FDDI for Process Control Network

  • Thread starter Baskar K.G. (GPPD)
  • Start date
B

Thread Starter

Baskar K.G. (GPPD)

Dear all,
Does anyone have experience working with FDDI network? We are planning to use FDDI as a backbone (outdoor) for Process control network for connecting Win-NT servers. Help me by providing information on the reliability, robustness, redundancy, ease of maintenance
of FDDI for process control network.

Is there any other field proven, reliable alternatives available to FDDI for outdoor cabling to support high speed and redundant
network?

I appreciate your attention and help.

Baskar.K
[email protected]
 
E
About five years ago we installed and maintained a number of FDDI systems on large industrial sites (mostly pulp mills). It worked very well and had the best redundancy scheme of any system I have worked with. The only problem we have was
the distance limitations of 1km when you used the redundancy (2 km if you didn't use redundancy).

Unfortunately, we see it as a dying protocol. The cost for FDDI is getting far too high as compared Switched Fast Ethernet and as a result we haven't done a new FDDI install in over 3 years. We now use ATM (expensive but wonderful) or Fast Ethernet. The trouble is that the standard redundancy scheme used in Ethernet implementations (Spanning Tree) is a real pain and we have had to turn it off in several sites. 3COM and Cisco both have good proprietary redundancy schemes but you need to stick with a single vendor.

So my answer is FDDI is wonderful (under 1km) but so was OS/2 and Beta Tapes...
You may find that you have to live with Fast Ethernet.

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

Rob Hulsebos

>...new FDDI install in over 3 years. We now use ATM (expensive but
>wonderful) or Fast Ethernet. The trouble is that the standard
>redundancy scheme used in Ethernet implementations (Spanning
>Tree) is a real pain and we have had to turn it off in several sites.

Is the fault detection time too long in Spanning Tree, or is it something else?

>3COM and Cisco both have good proprietary redundancy
>schemes but you need to stick with a single vendor.

Can you elaborate on their merits?

Greetings,
Rob Hulsebos
 
E
Hi Rob & Pravin

Sorry I have taken so long to respond. First Pravin asked for a brief description of what Spanning Tree is... Good question, so here is my brief answer to that...

"The spanning tree algorithm (STA) is an IEEE standard (IEEE 802.1d) that ensures the existence of a loop-free topology in networks that contain parallel bridges. A loop occurs when there are alternate routes between devices. If there is a loop in a network, bridges may forward traffic indefinitely, which can result in increased traffic and eventual network failure. STA prevents loops by allowing bridges to detect parallel paths and blocking one of these paths."

As for the problems we faced with Spanning Tree, they are due to STA's topology discovery time. Whenever the bridges or switches on the network detect a new bridge or switch they can go into listening or learning modes. During this time
all traffic on the network stops and in large networks it can stop for up to 30 seconds! In one pulp mill it was long enough to cause all SNA sessions to drop.

Generally TCP connections would survive but it is not great having all network traffic blocked for at least ten seconds. And this wasn't just when there was a network fault, it was times when a switch was added or removed anywhere in the
network. (For more info on this see
http://www.cisco.com/warp/public/cc/sol/mkt/ent/ndsgn/spane_an.htm.)

As for the vendor redundancy schemes, they are typically based on monitoring the state of a single pair of links rather than a network wide view. For example, you can connect one hub (or switch) back to two separate switches that are both on the backbone.

SWITCH<---------->SWITCH
\ /
\ /
---->HUB<----

Under normal design rules we would have created a loop and if Spanning Tree was active, one link would be blocked. With the vendor schemes, you define the two links as a primary/redundant pair and the hub monitors both for "aliveness". If
it sees that the primary link is alive it blocks the redundant link. If it sees that the primary link is dead it activates the redundant link. Unlike Spanning Tree, it doesn't care what the rest of the network is doing so there is no
learning when other devices are added or removed.

The disadvantages of the Vendor schemes are:
a) they are proprietary and
b) they can't detect if you have accidentally created a loop (and then save your butt).

Vendor schemes expect you to know what possible loops exist and set them up as redundant link pairs. And this isn't always obvious if you have a complex redundancy scheme. For example, a student in the ISA Industrial Networks class last month described an ongoing problem where their plant network occasionally shoots over 70% traffic loading after certain field events (it then crashes). I obviously didn't get a chance to visit the site, but my guess is they have a
intermittent loop in their rather complex redundancy scheme. Spanning Tree would prevent this, but at the cost of long resolution times when the network was modified.

Regards
Eric Byres, P. Eng.
[email protected]


P.S. Cisco claims to have addressed the STA concern by developing something called the Spanning Tree UplinkFast feature. I haven't played with it yet.
 
Top