Resource-aware Virtualization for Industrial Networks
A Novel Architecture Combining Resource Management, Policy Control
and Network Virtualization for Networks in Automation or Supervisory
Control and Data Acquisition Networks
Hans-Peter Huth and Amine M. Houyou
Siemens AG, Corporate Research, Munich, Germany
Keywords: Network Virtualization, Quality-of-Service, Automation Networks, Industrial Ethernet, Network
Management and Control, Multi-tenancy.
Abstract: Within the EU Project IoT@Work future Internet techniques such as network virtualization are investigated
in order to provide communication services in automation systems. The physical network is modeled as an
abstract pool of resources that is allocated through virtual slices to different automation applications. Each
slice of the network resources fulfils the application communication needs in terms of security, QoS, and re-
liability. The proposed architecture is targeted at constrained networks, where a centralized control and
management point could be accessed through an API to create and implement virtual networks which could
be mapped to a real physical infrastructure. This approach would impact not only the task of engineering by
separating application programming from configuring networks. It also hides the network dynamics and
self-management connecting ever more dynamic, adaptive and smart automation systems.
1 INTRODUCTION
Networks play a major role in all distributed applica-
tions and any network has to support many, some-
times very different, data flows which are potentially
planned, implemented and operated by different
parties. Consequently configuration of networks to
meet those requirements is a crucial task. In tele-
communication networks and to a lesser degree also
in enterprise networks the solution to achieve the
desired quality of service (QoS) is to construct virtu-
al networks dedicated and optimized for the respec-
tive applications which try to optimize the use of the
underlying physical network. There exists a broad
range of possible network virtualization solutions
adapted to the respective domains (Duan et al.,
2012)
In the last years, networks controlling machines
in industry automation, buildings or in Supervisory
Control and Data Acquisition (SCADA) systems are
using more and more technologies and standards
which are built upon Ethernet and Internet technolo-
gies. However, those networks face very specific
requirements which cannot easily be met by apply-
ing technologies from the telecommunication or
enterprise networks. Still, there is an ever increasing
need for interoperability driven by higher integration
between clouds, enterprise networks, on the one
hand, and the industrial networked systems on the
other. Several trends in these areas demand for new
networking concepts. First, the upcoming Internet-
of-Things will vastly increase the number of devices
and applications managed and to be supported in the
network and will demand for more agility, see also
Houyou et al., 2012. Second, the market demands
for an ever-increasing flexibility to support new
business models and to be able to quickly adopt new
technologies. In the industrial domain, however, it is
crucial to provide a reliable and deterministic opera-
tion. This is the major reason why operators of in-
dustrial networks are reluctant to change a running
system.
This work introduces a system architecture
which adapts the network virtualization and resource
control ideas to the industrial domain to achieve a
greater agility, lower management and control costs
while still fulfilling industrial requierements.
Specifically we use the industry automation as a
major focus as it has both high complexity and tight
requierments so it is a good benchmark for the novel
solution.
44
Huth H. and M. Houyou A..
Resource-aware Virtualization for Industrial Networks - A Novel Architecture Combining Resource Management, Policy Control and Network
Virtualization for Networks in Automation or Supervisory Control and Data Acquisition Networks.
DOI: 10.5220/0004508900440050
In Proceedings of the 4th International Conference on Data Communication Networking, 10th International Conference on e-Business and 4th
International Conference on Optical Communication Systems (DCNET-2013), pages 44-50
ISBN: 978-989-8565-72-3
Copyright
c
2013 SCITEPRESS (Science and Technology Publications, Lda.)
Industrial automation and control are
characterized by a great variety of communication
standard. There is still a large installation of non-IP
based so-called fieldbus systems, but in this study
we address IP-based technologies only as this is
commonly considered to be the future in automation
(see Jasperneite et al., 2009).
Requierements for automation networks differ
significantly from their counterparts in enterprise
and telecommunication. Jasperneite 2009 lists, e.g.,
low configuration costs, IEEE-802.3 compatibility,
hard real-time, and reliability requierments far
beyond multimedia for automation networks.
Another non-technical requierement has its origin in
the long life cycles and large base of installed -and
often very expensive- devices in industrial domains.
Thus any new concept must provide a clear
migration path that enables gradual modernisation
by re-using as much as possible from existing
installations. Shielding control applications from
other traffic in the LAN is requiered for reliable
operation (Martin, 2011). Devices in the embedded
domain typically also differ from enterprise or
telecommunication equipment. These devices are in
general much smaller in terms of computing power
and number of ports. Also, - as opposed to applica-
tion-based programming- many real-time related
functions are supported by ASIC implementations
and specialized operating systems. However, most
of these devices already have powerful management
features by means of standard SNMP and propietary
extensions. There is also a high cost pressure on
both network and end devices in automation. Thus
all solutions must be “lean,” i.e. they should not add
additional costs by requiering powerful CPUs or
larger on-device memory.
Furthermore, both network and applications
typically are subject of a planning step. After this
step there is the commissioning of the configurations
determined in the planning phase, followed by the
operation phase. One drawback of this approach is
that any change in the applications or in the network
may necessitate revision of the planning. If this is
the case, reconfigurations in many places might be
required. Additionally, many applications are tightly
bound to a select physical topology and specific
devices. This tight coupling of applications with the
physical network makes again any change a
challenge.
The solution presented here decouples
applications from the physical network. This is done
by providing an abstract view that is independent of
concrete technologies and that is also independent of
the physical network topology. Furthermore,
different applications can be seperated and
“shielded” from each other by means of on-demand
virtualization of the network and its resources. This
allows for independent changes in applications or in
the physical network. By adherring to this approach
one also isolates changes in a way that is not needed
to re-plan the complete automation applications if
only small changes have occured. Forming virtual
networks for applications also allows to integrate
different technologies. Our apporach has thus the
additional avantage of supporting heterogeinity and
multi-tenancy in a natural way.
Unlike telecommunication networks, our
approach cannot be realised by using overlays such
as VPNs or MPLS. The reason for this is that the
underlying network shall remain simple and well
suited for small, embedded devices. Also, the per-
packet overhead of tunneling technologies (overlays)
makes it impossible to meet ultra-fast forwarding
and switch times required in industrial-Etherent
standards. Even lean concepts like MPLS may not
be usable, since they requiere reading larger portions
of a packet before a forwarding decision can be met.
This procedure prohibits ultra-fast cut-through
forwarding found in many industrial switches (see
Prytz, 2008).
Since overlays cannot be used for virtualization
in automation networks, other means are needed.
Our approach attempts to re-use layer-2 features
such as, e.g., VLANs (see Chowdhury and Boutaba ,
2010) or Shortest-Path Bridging (see D. Allan et al.,
2010). We also consider the direct manipulation of
forwarding/routing tables in the the spirit ofOpen
Flow (see, e.g., Curtis et al., 2011). Unlike Open
Flow however, our concept allows to implement a
work split between an intelligent and central live
management tool and intelligent devices, which then
can operate at full speed and with some autonomity.
Because our architecture essentially partions a
network including resources into new virtual
networks, which -from an applications point of
view- behave more or less like physical networks,
we use the term 'slice' for a virtual network in our
architecture (see Chowdhury and Boutaba, 2010).
2 RELATED WORK
Network virtualization is not new. Well known
technologies and standards include IEEE 802.1ak
VLAN (IEEE Std 802.1Q and amendments) and
Multiprotocol Label Switching (Rosen et al., 2001).
These technologies allow marking and identifying
packets throughout the network and also
Resource-awareVirtualizationforIndustrialNetworks-ANovelArchitectureCombiningResourceManagement,Policy
ControlandNetworkVirtualizationforNetworksinAutomationorSupervisoryControlandDataAcquisitionNetworks
45
manipulating the way packets are treated (i.e. rout-
ing, QoS, security policies). This is achieved in all
underlying network elements such as switches or
routers. IP tunnelling or encrypted SSH tunnels is a
second form of network virtualization, also called
'Virtual Private Networks' (VPN). VPNs create an
overlay between peers and cannot easily be manipu-
lated by the underlying infrastructure.
Current network-virtualization approaches ad-
dress the following problems by manipulating the
route of packets between end points:
Logical separation of distributed applications or
end points over the same shared physical infra-
structure (path isolation). The result of such a logi-
cal separation is frequently referred to as 'sub-
strate'.
Allow additional security features (encryption,
access control, and authentication) and limiting
connectivity to parts of the network.
Less common, but also possible, is to tailor differ-
ent QoS dimensions (e.g. bandwidth, delay guaran-
tees, reliability needs, security, etc.) for each virtu-
al network. This is done by hiding the substrate
choices or the methods involved in establishing the
virtual network. Therefore, applications are una-
ware of the underlying resource management and
resource allocation. Both are managed on a per-
virtual-network basis.
Additionally, one needs means to (a) calculate
optimal paths through the network (traffic engineer-
ing) and (b) means to commission rules to network
elements. The latter is necessary for configuring the
virtual network. The state of the art is to rely on
costly and complex tools that are inherently bound
to a specific technology. Step (a), traffic engineer-
ing, is typically done off-line and then commis-
sioned once before network boot. There are rare
exceptions, for example the MPLS- and IP-routing-
based approach described by Hoogendoorn 2003.
Similar to Hoogendoorn work, this paper also intro-
duces a domain controller that is responsible for part
of the network. This controller also implements on-
line traffic engineering and automatic configuration
of network elements.
Together with a better management of resources
in the network and a classification of communication
needs of “factory” applications, we aim at making
the network management happen in an autonomous
way. This is done in such a way that requests for
resources are dealt with in a central entity which also
may established and/or extended virtual networks as
needed. Also, changes (for optimization purposes, or
dealing with limited failures) in the network are
hidden from both the operator and the application as
much as possible. By so doing one limits the run-
time management overhead to dealing with critical
situations (e.g., large-scale hardware failures or
system breakdowns).
However, today's solutions are pre-configured
by means of network management based on pre-
planned rules, and resource management or QoS
aspects are typically not or only rudimentary han-
dled. The granularity of a solution is typically on a
domain or application-group level. Examples in-
clude multi-tenancy networks enabling many com-
panies to share a common infrastructure or to sepa-
rate various applications such as voice-over-IP,
network management and automation. Our approach
shall support a finer-granularity mapping of services
to virtual networks. While this is of course feasible
with today's approaches, the incurred management
costs prohibit implementation and management of a
(potentially) large number of virtual networks. In
fact, today's networks do not have any interface to
applications for creating, changing, or even monitor-
ing properties of a virtual network. The proposal of
combining several Ethernet extensions and protocols
with virtualization has been addressed in previous
work. However, the proposed combination by Finn
2012 does not address the network-design issue and
the interface between applications and the network.
Rather than using a zoo of specialized protocols
another approach is to directly configure the behav-
iour of network elements. Notice that the this strate-
gy is also pursued by the OpenFlow community
(The GENI Project Office, 2008). OpenFlow how-
ever imposes a high signalling overhead, as it typi-
cally attempts to control the network on a per-flow
level. Also, OpenFlow requires new network devices
(Curtis et al., 2011). Another weak point of Open-
Flow is error handling: rearranging or recreating
paths and states may take a significant amount of
time due to a high signaling load. To meet industrial
reliability demands, very fast and locally imple-
mented failure recovery is a must.
The interface to a virtual network for applica-
tions is typically mapped to a physical or a virtual
network interface (see Fischer et al., 2013). There
are basically two strategies for achieving this map-
ping:
An end device is attached to an edge device (i.e. a
switch) and the virtualization begins/ends at the
respective edge port (i.e. port-based VLAN). In
this case, all applications on the end device will
share the same virtual network and are not be able
to detect it.
A virtual network ends in the end device and a
virtual interface provides network access. In this
DCNET2013-InternationalConferenceonDataCommunicationNetworking
46
case, applications can use a specific virtual net-
work by simply using the respective virtual inter-
face.
Building on these existing protocols and technol-
ogies, our communication service adds means for
creating and tearing down a virtual network on de-
mand, and for managing resources “behind the
scenes.”
In the state-of-the-art solutions presented above,
the network does not -or only in a very restricted
way- communicate with applications. Thus, applica-
tions that need to react to network issues such as
broken links or bandwidth problems, need to imple-
ment either their own end-to-end monitoring system
or they need to use the minimal error messages pro-
vided by the socket layer via ICMP. Our approach
on the one hand shall provide a richer interface for
network state information. On the other hand it shall
hide and repair many errors by means of resilience
mechanisms.
Network virtualization is in principle independ-
ent from quality-of-service or other traffic policies,
but in telecommunication a combination of both
already in use. One example is the use of MPLS to
create virtual networks and the use RSVP-TE to
apply QoS policies to that path (Rodriguez-Perez
and Gonzalez-Sanchez, 2007). This approach may
employ different granularities, i.e. it may operate on
aggregates or per-flow.
There are QoS concepts on different OSI layers
that are in principle independent of network virtual-
ization. For example IntServ (Wroclawski, 1997) is
based on reservations for application-layer flows. A
similar idea is used on layer-2 by the Audio/Video
Bridging (AVB) standards promoted by IEEE 802.1
(Imtiaz et al., 2009). These approaches, however,
rely on a routing or path finding mechanism that are
hard to manipulate in order to allow full-blown traf-
fic engineering similar to MPLS (i.e. assigning dif-
ferent parallel traffic to different paths).
It should be noted that today’s concepts of net-
work virtualization and network-resource control are
also driven by cloud-computing and data-center
applications. In these case, creation of virtual net-
works on-demand is an essential functionality, the
so-called 'Network-as-a-Service' (Costa et al., 2012).
3 SOLUTION ARCHITECTURE
3.1 Overall System Architecture
The solution presented here proposes, at its core, a
network-abstraction layer, the 'slice layer'
(see fig. 1).
Figure 1: Layers of Network Abstraction.
This slice layer provides a graph view of the net-
work: applications can only see edge nodes and
intelligent traffic-engineering algorithms can operate
on the graph without knowing details of the underly-
ing substrate implementation. Important architectur-
al components are clean interfaces meaning they
must provide clearly defined responsibilities and
enforce a "separation of concerns" (Laplante, 2007).
Applications (including planning and management
tools) use the 'Communication Service Interface'
(CSI) for setting up and using the network. The slice
view is provided and controlled from a central com-
ponent, the slice manager (SMGR). The SMGR is
responsible for a network, or a part of a network. If a
network virtualization (= slice) must be constructed,
the SMGR receives the necessary specifications
from the application via the CSI. The SMGR then
calculates an optimal routing sub-graph and finally
commissions the required rules to the physical net-
work using a 'driver layer'. The driver layer has
knowledge of the concrete methods for accessing the
devices and this layer also provides device abstrac-
tion for the slice view in the SMGR. The driver layer
forms the base for including a multitude of different
standards or vendor-specific device interfaces. It is
an important component for making the slice layer
agnostic to the underlying device and protocol het-
erogeneity. Unlike approaches like OpenFlow or
MPLS based networks, this layered approach does
not mandate for specific interfaces. The slice man-
ager can handle all of the interfaces simultaneously
by means of driver "plug-ins". Because of that, the
slice manager constitutes an upgrade-friendly solu-
tion.
Devices, be it end devices or network elements,
must be able to perform routing or forwarding ac-
cording to the policies defined by the SMGR. In our
Resource-awareVirtualizationforIndustrialNetworks-ANovelArchitectureCombiningResourceManagement,Policy
ControlandNetworkVirtualizationforNetworksinAutomationorSupervisoryControlandDataAcquisitionNetworks
47
architecture, we bundle the needed functionality plus
signaling interface towards the SMGR in a logical
function called 'Slice Enforcement Point' (SEP, see
fig. 2). The SEP can be implemented on that device
or it uses other means (e.g. SNMP) to remotely
perform its tasks. End devices that are not under
control of a SEP -hereafter called 'legacy devices'-
can be attached to an adjacent network element with
an SEP ('edge SEP' in fig. 2). This edge SEP con-
trols the corresponding network interfaces and en-
forces the needed policies for the legacy devices.
This solution is similar to a port-based VLAN,
which indicates a migration path from today’s
Ethernet technology toward our slice solution.
Devices that are specifically dedicated to support
the new approach will have their own local SEP
('integrated SEP', fig. 2) so that slices can origin in
that device.
Figure 2: Architecture Birds View.
3.2 The Slice Manager
The slice manager controls all devices it is responsi-
ble for and it provides the interface for managing
slices. Fig. 3 shows the internal functional architec-
ture of the slice manager.
Figure 3: Simplified Slice Manager Architecture.
The slice manager needs exposes both a manage-
ment interface and an interface for accessing devices
in the network. The management interface allows
controlling (i.e. establishing, tearing down and
changing) slices without interventions of applica-
tions. The management interface can also be ac-
cessed by planning tools or by an operator. The
interface towards the physical devices follows an
object-oriented paradigm with one proxy object per
device managed by this SMGR instance. These
proxy objects also provide a standardized abstract
view of a device and they contain drivers for access-
ing the device. For example, an industrial-Ethernet
device and a legacy-Ethernet device have different
proxy implementations and the algorithms in the
SMGR only see different device properties but
access to the devices (viz. the interface) is of the
same type for both.
In order to manage and optimize slices, the
SMGR must not only know all devices and their
capabilities, but also the topology of the network.
One way to accomplish this is to collect all neigh-
borhood information from the devices and infer the
network graph from the information collected. As
SNMP and LLDP are already frequently used in
industrial devices, this info is usually present even in
today's devices.
Notice that fig. 3 does not show an application
interface. We propose to re-use the device interface
(which uses SEP signaling) as an application signal-
ing channel for several reasons: (a) it is easy for
applications to discover a local SEP, so there is no
need for another service discovery. (b) Access con-
trol becomes easier as it already can be performed in
the SEP. (c) No need for an additional asynchronous
and multi-client interface, a fact that simplifies im-
plementation of the SMGR.
3.3 The Slice Enforcement Point
As explained before, the SEP is a functionality
bound to a device. It is responsible for signaling
towards the SMGR or the applications, and it con-
trols the network interfaces of that device. For ex-
ample, the SEP must be able to manipulate forward-
ing or routing tables and to set QoS rules. The actual
implementation can re-use existing control interface
such as SNMP (Case, 2002) or IETF FORCES
(Yang et al., 2004). However, we believe that a
small dedicated agent forming the SEP is beneficial
because it can be tailored for that purpose and it may
add local intelligence which, e.g., enables quick
recovery in failure cases without contacting the
SMGR. Notice that due to the object-oriented archi-
tecture of the SMGR, a mix of different SEP imple-
mentations is of course possible.
3.4 The Applications Interface
to Communication Services
In order to use slices, applications need a
DCNET2013-InternationalConferenceonDataCommunicationNetworking
48
communication interface (User Plane) and signaling
means (Control Plane) for attaching to a slice and for
specifying properties such as e.g. bandwidth con-
straints or QoS. For the user plane we propose the
use a virtual layer-2 interface, similar to solutions
found in server-virtualization environments. The
signaling can use the next SEP as entry point, which
is either located on a device or on an adjacent edge
device (switch or router).
Specification of a slice includes at least a specifi-
cation of the required QoS, but in order to enable on-
line traffic engineering, a specification of the traffic
is also beneficial. Notice we propose to use slices for
aggregate flows of an application and not for single
flows, although the latter is conceptually possible.
Beside pure QoS, industrial applications may
have tight reliability requirements. Further more,
some of them will have a high importance (i.e. the
safety application). Thus we propose to add a notion
of resilience and importance to the specification of
traffic requirements. A traffic-engineering algorithm
can use this information to, e.g., establish redundant
paths or to resolve resource shortages.
3.5 Use Cases
The following sample use cases illustrate how the
slice system can be used.
First, consider a data acquisition application ac-
cessing many sensors and a server collecting the
sensor data. In a planning step, a traffic matrix is
calculated and a slice is defined. The slice definition
can be forwarded to the slice manager. The slice
manager then calculates an optimal path for the slice
and commissions it to the network devices. After
that, the end devices (sensors and server) simply
attach to the predefined slice. In the 'attach' process,
a virtual network interface is created by the device
SEP and the interface is bound to the slice in ques-
tion.
A second example is an automation application
that uses a PLC for controlling actuators and reading
sensors. The application requires “hard” real-time
communication and is programmed by a planning
tool. This tool can estimate and generate the traffic
matrix and also the slice specification. The latter is
then pushed into the slice manager. Unlike today,
the planning tool does not need full knowledge of
the topology and capabilities of the network. The
slice manager in turn - knowing the properties of the
network - can construct an optimal slice, eventually
re-using specific means from one of the many Indus-
trial Ethernet standards (Jasperneite et al., 2009).
While the first two examples assume pre-
planning, dynamic slice set-up in the running system
can is also feasible. For instance, consider that a
remote support needs to access a robot for mainte-
nance. In this case the factory operator may install
means for granting access for the remote service
over the firewall. In order to facilitate this, the op-
erator installs a slice on the fly. This slice might
support best-effort communication and an upper
bandwidth limit. Additionally, this slice only expos-
es the devices needed for the remote service.
These sample use cases also illustrate that the
'normal' use of the slice system is to construct semi-
static or longer living slices for aggregates (i.e.
flows belonging to one application). In many cases
the slices will be instantiated by means of a configu-
ration/management interface rather than application
signaling.
3.6 Security Aspects
Slices or network virtualization in general is not a
security concept; rather it is a complements security
tools for constructing application 'sand boxes'. Fur-
thermore, the proposed approach allows not only to
dynamically grant or deny access, but it also sup-
ports 'on-demand' control of the resources that a
certain slice consumes. By virtue of this control the
slice manager prevents DoS attacks (see also the
third use case above). We believe that this is an
essential property for supporting multi-tenancy and
applications sharing the same infrastructure.
4 CONCLUSIONS
The work presented specifies a system architecture
for network virtualization in industrial networks.
Concepts from Enterprise and Telecommunication
have been mapped and adopted to fulfill industrial
requirements. A domain controller, the Slice Man-
ager, directly manipulates a potentially heterogene-
ous network while providing a simple abstract view
to planning and management applications. Applica-
tions run in "slice" which is a virtual network with
clear QoS guarantees and bandwidth policies.
Ongoing research focuses on three issues:
Scalability: our current approach assumes one slice
manager. In order to cover larger physical areas
(signaling lag!) or larger numbers of devices, we
investigate the use of multiple domain controllers,
which, of course, need to co-ordinate their work.
Optimal use of Industrial Ethernet features: current
industrial standards provide many advanced resili-
Resource-awareVirtualizationforIndustrialNetworks-ANovelArchitectureCombiningResourceManagement,Policy
ControlandNetworkVirtualizationforNetworksinAutomationorSupervisoryControlandDataAcquisitionNetworks
49
ence features, sophisticated QoS beyond standard
Ethernet, and many other features optimized for
industrial requirements. Re-using these capabilities
in an optimal manner is another challenge for fu-
ture work.
Traffic Engineering adapted to the slice system:
considering the specific capabilities and require-
ments in this industrial domain, TE algorithms
known from other areas must be analysed and
adapted.
Further more, implementation issues and stand-
ardization have to be considered. The first topic is
addressed by a Linux-based proof-of-concept, which
forms a test bed for future enhancements. Concern-
ing standardization, the authors believe that engag-
ing IETF Forces or the OpenFlow community may
be a way to proceed, but also Industrial Ethernet
standardization must be addressed.
ACKNOWLEDGEMENTS
This work has been carried out in the European FP7
funded research project IoT@Work.
We would like to thank Jo Walewski for his use-
ful review and helpful discussion.
REFERENCES
Houyou A., Huth H.-P., Kloukinas C., Trsek H., Rotondi
D., "Agile Manufacturing: General Challenges and an
IoT@Work Perspective", 17th IEEE International
Conference on Emerging Technologies and Factory
Automation (ETFA 2012); 09/2012
Duan Q, Yan Y., Vasilakos A., "A Survey on Service-
Oriented Network Virtualization Toward Convergence
of Networking and Cloud Computing", IEEE Transac-
tions on Network and Service Management, Vol. 9,
No. 4, Dec. 2012
Jasperneite J., Imtiaz J., Schumacher M., Weber K., "A
Proposal for a Generic Real-Time Ethernet System",
IEEE Transactions on Industrial Informatics, 06/2009
Hoogendoorn C., Schrodi K., Huber M., Winkler C.,
Charzinski C., "Towards Carrier-Grade Next Genera-
tion Networks", Proc. ICCT 2003, Beijing, China, Apr.
2003
The GENI Project Office, "The GENI System Overview",
Document ID: GENI-SE-SY-SO-02.0, September 29,
2008, Cambridge, US, geni.net
Rosen E., Viswanathan A., Callon R., "RFC 3031: Multi-
protocol Label Switching Architecture", IETF, 2001
McKeown N., Anderson T., Balakrishnan H., Parulkar G.,
Peterson L., Rexford
Curtis, A., Mogul, J., Tourilhes J., Yalagandula P., Sharma
P., Banerjee S., "DevoFlow: Scaling Flow Manage-
ment for High-Performance Networks", SIGCOMM
11, August 15-19, 2011, Toronto, Canada
Martin K., "Network Infrastructure in Large, Integrated
Control Systems", Cement Industry Technical Confer-
ence, 2011 IEEE-IAS/PCA 53rd
Costa P., Migliavacca M., Pietzuch P., Wolf A. 2012.
"NaaS: network-as-a-service in the cloud." Hot-
ICE'12. USENIX Association, Berkeley, CA, USA
Case J., "Introduction and Applicability Statements for
Internet Standard Management Framework", RFC
3410, IETF, Dec. 2002
Yang L., Dantu R., Anderson T., Gopal R., "Forwarding
and Control Element Separation (ForCES) Frame-
work", RFC 3746, IETF, April 2004
Wroclawski J., "The Use of RSVP with IETF Integrated
Services", RFC 2210, IETF, Sept. 1997
Imtiaz J., Jasperneite J., Han L., “A performance study of
Ethernet Audio Video Bridging (AVB) for Industrial
real-time communication”, Proceedings of 12th IEEE
International Conference on Emerging Technologies
and Factory Automation, Palma de Mallorca, Spain,
22-25 Sept. 2009
Chowdhury N., Boutaba R., “A survey of network virtual-
ization”, Computer Networks, 2010 – Elsevier, Vol-
ume 54, Issue 5, 8 April 2010, Pages 862–876
Allan D., Ashwood-Smith P., Bragg, N. Farkas J., Fedyk
D. Ouellete M., Seaman M, Unbehagen P., "Shortest
path bridging: Efficient control of larger ethernet net-
works," Communications Magazine, IEEE , vol.48,
no.10, pp.128-135, October 2010.
Rodriguez-Perez F., Gonzalez-Sanchez J., "Extending
RSVP-TE to support Guarantee of Service in MPLS
",
Innovative Algorithms and Techniques in Automation,
Industrial Electronics and Telecommunications,
Springer 9/2007, DOI:10.1007/978-1-4020-6266-7_28
pp 149-154
Finn N., “Improvement in Ethernet Standards to Further
Reduce Latency and Jitter”, Presented at the ODVA
Industry Conference and 15
th
Annual Meeting, Oct.
2012, Stone Mountain, Georgia, USA.
Fischer, A.; Botero, J.; Beck, M.; De Meer, H.; Hessel-
bach, X., "Virtual Network Embedding: A Survey",
Communications Surveys & Tutorials, IEEE , vol.PP,
no.99, pp.1,19, 0 doi: 10.1109/SURV.2013.013013.
00155
Prytz, Gunnar. "A performance analysis of EtherCAT and
PROFINET IRT.", IEEE International Conference on
Emerging Technologies and Factory Automation ET-
FA, 2008.
Laplante, P. " What Every Engineer Should Know about
Software Engineering", CRC Press, 2007, ISBN
1420006746.
DCNET2013-InternationalConferenceonDataCommunicationNetworking
50