Most commonly, you build your control hierarchy in two layers. PLCs, loop controllers, and SIXTRAK stations are joined to the supervisory computer with serial links or dedicated proprietary networks. We call this the control network. Then, the supervisory computer is linked to our enterprise-wide LAN (Local Area Network), often with Ethernet. This classic layering addresses the concerns that unpredictable network traffic on the top end LAN may slow down delivery of the vital control data.

Your plant-wide network is probably an Ethernet network. We suggest that your control network be a separate dedicated Ethernet link. This lightly loaded control network will give you real-time performance because it is free from the heavy and often unpredictable traffic on the plant-wide network. By using Ethernet for both network layers, you enjoy many benefits including high performance, low cost, common training and reduced spare parts. It should be noted here that Windows 95, 98, NT, 2000, and Millennium support up to four simultaneously installed Ethernet ports (separate networks).

The control network only carries small I/O packets and is therefore lightly loaded. By only using a fraction of the network bandwidth (available message capacity) you are assured that data will always arrive quickly. Lets look at a numeric example:

Consider a system with 200 analog and 400 discrete I/O to update in each of five stations (1000 analog and 2000 discrete channels total) and a polling requirement of 50 mS.

Each analog value takes two bytes and eight discrete I/O pack into a single byte. If you figure a message efficiency of 50% (a conservative estimate that assumes that half the network characters are overhead such as addressing and error checking bytes), then 4500 characters (bytes) is sent during each system update. At 10 Megabits per second (1,000,000 characters per second), the total transmission time would be 4.5 mS. If we set the update requirement at 50 mS, then we are actually only using less than 10% of the Ethernet bandwidth. Even if errors occur, and each message is sent twice, during each 50 mS update cycle we would still only use less than 1/5 the network capacity.

This example demonstrates that the control network, free from large bursts of information that occur on the plant-wide network, is capable of real-time performance.