M3: Machine-to-Machine Management Framework
Heikki Mahkonen
1
, Tony Jokikyyny
1
, Jaime Jimenéz
1
and Sławomir Kukliński
2
1
Ericsson Research, Kirkkonummi, Finland
2
Orange Labs, Warsaw, Poland
Keywords: Machine-to-Machine, Internet of Things, Cognitive Network Management, Distributed Hash Tables,
Publish Subscribe Networking.
Abstract: The number of deployed sensor devices with Internet connection is expected to exceed 50 billion units.
Many of these devices spend most of their time in sleep mode to conserve energy. This sets new kinds of
requirements for network management, and creates the need of redesigning conventional network
management. Hence, most of the manual deployment, configuration and operation tasks need to be
automated in a scalable fashion, using protocols that can deal with the uncertainty caused by the intermittent
nature of the devices. For scalability reasons, the network management logic needs to be distributable in the
network management architecture. In this document we describe our management framework for M2M
networks. It is also shown, how the framework has been implemented as a prototype testbed. We have used
the testbed to study centralized and de-centralized M2M network management logic for different
management scenarios.
1 INTRODUCTION
In future networks the amount of users and M2M
devices are growing. This sets new requirements for
network management. Typical users do not have the
expertise to deploy and configure M2M networks by
themselves. Neither ISPs have the manpower to
manually configure all the expected Machine-to-
Machine (M2M) network deployments. For this
reason, new automated and scalable ways of
deploying and managing network equipment are
required.
This work is related to a CELTIC project, called
COgnitive network ManageMent under UNcErtainty
(COMMUNE) (COMMUNE, 2013), which studies
cognitive network management under uncertainty.
The main goal of the project is to design a network
management architecture that can distribute
programmable network management algorithms to
different parts of the network. In a typical case, these
algorithms are cognitive in nature, allowing them to
learn and adapt to changes in the environment where
they are running. We have implemented a M2M
management testbed as part of COMMUNE work.
The purpose of the testbed is to study different
network management algorithms in a real M2M
networking environment.
In this paper, we describe the current state of our
testbed and evaluate some of the implemented
network management functionality. The paper is
structured as follows. In Section 2 we give
background on M2M networks, and current work on
network management. Section 3 discusses our
cognitive M2M network management framework.
Section 4 describes the testbed implementation and
experimented scenarios. Finally, Section 5 concludes
the work.
2 BACKGROUND
2.1 M2M Networks
European Telecommunications Standards Institute
(ETSI) describes a high level architecture (TS-
102.690, 2011) for M2M networking and for
Internet of Things (IoT) services and applications. In
this architecture, a functional split can be made
between the constrained M2M devices, a
middleware layer with more logic and processing
power, and an IoT service and the application layer.
M2M devices may use multiple different
communications protocols e.g. 6LoWPAN
(Montenegro, 2007), CoAP (Z. Shelby K. H., 2013).
139
Mahkonen H., Jokikyyny T., Jimenéz J. and Kukli
´
nski S..
M3: Machine-to-Machine Management Framework.
DOI: 10.5220/0004806501390144
In Proceedings of the 3rd International Conference on Sensor Networks (SENSORNETS-2014), pages 139-144
ISBN: 978-989-758-001-7
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
The main differentiator between M2M device and
any other device is that M2M device is a battery,
memory, and CPU constrained device, serving a
certain predefined purpose. Device management for
such a device is required to enable automated
configuration and management of the service.
To connect M2M devices to Internet services
different kinds of M2M middleware services can be
used e.g. CoAP, HTTP REST (Fielding, 2000), or
CoRE Resource Directory (Z. Shelby S. K., 2013).
Management functions for the M2M middleware
need to deal with configuration of the Personal Area
Network (PAN), device identification and lookup as
well as providing semantic descriptions of devices.
Internets of Things (IoT) service platforms are
handling service and application capabilities.
Typically, it provides users a web service interface,
through which users can view M2M device data and
use M2M devices. For scalability reasons, IoT
services are often implemented into a cloud
platform, e.g. OpenStack (OpenStack, 2013).
2.2 Network Management
The Telecommunication Standardization Sector of
the International Telecommunications Union (ITU-
T) (M.3400, 2002) has defined a model and
framework for network management, called FCAPS.
This model has been widely used as a basis, when
designing network management frameworks and
protocols.
SNMP is an Internet Engineering Task Force
(IETF) standard for managing devices on IP
networks (Rose, 1991) (Case, 1990). The main
problem of using FCAPS and SNMP in M2M, is the
scalability when the volume of devices and gateways
can reach up to 50 billion units.
2.3 Self-Organizing Networks
The problems with classical network management
have already been identified about 10 years ago,
generating the concept of autonomic or self-
organized networks. We went through some of the
most interesting work; however, as we noticed in
these activities the problem of M2M and IoT
management has been ignored.
The aim of EFIPSANS FP7 project (EFIPSANS,
2009) was to expose the features in IPv6 protocols
that could be exploited or extended for the purposes
of creating autonomic networks and services. The
project implemented autonomic networks and
services through a Generic Autonomic Networking
Architecture (GANA). This approach seems to be
too complex to use it for M2M management.
The 4WARD FP7 (Ghader, 2009) project has
similar objectives as EFIPSANS, although it is not
focused on IPv6. The approach is also less
hierarchical. The idea of this In-Network
Management (INM) system is to execute
management functions on its own.
The Self-Organizing Networks (SON) solutions
for cellular networks are currently being defined in
the 3rd Generation Partnership Project (3GPP)
standardization (TS-32.500, 2013). The problem
with SON is that it is focused on plug-and-play
deployment of new 3GPP radio base stations and
therefore cannot be used in other networks.
2.4 Distributed Management
The nature of M2M calls for distributed network
management. There are already several distributed
management examples, such as (G. Goldszmidt,
1995) allowing the distribution of management, and
(Waldbusser, 2006) extending the functionality of
SNMP’s MIB. More recent work provides a
management framework for a distributed Machine-
to-machine network, using Chord, (Y. Peng, 2012)
(I. Stoica, 2001). Chord provides very efficient
lookup with a key-based routing system (P.
Pietzuch, 2007). A similar example is the SNMP
Usage for RELOAD (Gupta, 2012) that uses
RELOAD as lookup mechanism for SNMP.
One of the advantages of having decentralization
in management logic is that it enables management
while some devices are offline. Another advantage is
that distribution can provide autonomous monitoring
by delegating some of the monitoring tasks into the,
often very powerful, monitored devices themselves.
Finally, in scenarios with high churn a distributed
approach would ensure the scalability of network.
The current management systems are not
completely distributed. For example, in order to
authenticate the devices and their operations on a
distributed network, there is usually a central
enrolment server serving as a trust anchor and
Certification Authority (CA) for the whole overlay.
In this work we propose a new kind of network
management framework for M2M and IoT service
management that is autonomous and distributed. The
autonomous features minimize the required human
intervention. The distribution of the management
logic and signalling enable the framework to be
scalable even in most demanding network scenarios.
SENSORNETS2014-InternationalConferenceonSensorNetworks
140
3 M3: FRAMEWORK
In this section we describe our M3 framework, based
on the GARSON model (S. Kukliński, 2012).
GARSON is a generic paradigm that allows for
centralized and distributed management operations.
It divides the network into smaller management
domains. Moreover, it suggests a way in which the
management functions should be decomposed, what
provides profits related to reusability of components
and more systematic and simpler network
management programmability.
Figure 1: GARSON architecture for M2M and IoT
Network Management.
Figure 1 shows how the hierarchy proposed by
GARSON is applied to M2M management. We have
defined two management domains for M2M
networking and IoT services. Common Inter Domain
Level (IDL) can handle bot domains. In the Top
Level Management (TLM), the proxy server
provides interfaces for system administrator and
users to control the M2M management using policy
updates.
3.1 Distribution in M3 Framework
The M3 framework allows both centralized and
distributed operations simultaneously. Centralized
network management is required in some use cases
e.g. authentication, security, accounting, and
bootstrapping of distributed features. Most of the
network management should, however, be run in
distributed and autonomous fashion.
In GARSON, distribution of the management
decisions can be implemented in the intra-domain
and inter-domain levels of the architecture.
Distribution of management data and commands
between these autonomous management logics can
be done either with DHT (H. Balakrishnan, 2003) or
publish/subscribe (P. Jokela, 2009) networking.
To support the programmability and
functionality update, a suitable execution
environment is needed that can handle distribution
of execution modules. In our framework we used
e.g. Java OSGi (OSGi, 2013). Due to this approach
new management functions can be easily added or
existing updated on fly.
3.2 GARSON System Model
Figure 2 shows the general system model for
autonomic and cognitive management that we used
in our work.
Figure 2: System model of autonomic and cognitive
network management.
This model provides management system
decomposition into internal layers to provide
monitoring (AMON), knowledge based autonomic
reasoning (KNOW), cognitive reasoning (COG)
logic, and actuation (ACT) in different network
elements. In addition, the functionality of these
processes can be controlled through policies.
Components of each layer can be implemented in
different networks nodes and intra-layer
communication mechanisms are provided in order to
achieve the assumed goals:
AMON: A programmable monitoring
functionality, needed to produce network
management input for management decision logic.
KNOW: Knowledge base reasoning algorithms
using input from AMON and producing decisions
ACT.
COG: Cognitive functions e.g. machine learning
required to autonomously learn from previous
actions.
ACT: An actuation layer is required to format
reasoning decisions into management commands.
POLICIES: Administrators can control the
management through policies.
3.3 Communication Models
Figure 3 a) shows a traditional manager-agent
communication pattern used for centralized network
M3:Machine-to-MachineManagementFramework
141
management. In this model, agent is the managed
entity and the manager implements the management
function. To enable distribution of the management
reasoning logic and functions communication
without the need of direct connection a
publish/subscribe overlay can be used. This option
enables agent discovery via the overlay, still having
the MIB within the manager entity Figure 3 b).
Figure 3: Some possible network topologies: (a)
Centralized approach, (b) Multiple Distributed Agents, (c)
Multiple Distributed Managers, (d) Fully decentralized.
In a similar way, multiple managers can manage one
or more agents via a P2P overlay, without the need
for establishing and maintaining direct connections
between them, and having the MIB distributed only
among the managers Figure 3 c).
Finally, to solve the fully distributed
management scenario, an overlay of peers with the
functionality of both managers and agents can be
used, see Figure 3 d). In this scenario, peers will
forward and route their messages via each other. An
extra logic is required on the peers to establish the
preference, when sending commands to each node.
4 M3: TESTBED
A testbed was created to enable experimentation of
M2M network and IoT service management
algorithms. Figure 4 shows overview of the testbed
architecture.
The M2M devices in our testbed are Arduino
(Arduino, 2013) devices with various kinds of
sensors and actuators. Each device is connected to
the M2M network wirelessly.
The M2M gateway is running a Java OSGi
framework. The gateway handles M2M
communication over Digi XBee (Digi, 2013). In
addition, it connects to publish/subscribe and DHT
management frameworks.
We implemented the DHT and publish/subscribe
overlay using Hazelcast (Hazelcast, 2013) in the
Figure 4: Overview of the cognitive network management
testbed.
testbed. The Hazelcast API provides easy to use API
for DHT implementations. In addition, it provides a
Topic posting feature that was used to implement the
publish/subscribe mechanism.
The IoT service platform in the testbed is an
OpenStack cloud platform (OpenStack, 2013). In
this environment, we have a capability to instantiate
virtual machines in real time depending on our
needs.
The access controller functionality is in charge of
connecting M2M devices to management framework
by configuring them with publish/subscribe
information and bootstrapping DHT.
Communication to it is secured with Generic
Bootstrap Architecture (GBA) (TS-33.220, 2013).
4.1 Device Description Management
The device description management is needed to
identify and configure M2M. We use the
publish/subscribe channel to connect attaching M2M
devices to different management entities. The
management entities can be added and removed by
subscribing and unsubscribing them. This
architecture provides an easy way to introduce new
management functionality without updating the
software of a management server. Different users
may have different requirements for the
management of M2M devices, e.g. ISP, device
vendors, users, etc.
Figure 5: Device Description Management.
Figure 5 shows that the AMON functionality is
distributed into M2M gateways that collect device
information. The device identity that is a 64-bit
serial number is published to a well-known
SENSORNETS2014-InternationalConferenceonSensorNetworks
142
management channel that connects the information
to KNOW logic built into management entities. The
management entity maps the device identity to a
device description and publishes the information to
the channel as ACT command. The M2M gateway
that is responsible for managing the device will
receive the device description and actuate the
management commands based on it.
Figure 6: Network management publish subscribe channel
for device management.
As shown in Figure 6, the publish/subscribe
interface supports subscription from multiple
network management entities. These entities can
have a single network management function that
they are responsible for. If there are multiple
management functions, the gateway gets a number
of device description fragments as a reply to a
device description request. The device description is
then constructed by combining the fragments.
4.2 WPAN Coordination
We used the DHT implementation to cluster the
management information from M2M gateways that
reside in particular geographic area. This offers us a
way to distribute some management tasks as the data
required is available in the DHT e.g. WPAN
coordination.
Figure 7: Wireless Private Area Network Management.
Figure 7 shows the monitoring information we can
store into the DHT. The AMON layer is shared by
each gateway through the DHT based on
geographical location. Each gateway implements the
KNOW layer and has a capability to calculate local
optimum for their WPAN configuration as they
know the network configurations around them. The
ACT layer is implemented inside each gateway and
acts according to the output from KNOW layer.
Figure 8: M2M network emulation GUI.
In our testbed each gateway stores the WPAN
configuration into the DHT with a key that has
geographical significance. This way the stored
information also maps to a location or area. By using
the same key the gateways can collect all WPAN
information in their vicinity. We have implemented
a simple algorithm in all gateways to optimize
WPAN channel usage based on the stored
information. The algorithm constructs a coverage
map based on the information in DHT and selects
the channel based on the minimum overlap on other
WPANs.
An emulation interface was used to evaluate our
self-optimization functionality for WPANs. The
interface depicted in Figure 8, shows our physical
gateways and their coverage areas, as well as
emulated gateways and their expected coverage
areas. Channels are colour coordinated. The physical
gateways are separated in the map by the marker.
This setup allows us to study scalability issues in
M2M management.
5 CONCLUSION AND FUTURE
WORK
In this paper we described a new kind of network
management framework, designed for M2M
network and IoT service management. We also
described our current implementation of the
framework. The capability of our testbed was shown
in two example scenarios. Two ways of distributing
M3:Machine-to-MachineManagementFramework
143
the monitoring and reasoning logic for network
management were described.
We showed, that the management functions can
be distributed into different nodes and
interconnected using a publish/subscribe interface.
This enables multiple separated network managers
to correspond to a single signal that is published.
One future work item is to look into how SNMP
could use the publish/subscribe interface as its data
transport mechanism.
To enable independently running distributed
management algorithms, the input data needs to be
distributed. We can achieve distribution by using
DHT to store the monitored input data. Each
managed node can run the reasoning logic locally
and use monitoring information collected globally,
or from locations near to the entity. The coordination
of the distributed management logic, COG layer
support and policy control is left for future work.
REFERENCES
Arduino. (2013). http://www.arduino.cc. Arduino.
Case, J. F. (1990). Simple NetworkManagement Protocol
(SNMP). RFC 1157.
COMMUNE. (2013). http://projects.celticinitiative.org/
commune. CELTIC EU Project.
Digi. (2013). http://www.digi.com/xbee. Digi XBee.
EFIPSANS. (2009). Third draft of autonomic behaviours
specifications (abs) for selected diverse networking
environments. EFiPSANS EU Project.
Fielding, R. (2000). Architectural styles and the design of
network-based software architectures. http://
www.ics.uci.edu/fielding/pubs/dissertation/top.htm.
G. Goldszmidt, Y. Y. (1995). Distributed management by
delegation, in proceedings of the 15th international
conference on distributed computing, pp. 333-340.
IEEE.
Ghader, M. (2009). Network management architecture in
the future internet. Spain: ICT MobileSummit.
Gupta, N. (2012). Management of decentralized dht based
m2m network.
H. Balakrishnan, M. F. (2003). Looking up data in p2p
systems. In Communications of the ACM.
Hazelcast. (2013). http://www.hazelcast.com. Hazelcast.
I. Stoica, R. M. (2001). Chord: A scalable peer-to-peer
lookup service for internet appli-cations. in
Proceedings of SIGCOMM’01 Conference.
M.3400. (2002). M.3400 (02/200): Tmn management
functions. ITU-T.
Montenegro, G. K. (2007). Transmission of IPv6 Packets
over IEEE 802.15.4 Networks. RFC 4944 (Proposed
Standard). Updated by RFCs 6282, 6775.
OpenStack. (2013). http://www.openstack.org. OpenStack.
OSGi. (2013). Open services gateway initiative.
http://www.osgi.org.
P. Jokela, A. Z. (2009). Lipsin: Line-speed
publish/subscribe inter-networking. Barcelona, Spain:
in proceedings of SIGCOMM ’09.
P. Pietzuch, D. E. (2007). Towards a common api for
publish/subscribe. Canada: in Proceedings of
DEBS’07.
Rose, M. T. (1991). The simple book: an introduction to
management of TCP/IP-based internets. Englewood
Cliffs, NJ;: Prentice Hall, 2nd edition.
S. Kukliński, M. S. (2012). Garson: Management
performance aware approach to autonomic and
cognitive networks. IEEE MENS.
TS-102.690. (2011). Machine-to-machine communications
(m2m) functional architecture. ETSI.
TS-32.500. (2013). Telecommunication management self-
organizing networks (son) concepts and requirements.
3GPP.
TS-33.220. (2013). Generic bootstrapping architecture
(gba). 3GPP SON standardization.
Waldbusser, S. (2006). Remote Network Monitoring
Management Information Base Version 2. RFC 4502.
Y. Peng, W. H. (2012). An snmp usage for reload. IETF
draft.
Z. Shelby, K. H. (2013). Constrained application protocol
(coap). IETF draft.
Z. Shelby, S. K. (2013). Core resource directory. IETF
draft.
SENSORNETS2014-InternationalConferenceonSensorNetworks
144