw)
data. Advanced systems and innovative
technologies will effectively process huge
amounts of «raw» data and highlight useful
information. This will require a concentrated
effort of IoT researchers and practitioners.
The achievement is an overview of the
current and future uses of rich data in the
IoT environment, but there is no claimed
disclosure of the architecture of the IoT
environment using rich data by abstraction
layers [9].
Using the services of Hadoop, YARN,
HDFS and others simplifies the work of
information processing. So, Hadoop is a
freely distributing set of utilities, libraries
and a framework for developing and
executing distributed programs running on
clusters of hundreds and thousands of nodes
– this is the world of big data. Highload starts
writing logs, then big data is already required
for their processing. In general, big data is a
very broad field with machine learning and
semantic analysis. Actually, there exist four
tasks: the first is reading data and three data
processing tasks. These tasks are proposed to
be solved using Hadoop.
YARN is Yet Another Resource
52
Моделі та методи машинного навчання
Negotiator, which is responsible for running
tasks on multiple machines, manages the
computing resources of the cluster, and
simply submits the task for execution. It
doesn’t know what tasks it passes off, and
MapReduce directly executes the tasks that
it starts.
When designing HDFS, we
simultaneously took into account the ideas
that the system should be fault-tolerant, that
is, data should be stored on different servers.
The system should be distributed so that it
can be easily added, removed new servers. A
file represented by a name and their content,
the content is broken into a set of blocks.
When developers and admins work with
the command line and data, the work goes
through an API written in Java
The second part related to data
processing is the second, third and fourth
tasks. This is Yet Another Resource
Negotiator/MapReduce. The principle of
construction is the same as HDFS – there is
a master node that controls the entire process
and there are nodes – managers – these are
daemons running on each of the servers on
which data is processed, and with the known
availability of resources on this server, you
can run any task.
You can try to break them down into
data management categories, that is, into
files, resources, and allocating computing
resources. A framework is the organization
of a calculation, the launch of processing
tables or other programs – the code of which
the user does not write.
Hadoop usually runs on a powerful
computing platform using a lot of memory
at startup, so if memory is low on the server,
Hadoop can be run but complex calculations
can’t be done. But if there are many small
projects, all the data of these projects can
be combined within one Hadoop cluster.
Hadoop works well with low power servers
and can extend their lifespan.
Adding deep learning to the system
with the construction of clusters with the
ideology of neural networks removes all
difficulties and takes storage and processing
of huge amounts of information to a new
level, otherwise a deadlock is inevitable.
The persistent need to consider large-
scale graph-structured data in machine
learning and data mining (MLDM) is a
significant challenge. Since the sizes of
datasets decay, statistical theory suggests
that we should use more complex models
to eliminate the influence of simpler models
and determine more complex signals from
the data. At the same time, the computational
and storage complexity of large models,
combined with rapidly growing datasets,
has exhausted the limit of single-step
calculations.
2. Limitations of service-oriented
architecture and its combination with
cloud computing
Cloud computing is dynamically
scalable resources and SOA (Service
Oriented Architecture) is the concept of
loosely coupled services. Each service is
independent of the other. Together they can
form a complete system. We will give an
overview of cloud computing and SOA, and
also propose a decision that SOA should be
combined with the cloud in order to eliminate
the limitations of SOA. Combining the cloud
with SOA will increase the availability and
reliability of SOA and reduce messaging
costs.
Different documents have different
definitions of Service Oriented Architecture
(SOA). From an architectural point of view,
a service is defined as a way to access the
functionality of a system or any individual
function. Some standard interfaces are
used to access these features, and there are
also some predefined rules for accessing
these services, which are set out in the
service description. The service is defined
in terms of its elements as an organization
for the promotion of structured information
standards. SOA is based on loosely
connected software parts; each software
component provides an individual service.
The fundamental idea behind SOA is open
access and encapsulation. Although SOA
does not include a cloud architectural style,
in order to get the maximum benefit from
SOA, we must combine service-oriented
architecture with cloud computing. This will
help us overcome the limitations of SOA.
The main concept of the cloud is
53
Моделі та методи машинного навчання
fast delivery and resource scalability. The
resources offered by cloud computing are
dynamically scalable. Another advantage of
cloud computing over traditional computing
is its low cost and location independence.
Through SOA, different components of
application development can be integrated.
New features can be added to existing
applications or can be retained, but very little
research has been done on SOA maintenance
tools. Data can be collected with the system
by deploying it in the environment in which
it is intended to operate. The collected data
is analyzed [10].
From an engineering point of view,
a service-oriented architecture has the
following advantages:
• Language-independent integration:
Services in a service-oriented architecture use
the XML standard. It focuses on converting
data generated in XML format and passing it
to another component.
• Multiple components: SOA is
based on the concept of loosely coupled
components; once these components are
developed, they can be used separately.
These components can be used with proper
reliability and safety assurance. In addition,
these services can be combined together to
create a new system with higher capabilities.
• Rapid Application Development:
Offered by components that are developed
using a Service Oriented Architecture;
meet some of the organization’s business
requirements. These blocks can be used
separately and later they can be quickly
integrated.
• Definition of existing system service
and setting standards. With Service Oriented
Architecture, we can integrate new systems
with old legacy systems without rewriting
the new system. This will save capital costs
as well as time. Thus, organizations without
the overhead of developing new systems
from the core.
There are many reasons for the
adoption of service-oriented computing
in enterprise environments. It’s just that
the services provided by service-oriented
computing are reusable and flexibly
integrated into new services.
The cloud computing model deals
with dynamically scalable resources. These
resources are provided over the network.
There are many other benefits that can be
taken from cloud computing. This reduces
hardware costs, maintenance costs, and the
cost of installing access to hardware and
software in the cloud. In addition, it promotes
resource agility, scalability, and reliability.
Locations and devices are independent,
adaptive and resilient. There are three unique
cloud layers depending on the resource type.
Software as a Service (SaaS) is the
most popular layer that provides users with a
ready-made program. It ensures that the user
will use the Internet host software without
using client resources such as installing and
running applications on the client’s local
computer. Each data item has a read lock or
a write lock, and there is a distributed cloud
consistency and convergence mechanism.
Platform as a Service (PaaS) is the
second layer that contains the environment
for running the software. An application
server can be one that allows developers to
deploy a web application without purchasing
or configuring their actual servers. The
purpose of this model is to ensure data
protection, which is very important in an
environment where we consider storage as a
service. To provide a load balancing service,
it is important to ensure safety against
shutdowns.
Infrastructure as a Service (IaaS) –
This layer shows the exchange of hardware
resources to perform services, classically
using virtualization technology. With this
approach, many users can use the available
resources and the resources can be increased
Fig. 1. Service-oriented architecture.
54
Моделі та методи машинного навчання
on demand. Resources in IaaS are virtual
machines that must be managed.
Radar Data cloud computing is
implemented using a hybrid or private cloud.
Hybrid clouds: types of clouds that have
some of the characteristics of public clouds
and some of private ones; so we can say that
it is a cloud that has a mixture of properties
and belongs to the public or a private
cloud. Organizational responsibilities are
often shared between cloud providers and
developers. This cloud infrastructure is more
agile in handling critical processes, as the
user can store their important sensitive data
in the private cloud and use the public cloud
for normal routine services.
Overlapping of CLOUD and SOA
functions. SOA has many benefits, but
there are also some limitations. The very
first problem that SOA faces is the use
of an inefficient XML messaging format.
To a certain extent, service-oriented
architectures and cloud computing are
related. A service-oriented architecture
provides an architectural template. Whereas
cloud computing offers highly scalable,
dynamic resources and a flexible platform
for a service-oriented architecture. SOA and
cloud computing can exist simultaneously,
supporting and balancing each other. One of
the main advantages of cloud computing is
the execution of the same request on multiple
servers, resulting in low communication
costs. Cloud computing and SOA are mutual.
So, everything that happens in a service-
oriented architectural environment will
be sent as an event to the cloud. The event
can be a data transaction or a user service
request. The request can be used for any
hardware resource or data. Therefore, it will
be easy to add a new service to the program,
and resources will become more scalable.
A solution to one of the limitations of
SOA is proposed. It uses the XML format for
messaging, which consumes a lot of network
bandwidth. Combining Cloud with SOA,
given the common points, will increase the
reliability, availability of SOA and reduce the
cost of its messaging. The use of Artificial
Intelligence will lead to the development of
a collaborative service and the development
of new computing architectures [11].
3. Virtual machines for mobile radar
computing in the cloud environment
Virtual machines have become an
attractive approach for the radar computing
platform, as applications running on virtual
machines are isolated from each other to
counteract the propagation of failures and
security, but can simultaneously use physical
machines. An important property of a virtual
computing platform is how quickly it can
respond when a request changes. That is,
we can talk about the reaction of the virtual
computing system to reassign resources as the
flexibility of the platform to rebuild when the
task changes. We are talking about a hosting
Fig. 2. Three basic cloud services
Cloud
SaaS
PaaS
IaaS
Applica�on browser
Web Services
Access technologies
Fig. 3. SOA infrastructure
Fig. 4. SOA message transfer
XML HTPSOAP Presentation TCP IP Network Physical
NETWORK
NetworkPhysical IP
SERVER
Physical Network IP TCP Presentation HTP SOAP XML
CLIENTCLIENT
55
Моделі та методи машинного навчання
service provider environment where the
entire platform is under the control of a single
administrative domain and applications often
create clusters at the application level. In this
work, resource reassignment mechanisms in
these applications are explored from the point
of view of flexibility and a new mechanism
that uses the properties of the virtual utility
of the computing platform.
This new mechanism uses haze
virtual machines (VMs) that participate in
application clusters but do not process client
requests until they are enabled by system
resource management. We’re evaluating
this, along with other mechanisms in the
service computing testbed. The results
show that this VM approach outperformed
other approaches in terms of flexibility and
allowed a new virtual machine to be added
to an existing application. This approach,
increasingly used in hosting platforms, has
been confirmed by the observation that a
virtual machine can quickly recover from a
suspended state, provided that the suspended
VM remains in memory. Unfortunately,
servers typically implement a particular
application in a software cluster; on the
application servers on client machines, the
cluster will be considered unhealthy and, as
shown later, it will take a long time to re-
enter the cluster when reactivated.
So, we propose to have a small
margin of virtual machine resources on each
physical machine that remains active but
disconnected from the Internet. Since they
are not available to clients, these virtual
machines will not serve requests. Therefore,
they are sometimes called ghosts because
their existence is invisible to receiving and
redirecting the content of client requests.
Ghost VMs consume a central cluster
management resource. But CPU time is
negligible because only a small fraction of
the ghost machines are kept on each physical
machine. Remapping the source only
constitutes a reconfiguration of the content
selector.
By migrating a virtual machine from
an overloaded host to a relatively free host,
we can achieve better load balancing and
use our resources more efficiently. There are
several ways that a migration can be done.
1. Migration of a virtual machine can
be achieved by stopping the virtual machine
on one host and booting a previously saved
mirrored virtual machine on a different host
machine. In this case, it is the destination
virtual machine that does not display the
current state of the source VM.
2. The virtual machine of one host can
be cloned to another. This method requires
first suspending the VM on the output host,
then copying it between hosts, and finally
restoring it on the target host. The source
VM state will be preserved, but this method
requires moving the VM on machines in the
critical path.
3. VM migration methods have been
developed, where the source virtual machine
is not stopped and the source of information
is updated. The state of the virtual machine
is transmitted to the target host when
running. This method drastically reduces
downtime during migration, but by itself
does not prove the flexibility of reallocating
resources. First, the target host will only be
able to start doing its job after a delay similar
to cloning technology. Secondly, the egress
machine will not receive overload relief due
to the operation of the outgoing VM until the
migration is complete.
Promotion of the VM from ghost
to active state can be combined. In most
cases, most of the delay occurs just before
the network switch reconfigures. This VM
forms the basis for a flexible reallocation of
resources. At the initial stage, an active list
of virtual machines is formed. This is enough
to serve the client’s requests. If we migrate
a virtual machine by stopping it, it is in the
ghost list.
We have considered an important
issue of virtualization of utility computing
platforms – the need to quickly respond
to changing requirements. We mean the
response time of a platform as its flexibility.
We are targeting a provider environment
hosting utility where the entire management
platform for a single administrative domain
and application resides, often clustered at the
program level. Ghost virtual machines are
used that contain running application servers
and are part of a cluster with active virtual
machines. Considered a more advanced test
56
Моделі та методи машинного навчання
client – tools for this environment, in addition
to load generators, in more representative
test programs to understand the time of
rapid redeployment of resources in the
application computing platform and a better
understanding of the underlying resources.
The concept of ghost VMs introduces
a hierarchy for a virtual machine on a radar
computing platform: active ghost, paused,
and not booted. This hierarchy provides more
opportunities for global and local resources,
control algorithms.
The new vision of mobile computing
relieves mobile devices of serious resource
constraints, allowing resource-intensive
programs to use cloud computing without
delays, jitter, congestion and failures in the
global network. Through the efforts of many
researchers, basic concepts, methods and
mechanisms have been developed to provide
the basis for this area of rapidly developing
informatics [1]. Mobile computing enhances
cognition through computationally intensive
capabilities such as computer vision and
graphics, machine learning, planning and
decision making.
A cloud-based mobile computing
strategy using a temporarily configured
infrastructure as the mobile device moves
with its user in the physical world. The crisp
interactive response needed to seamlessly
expand human cognition is easily achieved
in the architecture due to the physical
proximity of the cloud and network latency
per clock. The use of a cloud package also
makes it easier to meet the peak throughput
of multiple users who interactively generate
and receive radar imagery, high-definition
video, and high-resolution images. Rapid
infrastructure setup for various applications
is becoming an important requirement.
Lack of resources is a major hurdle
for many applications to seamlessly expand
human cognition, as such programs typically
require processing and energy far beyond
what mobile hardware can provide.
The obvious solution to the problem
of resource scarcity of mobile devices is the
use of cloud computing. The mobile device
can execute a resource-intensive program
on a remote high-performance computer
server or computing cluster and support user
interaction with the thin client program over
the Internet.
In wireless networks, a common
power-saving technique is to turn on the
mobile device’s receiver only for short
periods of time to receive and acknowledge
packets buffered at the base station, which
increases the average end-to-end packet
delay as well as jitter. On the other hand,
throughput is unlikely to be affected by these
methods, since it is an aggregate, not an
instantaneous figure. While throughput will
continue to improve over time, latency is
unlikely to decrease rapidly.
Wireless LAN throughput is typically
two orders of magnitude greater than the
wireless Internet bandwidth available to a
mobile device. For instance the nominal
throughput of the fastest currently available
wireless LAN (802.11n) and HSPDA 400
Mbps wireless Internet and 2 respectively.
From a user experience standpoint, the
difference in transmission latency across
these bandwidths can be very significant:
80 milliseconds instead of 16 seconds for a
4MB JPEG image, a huge difference for deep
immersive programs.
Instead of relying on a remote «cloud»,
we could bridge the resource gap of a mobile
device with the help of a nearby resource-
rich cloud. Thus, we could meet the need for
real-time interactive response through low
latency, one-click, high bandwidth wireless
access to the cloud packet.
Conclusion
The mobile device works as a thin client,
and all significant computing takes place in the
neighboring cloud. The physical proximity of
this cloud is important: the end-to-end response
time of applications running on it must be fast
(a few milliseconds) and predictable. If there is
no cloud nearby, the mobile device can flexibly
switch to a fallback mode that includes a
remote cloud or, in the worst case, only its own
resources. Full functionality and performance
may return later when the device detects a
nearby cloud [12,13].
The virtual machine approach is more
reliable than process migration or software
virtualization. It is also less restrictive
and more general than language-based
57
Моделі та методи машинного навчання
virtualization that requires programs to be
written in Java or C#. Another approach is
called dynamic VM synthesis. The mobile
device provides a small VM add-on to the
cloud software infrastructure that already has
the underlying VM from which the add-on
was derived. Summing up, we can say that the
topic of convergence of mobile applications
in portable devices with cloud computing has
become a revolution in the organization of
distributed computing in the direction of closer
interaction between hardware and software,
the use of elements of artificial intelligence,
and deep learning of a neural network.
References
1. Y. Simmhan, B. Plale, D. Gannon. A survey
of data provenance techniques 2005. Com-
puter Science Department, Indiana University,
Bloomington IN 47405, 69
2. D. Gannon. Theoretical Problems in the De-
sign of Tools to Aid in the Construction of Par-
allel Programs. 1989. Opportunities and Con-
straints of Parallel Computing, 39-48
3. A framework for collecting provenance in
data-centric scientific workflows 2006. IEEE
International Conference on Web Services
(ICWS’06), 427-436
4. F. Bodin, P Beckman, D Gannon, S Narayana,
SX Yang. Distributed pc++ basic ideas for an
object parallel language, 1993. Scientific Pro-
gramming 2 (3), 7-22.
5. C. Catlett, W.E. Allcock, P. Andrews, R. Aydt,
R. Bair, N. Balac, B. Banister, ... Teragrid:
Analysis of organization, system architecture,
and middleware enabling new types of appli-
cations. 2008. High performance computing
and grids in action, 225-249
6. Kevin Echton. (2009, June). Things about the
Internet of Things In the real world, things
matter more than ideas. http://www.rfidjourn-
al.com/article/print/4986
7. Pankech Patel, Animech Pyatak, Tiago Teich-
ery « On the way to the development of sup-
plements for the Internet of speeches», New
York, USA. 2011, p.5:1-5: 6. http://doi.acm.
org/10.1145/209.190.2093195
8. Pol Zikopulos. Big Data IBM: What is Big
Data., pat 1 and 2. 2012. http://www.youtube.
com/watch?v=B27SpLOOh
9. Е. Akerman and Е. Huizu, «5 technologies,
what formed net», Spectrum, IEEE, vol. 48, p.
40-45, June 2011 р.
10. K. H. Bennett, V. T. Rajlich and N. Wilde,
“Software Evolution and the Staged Model of
the Software Lifecycle”, Advances in Comput-
ers, Volume 56, Academic Press, pp. 1 – 54,
2002.
11. Gaoyun Chen, Jun Lu and Jian Huang, Zexu
Wu, “SaaAS – The Mobile Agent based Ser-
vice for Cloud Computing in Internet Environ-
ment. Sixth International Conference on Natu-
ral Computation, ICNC 2010, pp. 2935-2939,
IEEE, Yantai, Shandong, China, ISBN: 978-1-
4244-5958-2, 2010.
12. Nancy Wilkins-Diehr, Chaitan Baru, Dennis
Gannon, Kate Keahey, John McGee, Marlon
Pierce, Rich Wolski, Wenjun Wu: “Harness-
ing clouds and software services for science”.
Journal “Cloud Computing and Software Ser-
vices”, 2010, p.17.
13. M. P. Papazoglou, P. Traverso, S. Dustdar and
F. Leymann, “Service-Oriented Computing
state of the art and research.
Received: 14.04.2023
About the authors:
Mykola Коsovets
Leading Constructor,
Number of scientific publications in
Ukrainian publications -56,
Number of scientific publications in foreign
publications -17,
Index Girsh – 5
https://orcid.org/0000-0001- 8443-7805
Scopus Author ID: 5644007500
Lilia Tovstenko
Leading Software Engineer,
Number of scientific publications in
Ukrainian publications -24,
Number of scientific publications abroad -8.
Index Girsh -7.
Place of work:
Mykola Коsovets
SPE “Quantor”, Chief
03057, c. Kyiv-57, str. Е. Potye, 8-А
58
Моделі та методи машинного навчання
Ph.: (380)66-2554143
E-mail: quantor.nik@gmail.com
Lilia Tovstenko
Glouchkov Institute of Cybernetics of
of the National Academy of Science of
Ukraine, 03187, Kyiv-187, Academician
Glouchkov Avenue, 40.
Ph.: (380)67-7774010
E-mail: 115lili@incyb.kiev.ua
|