Could it be possible to have a communication redundant link Ethernet TCP/IP in a programmable electronic system with architecture triple modular
redundant by a LAN and/or WAN network? In what layers of OSI/ISO, could we have troubles?

Or maybe, the troubles could be in the hardware or software of PES?

 

Hi Juan,

Your question has two answers :

first : for WAN.

in this case the redundance depend of the configuration of your network and in particular of the routers . It is possible to configure them for redundancy (of course you need a network configuration in which it is possible to reach the nodes through different way) by adding the various routes in the routing table of the routers (in many case, it is possible to
configure the router in a manner that it automatically find the various routes -using ARP and RARP protocol). These configuration depends on the material you use.

second : for LAN

For the lan, the actives nodes (hubs, switch and repeaters) must accept the "spanning tree" which authorize the redundancy.

In both cases, you should not have any problem with your software in the sense that this redundancy affect only layer 2 (LAN) and 3 (WAN) of the OSI model.

Regards

S.Monnet
SMAPI2

 

There are several things:

Use redundant Ethernet link transceivers -- they will take a Fiber or UTP connection and monitor the Ethernet "heartbeat". If a failed link goes down, it will use the alternate fiber pair. This can respond in under a milli-second.

Ethernet switching hubs can failover on a redundant fiber pair. They use similar technologies as the redundant Ethernet link transceiver. You probably want to setup a RMON or SNMP trap so it will report an alarm to the network management console when the link goes down.

Routers support multiple paths (routing). (They operate at level 3 of the ISO model).
ICMP will respond in milli-seconds up to seconds--if there is downed path.

IPv4 is limited in redundancy. You may want to use multi-cast technology for updating your real-time database. ISO protocols have better support for redundancy, but are dead in the market-place.

Sisco has expertise in both IP and ISO. They were very helpful on the UCA, in looking at redundancy issues at the application layer perspective.

Most problems occur at the Application layer. For instance, you typically want to have multiple connections going to a device, so in event of a failure, you can quickly switch over to the "trusted" system.

Redundant systems, typically need incorporate finite state machines, where you go back to the last known good-state before going ahead.

William F. Hullsiek
Software Engineer
Renewal by Andersen
Voice: 651.501-4000
Fax: 651.501-4080
Pager: [email protected]
Cell: 651.402.0946

 

From: "Steve Monnet" <[email protected]>
> Your question has two answers :
> first : for WAN.
> in this case the redundance depend of the configuration of your network and
> in particular of the routers . It is possible to configure them for
> redundancy (of course you need a network configuration in which it is
> possible to reach the nodes through different way) by adding the various
> routes in the routing table of the routers <Clip>

Well, it's not quite that simple... the routers will automatically support redundant paths between each other. However, your host (eg. PLC, PC, etc.), needs to connect to a single router which is its dedicated "Default Gateway". Lose the "Default Gateway" router and it's bye-bye TCP/IP connection. Most hosts support only a single Default Gateway.

Cisco has a solution to this where you can set up two routers create a third virtual router. If one router fails the other becomes the virtual router. You set your host's Default Gateway to point to the virtual router. This is not trivial.

> second : for LAN
> For the lan, the actives nodes (hubs, switch and repeaters) must accept
> the "spanning tree" which authorize the redundancy.

Again, it's not quite that simple... Spanning Tree (IEEE 802.1d) only works for switches, bridges and repeaters and not computers. Your host typically connects to a single hub or switch. Lose the connection to the hub (or the hub
itself) and it's bye-bye TCP/IP connection once again.

And putting two NICs in you PC doesn't help because the TCP/IP layer software won't know to automatically move your connections, ARP Cache, etc. to the second card if the first one fails.

To make matters worse, Spanning Tree can have a very slow resolve time to sort out failures. So slow that many connections drop out. We have had to turn off Spanning Tree in some applications.

Both 3COM and Cisco have a solution for this that involves a switch, two NICs and special driver software. But this assumes an NT or UNIX platform so if you are trying to do this in a PLC...

There is always a solution but it isn't always easy. Hope this clarifies things a bit.

Eric Byres, P. Eng.
Artemis Industrial Networking
[email protected]