Automatic ring redundancy configuration protocol_cr 6

Automatic ring redundancy configuration protocol_cr 6
Automatic device configuration for Ethernet ring redundancy protocols
Oliver Kleineberg, Michael Ries, Markus
Rentschler
Hirschmann Automation & Control GmbH
Stuttgarter Straße 45-51
DE-72654 Neckartenzlingen
{oliver.kleineberg, michael.ries,
markus.rentschler}@hirschmann.de
Abstract
In modern communication systems based on Ethernet
technology, the use of physical ring structures and ring
redundancy protocols has been common for some time in
the past. A challenge remains the configuration of such a
redundancy protocol on each device with the networks
stretching out over large areas, like in offshore wind
power stations, where each windmill houses at least one
Ethernet switch. The distance between windmills can be
several kilometers and configuration of the redundancy
protocol on each device is an elaborate process. In this
paper, a mechanism is proposed which automatizes the
configuration of ring redundancy protocols and
eliminates the need to configure each device separately.
The actual protocol implementation is intended to be
part of the next major software release of Hirschmann
Industrial Ethernet Switches.
1. Introduction and Motivation
Ethernet technology has become an essential part of
many modern communication infrastructures. In the past
years, it has been very successful in the field of
industrial automation systems. Recently, with further
technological advancement concerning reliability and
performance, Ethernet technology is transpiring further
and further into new application domains like power
utility automation in case of the IEC 61850, or train and
transportation systems. This trend is continuously
advancing; the technology is far from its limits. Ethernet
is claiming more and more application fields, and where
in the past, Ethernet installations where limited to
network systems with small physical diameters and low
numbers of devices, modern applications more and more
pose the need for distributed communication systems
over large areas with an increasing number of network
devices.
More and more, Ethernet based systems are also used
in mission critical applications. So the availability of the
network is getting more and more a critical issue. In [1]
Max Felser
Bern University of Applied Science
Engineering and Information Technology
Jlcoweg 1, CH-3400 Burgdorf
[email protected]
different requirements are outlined and in [2] different
methods for high available media systems are evaluated.
In [3] these different solutions are specified.
Distribution of communication networks also means
the distribution of technical devices, these
communication networks are based on, over large areas.
Therefore, now and even to a greater degree in the
future, Ethernet switches will be distributed over large
areas, out of the immediate physical reach e.g. from
service personnel or network engineers, with a
substantial increase in device numbers.
Another major challenge for Ethernet adoption in new
application fields is not a technical aspect, but a sociotechnical one: a specialist in an application field, where
Ethernet is considered to be a future means of
communication, potentially replacing non-standardized,
proprietary cabling and protocols, has to understand and
to handle what he or she is working with. He or she may
not be or even cannot be an Ethernet specialist, but must
be able to work with the technology and Ethernet must
not deviate him or her from the actual work, consuming
valuable time and resources. A power protection systems
electrical engineer wishes not and should not be forced
to configure Ethernet switches elaborately for power
utility communication systems to work.
Different protocols are developed to support the
installation and planning of such Ethernet installations
and give more and more practical value as shown in [4]
and [5].
A protocol, which is able to configure ring
redundancy
protocols
on
Ethernet
switches
automatically, therefore is of great use: Network
engineers can configure a large number of devices
already in the field, potentially distributed over a large
area, while people dependant on Ethernet technology as
communication infrastructure can be presented with a
solution that allows them to configure a redundant
network with little effort. The general idea how to
automatically configure ring network devices has been
given in [6] with the actual detailed implementation left
open.
A challenge to the actual implementation is the ability
for the protocol to detect additional physical loop
structures that are present in addition to the ring that
needs to be configured. The protocol implementation not
only needs to configure devices, but also needs to detect
and report these loops so that the configuration on the
intended physical ring can be made.
The physical ring structure is well suited for a
distributed network application, as it allows different
participants in communication systems to be allocated
over a large physical diameter with low cabling effort,
compared to other physical topologies like star or tree
topologies. Instead of connecting each participant to a
central distribution device and creating a star network,
all participants are interlinked, with the last device in the
list connecting to the first device, closing the ring.
The test frames will be received and identified by the
MRCs on one ring port and subsequently only be
transmitted on the corresponding other ring port. Each
test frame sent by the MRM on one ring port will
eventually be received by the MRM on the
corresponding other ring port. Reception of test frames
on both ring ports signals the Media Redundancy Master
that the network is in good health. As soon as a fault
occurs in the ring network, the MRP test frames will stop
being transmitted over the whole ring and subsequently,
the MRM will stop receiving its sent test frames.
Additionally, an MRC can inform the MRM about
changes in the topology via link change frames. This
signals the MRM that the ring network is broken at one
point in the topology and that the MRM needs to open its
formerly blocked network port for user traffic.
As soon as the network error is repaired, the MRM
will again receive its MRP test frames and will again
block one ring port.
2.1. Ring redundancy protocol example
There are several ring redundancy protocols in
existence and used on the market, like the Siemens OSM
ring, the Hirschmann HiPER Ring and many other
protocols from other manufacturers. The Media
Redundancy Protocol (MRP) however, has been
included in the international standard IEC 62439 – High
availability automation networks. The proposed
mechanisms in this paper can be implemented in e.g. a
network switch software amongst other for the MRP
protocol, but the mechanism is not limited to MRP, but
to all ring redundancy protocols that operate after similar
mode of operation.
The MRP works with a dedicated master – client
structure, where a ring is compromised of one master
node called the Media Redundancy Master (MRM) and
several client nodes called Media Redundancy Clients
(MRC). In order to break the physical loop introduced by
the physical ring structure on the logical level, the MRM
sets one of its ring ports into blocking mode, only
receiving
and
transmitting
MRPDUs
(Media
Redundancy Protocol Data Units) on this interface.
Normal user traffic is not relayed on the blocked port,
only on the second ring port, which remains in
forwarding state. This translates the ring structure to a
line structure without loops on the logical level.
For redundancy surveillance, the MRM transmits
MRP test frames from both its ring ports. These test
frames are sent to a specified MAC (Media Access
Control) multicast address which allows a switch to
identify PDUs for this protocol on the basis of the MAC
address and to handle the messages accordingly. By
default if no means of manipulation of any kind are
implemented on the switches, frames sent to a MAC
multicast address are handled like frames sent to a MAC
broadcast address: They are forwarded to every network
port except the port they were received at.
2.2. MRP and Profinet IO
Profinet is an automation network, based on and
compatible to Ethernet (IEEE 802.3) and specified in
IEC 61158-5-10 [7] and IEC 61158-6-10 [8] and IEC
61784-2 [9]. A Profinet IO-system consists of an IOcontroller, one or more IO-devices and possible IOsupervisors. The IO-supervisors are typically
engineering tools. In a typical Profinet IO-system an IOcontroller does control one or more IO-devices.
Reference [10] provides a good overview about the
functions of Profinet.
The specification provides three conformance classes
of Profinet IO-systems. These classes differ in the
supported
application-,
communicationand
redundancy-classes and specify the required features.
Higher classes are compatible to the lower ones.
Class A specifies certified IO-controllers and IOdevices with standard Ethernet interfaces and standard
Ethernet network infrastructure. Class B requires in
addition to Class A that the network infrastructure
conforms to the Profinet specification. Media
Redundancy Protocol (MRP) is required to support Class
B and thus basic redundancy network structures are
possible. Additional redundancy protocols are optional.
In class C Profinet IO-systems, additionally to the MRP
also the Media Redundancy Real-Time (MRRT)
protocol and the Isochronous Realtime (IRT) protocols
are mandatory.
With the MRP being an integral part of the Profinet
redundancy concept, the proposed automatic ring
configuration protocol can also be used to configure
Profinet redundant ring structures. With Conformance
Class B, in addition to MRP, the use of LLDP (Link
Layer Discovery Protocol) is mandatory. This further
adds to the usefulness in regards to device and capability
detection of the protocol in Profinet environments, as
described later in this paper in 3.6.
2. Ethernet Ring Networks and Protocols
3. Automatic device configuration
The basic fact why an automatic configuration of ring
devices is possible at all is that each ring node in a fully
functional redundant ring network has exactly two
distinguished network ports that are part of the ring.
These ports are identifiable and configurable e.g.
through a switches’ web configuration interface or a
command line interface via serial link or telnet.
A complete detection and configuration cycle from a
user’s point of view would be a detection initiation via
user command on the MRM and after successful ring
detection, which means that no additional loops were
detected, a subsequent configuration initiation via a
second user command. A one step solution cannot be
done, because as mentioned, before the configuration
takes place, it has to be certain that the ring which is
intended to be configured is the only loop structure in the
network.
figure 1. example ring network
Figure 1 shows a ring network consisting only of
switch nodes. The two ring ports on each device are
marked with a circle and a square. Without any further
interpretation, this network can be viewed as a simple,
undirected regular graph. When the functionality of
MRP test frames is applied to this network, this changes
the view of the network paths. From the MRM, test
frames egress both the circle and the square port. The
graph representing the network has exactly one link
between every two vertices, in this case represented by
network switches, and the frame will, unless the network
has been damaged, ingress the adjacent device on a ring
port. This means that, for a specific MRP test frame from
a port of the MRM, the ring represents itself as a directed
graph. In figure 2, it is shown that a frame traveling
clockwise from the MRM always egresses a network
port marked with a circle while counter-clockwise, a
frame always egresses a port marked with a square. In
this basic, trivial topology, a frame will travel exactly the
path the ring has to be configured. It will pass both ring
ports on all devices in the ring, even if the devices are
not already configured as participants in a redundant ring
network. So the basic idea is, that special configuration
PDUs, sent out from the MRM in a manner similar to the
MRP test frames, will be used to propagate configuration
information to all devices in a ring network and to test
the network structure against additional loops.
figure 2. Interpretation as directed graph
The ingress and egress ports of these configuration
PDUs (one from each direction in the network) mark the
ring ports, the device has to be configured with.
These configuration PDUs include in their SDU
(Service Data Unit) all relevant parameters for the ring
redundancy protocol that need to be configured on the
individual switches. In the case of MRP, this includes
e.g. the timing constraints, defined by the protocols’
consistent set of parameters, and a ring VLAN ID
(Virtual Local Area Network Identifier). In addition, it
also includes a hop count which indicates how many
devices the individual PDU has passed and which is
incremented in every device upon reception, a sequence
number, which is incremented in the sending MRM for
each set of PDUs, one PDU for each direction, and an
MRM port indicator, which identifies the original egress
port of the PDU on the MRM. The port indicators are
representations of the already mentioned circle and
square port identifiers.
3.1. Supported topologies and challenges
When the simple ring topology is extended, it
becomes clear that the characteristic of the configuration
frames, sent to a MAC multicast address, will introduce
some additional challenges.
Figure 3 shows the previous ring topology with MRC
1 also being the root element of an additional tree,
emerging from the original ring. In this case, an
additional switch structure is plugged into a non-ring
port of the ring switch MRC 1. If we now consider the
ring to be not configured with exception of the MRM,
and the MRM starts sending configuration frames from
its ring ports, these frames will eventually reach MRC 1.
Now, the ports which will be configured as ring ports are
not predetermined, and MRC 1 will receive the
configuration frame from the “circle” direction on one
which is being duplicated at MRC 2 and will ingress
MRC 1 on two different ports.
figure 3. ring with additional tree
port and forward it on all other connected ports except
the receiving port. This means that with the ingress from
“circle” direction, one ring port can be precisely
determined, while with egress in “circle” direction, two
possible ring ports are determined. With the
configuration frames in “square” direction, MRC 1 can
also identify its second ring port without a doubt through
ingress of the frame from the “square” direction, but
egress of the “square” frames takes place on all other
ports of the device. This means that the non ring
participant receives both configuration frames from the
circle and the square port of the MRM on one interface,
but as it has not received configuration frames on a
second port, it can identify that it is not part of the ring
structure, but part of a leaf link or tree, emerging from
the original ring.
But as soon as trees emerging from the ring are
meshes and contain loops themselves or the network
structure is more complicated, e.g. a ring with additional
links between the ring devices, the complexity rises
again.
3.2. Meshed network structures
Figure 4 shows a ring structure with an additional
mesh, connecting the ring nodes MRC 1 and MRC 2.
When the configuration frames travel through the ring, at
MRC 1, the “circle” frame will be forwarded to both
mesh links: On the one hand on the link which directly
connects MRC 1 and MRC 2, and on the other hand on
the link to the mesh containing the “non ring
participants”. The configuration frames over both
meshes will eventually ingress MRC 2, which without
further information, cannot ascertain which port he
received the configuration frame on will be the future
ring port. The only parameter that distinguishes the two
frames is the hop count, as two corresponding frames
always carry the same sequence number and port
identifer, because the frame has been duplicated at MRC
1. The same problem applies to the “square” frame,
figure 4. ring with mesh
In addition to the doubly received frames on MRC 1
and MRC 2, these switches would, without any further
control, forward the configuration frames to all ports
they did not receive them originally. This would lead, in
the long run, to a multitude of copies of configuration
frames, flooding the network. So the loop detection
mechanism must include means not only to detect the
loops, but to remove the configuration frames
themselves and prevent them from looping.
So it is necessary to save for each port of a network
device the highest received sequence number and the
lowest hop count for the sequence number. Per network
device, a loop detection flag and a loop source identifier
flag are needed. With this information in each network
node, it is possible to detect loops and to detect, whether
a network device is the source where a loop in the
topology is originating. In figure 4, the additional mesh
with the loop originates from MRC 1 and MRC 2. These
two devices will be reported to the MRM as loop sources
and will be identifiable by the administrator. The
mechanisms how the loop detection works on each
individual MRC and MRM are described in the
following paragraphs.
figure 6. information stored in each device
Figure 6 shows the information stored in each device
in an exemplary, figurative table per port. The loop
source identifier and loop detection flags are omitted in
this table, they are not stored once per port but once per
device. Because configuration frames travel in two
directions from the two ring ports of the MRM as
described above, for the frames of each direction, one
table column has to be maintained.
3.3. Configuration initialization and MRC behavior
The only device that needs to be configured by the
administrator is the MRM. On this device, all relevant
parameters for the ring redundancy protocol are
configured. Once this is done, the automatic ring
configuration is activated, and the MRM starts to send
configuration frames to the ring, initiating the loop
detection mechanism.
In addition to the MRM configuration, a protocol
removing loops on the logical layer, like RSTP (Rapid
Spanning Tree Protocol), must be activated on the
devices. If the topology is a simple ring and the ring is
the only loop structure present, the MRM itself
immediately breaks the loop after it is being configured,
but in case of additional loops besides the main ring or if
the MRM is in factory default state and not blocking one
ring port, a protocol like RSTP is needed to prevent
loops which would overload the network. For the
protocol to work properly, the configuration frames need
to pass ports that are blocked by RSTP on devices that
are aware to the automatic configuration protocol,
similar to RSTP BPDUs (Bridge Protocol Data Units) on
blocked ports of Switches that support RSTP. An MRM
receiving a configuration frame will remove the
configuration frame from the network and analyze the
source MAC address: If the source MAC address is not
its own MAC address, another MRM is present in the
ring. This is a faulty configuration, because with MRP or
any similar ring redundancy protocol, only one
redundancy manager may be present in a ring network at
any time. This fault is communicated to the
administrator, who has to remove the other MRM prior
to an automatic ring configuration.
An MRC, receiving a configuration frame on one of
its interfaces will analyze the SDU of the frame: The
sequence number and hop count are analyzed and
compared to the values stored in the switch. The analysis
of hop count and especially sequence number has to be
done because in switching queues, frames can be
reordered or if there are meshes present, multicast
frames will be doubled. This could lead to possible
ambiguities and misinterpretations on the devices.
The behavior of the protocol in an MRC is described
as follows:
1. When a device receives a configuration frame on
one port, it checks its port table and compares the
sequence number of the frame received with the
sequence number that is stored in the table for this
port:
a. If previously no frame with this sequence number
was received on this interface and therefore the
sequence number field for this port is empty or if
it contains a lower value, the entry in the table is
updated with the sequence number and hop count
from the frame and the frame is forwarded on all
ports except the port it was received at. If the
entry that was created or updated was the third
entry in the port table with this sequence number,
the loop source identifier flag is set.
b. If a frame with this sequence number was already
previously received on this interface, the frame is
dropped and the loop detection flag is set. If the
hop count of the frame received is lower than the
one stored, the stored value is updated with the
new count.
c. If the sequence number stored in the port table is
higher than the sequence number in the received
frame, the frame is dropped.
2. When the loop detection flag is set, a unicast
message to the MRM is sent, reporting a network
configuration which is unsuitable for automatic ring
configuration. This is subsequently reported on the
MRM to the administrator/operator. The MRM MAC
address is known to each MRC via the source address
of the configuration frame.
3. If the loop source identifier flag is set, a unicast
message is sent to the MRM, indicating the device to
be the source of a loop structure, which has to be
removed before an automatic ring configuration can
be made.
The saved information is kept for frames from both
directions, there are separately saved information for the
“circle” and “square” frames, as shown exemplary in
figure 6.
It is possible for a loop not to be detected with this
mechanism, if the frequency of configuration frames sent
to the network is too high, compared to the transit time
through a large network mesh. In a worst case scenario, a
frame with a higher sequence number n could always
arrive at a device before a duplicated frame with the
sequence number e.g. n-1 or frames with the same
sequence number could be received on more than two
interfaces, but in a timely context that never three entries
with the same sequence number could be present. This
would mean that loops which introduce frame
duplication would never be detected and the devices the
loops originate from cannot be detected. Therefore, the
sending frequency for configuration frames from the
MRM must not be too high, but in a sensible range for
large loops to be detected, e.g. around 20 – 30 seconds.
This gives the frames ample time to be transmitted,
duplicated, registered, received and removed on the
individual network devices so that the port tables will be
filled and the system goes into a steady state until the
next configuration frame with a higher sequence number
arrives. If, a new configuration frame arrives and an old
frame is still circulating, the old frame will be removed
when the port tables are updated with the higher
sequence number, as described above.
3.4. MRM data analysis and behavior
With the data transmitted from the MRCs, the loop
detection and loop source identifier information, a MRM
can determine whether a consistent ring configuration is
possible. The MRM will always remove any
configuration frames it receives on any port. In addition,
it will do the following:
1. If the MRM receives configuration frames with
identical sequence numbers but different hop
counts, it will detect a loop.
2. If the MRM receives a configuration frame with a
port indicator on the port this frame was sent on,
it will detect a loop.
3. If the MRM receives a loop detection indication
frame, it will detect a loop.
4. If the MRM after three consecutive configuration
frame sending operations has not received any
configuration frame on its other ring port, it
assumes that the ring is broken at some point and
physically not closed. This error needs to be
corrected before the automatic configuration can
be done.
5. If the MRM has not received a loop detection
indication or loop source indication frame from
an MRC or has not detected a loop itself after
three consecutive sent and received configuration
frames, it assumes that the ring configuration can
be made and it prompts the user to initiate the
configuration process.
If at any point the automatic configuration fails due to a
detected loop, the user will be informed to reconfigure
the network. The MAC addresses from the loop source
indication frames, where loops are detected, are also
reported. Via this MAC address, possibly in conjunction
with a network engineering tool or a network
management system (NMS), an administrator can
pinpoint the devices which are part of a loop structure
and initiate the appropriate actions for the
reconfiguration of the physical network.
3.5. Activation of the automatic ring configuration
If no additional loop is detected, the user is prompted
to start the configuration process with a user command
on the MRM. As soon as the user gives the command to
configure the ring, the MRM sends out a configuration
indication frame from both ring ports. A MRC receiving
the configuration indication frame will configure the ring
redundancy protocol for the previously received
parameters and after the configuration is done and the
device is ready to participate in the ring, the frame is
forwarded on the other ring port.
As soon as the MRM receives both configuration
indication frames on both its ring ports, it assumes that
the ring is configured and ready for normal operation. It
will then assume normal ring operation according to the
ring redundancy protocol configured.
3.6. Devices which do not support the protocol
Devices which do not support the protocol may be
present either in trees or meshes outside the desired ring
or may be ring devices.
In the case the devices are present in trees connected
to protocol sensitive devices that are part of a future ring,
the loop detection mechanism on the corresponding ring
device will be unaffected. In case a whole tree consists
of devices unaware to the protocol, potential loops in the
tree substructure will be administrated via RSTP or a
similar protocol, while the devices not aware to the
configuration protocol will treat the configuration frames
like standard MAC multicast frames.
In the case that they are ring devices, this is a critical
installation fault that cannot be detected via the
configuration protocols detection mechanisms alone. In
this case, the automatic configuration cannot be realized.
A detection mechanism to identify protocol unaware
devices on the ring via the protocol alone is not possible,
as these devices are completely transparent to the
protocol and therefore to any detection mechanism. Such
a faulty configuration can be detected, if a NMS is used
to monitor the network devices and to map a network
topology.
With Hirschmann devices, neighbor device detection
is implemented using the industrial standard protocol
LLDP [12]. LLDP is also implemented in Profinet
devices compliant to conformance class B or higher. If
there is no NMS in use to support the detection of
network devices not supporting the protocol, LLDP can
be used for device and capability detection [4]. But
generally, a setup with a protocol unaware device shall
be prevented from the start with a correct physical
device setup.
3.7. Devices with limited protocol support
In case a device has just exactly two distinguished
ports, it can implement the protocol with just the ability
to extract protocol configuration parameters from the
configuration frames and to forward the configuration
indication frame. This can be implemented e.g. on
common field level Profinet I/O devices with two ports,
offering MRP MRC support. This enables a slim
protocol implementation on I/O devices with very low
resource consumption.
4. Security: Protection against misuse
While configuration protocols often provide quick
and comfortable ways to configure network devices with
little effort, there is always the possibility of misuse and
sabotage.
In order to sabotage a network already in operation, a
denial of service attack with a very high number of
configuration frames sent to the network could be done
to overload the devices. This is prevented with a
mechanism to limit the number of actually received and
processed configuration frames to a defined low number.
Additionally, as a first step to secure a configured and
running installation against protocol frames inserted by
an attacker with malicious intend to reconfigure the ring
devices and to interrupt normal ring operation, the
automatic ring configuration will be locked on all MRCs
once the ring configuration is done. This lock can be
released by connecting to each MRC individually and
authenticate via command line, web interface login or
via SNMPv3 (Simple Network Management Protocol
Version 3). This assures that only service personnel with
correct login credentials can manipulate the switch
configuration. There is a distinct disadvantage to this: To
reconfigure a network or to initiate a new automatic ring
configuration cycle once the network is in operation, a
manual operation on each ring device is necessary.
To circumvent this, in future versions of this protocol,
an authentication mechanism is planned between an
MRM and the individual MRCs, reducing the actions
needed to initiate another configuration cycle to the user
authentication on the MRM.
frames that are additionally. The mechanism works the
same way for the frame from MRM ring port 2, just the
other way around.
Figure 8 shows the path of the frame(s) in a tree-like
structure, where a branch in the tree indicates a frame
multiplication. The frame travels on a step-by-step basis,
the switching times and line delays in the network are
assumed to be virtually identical between all devices and
links. In a real world network, this will not be the case
and the “tree” in figure 8 might look different. A step in
figure 8 can also be interpreted as the hop count after
ingress and increment in the respective device. The steps
and ingress directions are displayed in numbers and with
small arrows in figure 7.
5. Application to a use case
Figure 7 shows a fictional network to which the
automatic ring configuration will be applied. The MRM
is marked with “MRM” and all MRCs are given
characters for identification. The first step of the
automatic configuration is that the administrator
configures the MRM with all necessary ring protocol
parameters and initiates the automatic configuration
process by starting the ring check. The MRM will then
send out configuration frames from both its ring ports, in
order to deliver the parameters of the protocol to the
MRCs and to check if the topology is, with exception of
the ring to be configured, loop free.
figure 7. ring structure with mesh
In figure 7, only the paths a configuration frame sent
from ring port 1 of the MRM, in this case on the network
connection going upward towards MRC “A”, are
displayed. The paths a configuration frame takes from
MRM ring port 2 are not displayed so it is possible to
keep track of the frame from port 1 and subsequently the
figure 8. travel paths of configuration frames
The frame starts off from MRM Port 1. In MRC “C”,
it is multiplied for the first time, traveling to “D” and
“E”. As described in 3.3, on each ingress of a frame, its
sequence number is recorded for this port. The frame
traveling from “C” is again multiplied in “I”, and also
travels towards the MRM Port 2, where it is removed
from the network, indicated by a crossed-out box in
figure 8. It also marks all ingress ports on this network
path with its sequence number, and in the future, no
other frame with the same sequence number can travel
this network path, but will be removed. The frame
multiplications can be observed when following the
paths of the frames in figure 7 and figure 8. A special
case comes up in step 10, where frames from “E” and
“D” arrive at identical step count on different ports. It is
non-deterministic, which frame will arrive first and will
be processed first. In this example, it is assumed that the
frame from “E” did arrive first and travels all the way
back to MRM Port 1. The second frame will be received
and transmitted, but will be removed on the following
devices. MRC “C” can therefore, by the ingress of one
frame with an identical sequence number on 3 or more
ports, be identified as the source of a loop and will send
the loop source indication to the MRM. MRC “I” cannot
be detected as a loop source in this direction, but will be
identified by the frame traveling from MRC port 2 in the
opposite direction. With many devices sending the loop
detection indication and loop source indication, the
administrator will not be able to continue with the
automatic ring configuration. He or she has to remove
one physical loop, e.g. by administratively (temporarily)
blocking a port on MRC “C” or by removing physical
connections if they were not intended in the original
network design. The additional use of a NMS that can
graphically map a network topology or a Profinet
engineering tool can be of great use here. After the loop
is removed, the configuration can be started again. If no
loop was detected on the MRM after three consecutive
configuration frames sent and received, the administrator
can activate the ring configuration.
6. Implementation
Software
in
Ethernet
Switch
The automatic ring configuration is intended to be
part of the next major release of Hirschmann Ethernet
Switch Software. In the software itself, the functionality
will be called “ARC”, short for “Automatic Ring
Configuration”. Like with most switch functionality in
Hirschmann devices, the protocols configuration and
saved parameters are accessible via SNMP through a
dedicated MIB (Management Information Base). A new
MIB for the ARC has been written, containing all
protocol parameters and a list item for MAC addresses
that have been identified as sources of a loop via the
loop detection flag described earlier. This information
can then be accessed via SNMP to be used in a NMS.
The information is also used by the web user interface of
the devices. If a device fulfills the MRM role, the ARC
can be initiated in the same dialogue where the ring
redundancy is being configured.
7. Summary
The automatic ring configuration offers a quick and
easy way to configure many ring devices at once with
only a few configuration steps necessary. This aids
administrators by saving time in configuration processes
and enables the possibility of ring redundancy
configuration for application specialists who want to use
redundant Ethernet network structures, but don’t want to
be burdened with an elaborate configuration process.
The automatic ring configuration has the ability to
detect whether it will work in a certain network layout,
but, like with a manually configured network,
precautions in the deployment of actual physical network
devices have to be made in order not to mix devices not
supporting network protocols with devices supporting it.
The automatic ring configuration is especially useful
in application scenarios where the installation follows
the network structure which is intended with ring
redundancy protocols: one ring with simple leaf links or
connected trees. In these network structures, the protocol
immediately detects and configures network devices that
are part of the desired redundant ring structure.
References
[1]
Prytz, G.; “Redundancy in Industrial Ethernet
Networks”, Factory Communication Systems, 2006 IEEE
International Workshop on June 27, 2006 Page(s):380 385
[2] Kirrmann, H.; Dzung, D.; “Selecting a Standard
Redundancy Method for Highly Available Industrial
Networks”, 2006 IEEE International Workshop on
Factory Communication Systems, June 27, 2006
Page(s):386 – 390
[3] IEC 62439: “Industrial communication networks: high
availability automation networks”, CDV distributed on
2008-11-21 available at www.iec.ch
[4] Schafer I., Felser M.: “Topology Discovery in
PROFINET”, 12th IEEE Conference on Emerging
Technologies and Factory Automation, 2007. ETFA
2007, September 25-28, Patras, Greece
[5] Kleineberg O., Felser M.: Network Diagnostics for
Industrial Ethernet, 13th IEEE International Conference
on Emerging Technologies and Factory Automation
(ETFA 2008); September 15-18, 2008, Hamburg,
Germany, Work in Progress
[6] Rentschler, M.; Maisch, W. - Patent Nr.
US020080250124A1 - [EN] Redundancy-protocol
configuration in a ring network
[7] IEC 61158-6-10: “Industrial communication networks –
Fieldbus specifications – Part 6–10: Application layer
protocol specification – Type 10 elements”, available at
www.iec.ch
[8] IEC 61158-5-10: “Industrial communication networks –
Fieldbus specifications – Part 5–10: Application layer
service specification – Type 10 elements”, available at
www.iec.ch
[9] IEC 61784-2: “Industrial communication networks –
Profiles - Part 2: Additional fieldbus profiles for realtime networks based on ISO/IEC 8802-3”, available at
www.iec.ch
[10] PROFIBUS International: “PROFINET: Technology and
Application, System Description”, Document number:
4.132,
Issue
April
2006,
available
at
http://www.profibus.com
[11] Felser M.: Media Redundancy for PROFINET IO, IEEE
Workshop of Factory Communication Systems 2008,
Dresden, 20 to 23rd May 2008, Pages 325 to 330
[12] IEEE: IEEE 802.1AB-2005 - IEEE Standard for Local
and metropolitan area networks Station and Media
Access Control Connectivity Discovery; available at
www.ieee.org
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement