Performance Metrics for Industrial Ethernet Devices


Thread Starter

Steve Cliff

Hi List: I need to generate a list of performance metrics for industrial TCP/IP/Ethernet devices for use in the qualification and testing of devices to assist designers in the properly engineering of industrial applications. And maybe stir things up a bit as well. I am NOT asking about *network* metrics, but *device* metrics. Devices have to be matched so that clients do not flood servers beyond processing capabilities. All producers together must not generate so much data at to load the network too heavily. Yet data must be generated and processed fast enough to support the process itself. How can a designer confidently engineer a specific industrial network without these parameters? Certainly no one would attempt to design a structure or mechanical motion system or process system without the engineering data for the components of the system. How can there be confidence in the resulting design without component data? Device metrics should be concrete, measurable, objective, complete, reproducible, and useful. Think of the specification sheets that go with any component device you use in your day to day work. Many are measured by the manufacturer of the device, but for each class of device there are clear metrics for each class of device. The specs are how you pick the devices that fit your application. Today, in many networking and communications situations, the only specification available is the protocol supported. And the choices are so limited, that the matching of protocols is the only specification that is considered. The components are bought, assembled, and then the nightmare of system integration begins to try to find a way for the system to work. We must move beyond the just the protocol toward full metrics that will allow engineering of networks that work first time, every time. So, what metrics have you ever seen used to characterize network devices? What metrics do you think would be useful? Maybe these would help, may they would not: Maximum total number of incoming packets per second. Maximum number of incoming packets per second directed to this device. Maximum total number of packets produced per second. Maximum delay from arrival of data from the network to its delivery to the outputs. Maximum delay between the value in a network packet and the corresponding input change. Maximum response time to a command (may vary by command type). Largest packet size accepted. Largest packet size produced. Think about it. What do you really want and need to know to design a device into an engineered industrial TCP/IP/Ethernet application? Thanks for your thoughts, Steve. ---------------------------------------------------------------- Steven B. Cliff title: VP, Research & Development Control Technology, Inc. email: [email protected] 5734 Middlebrook Pike web: PO Box 59003 voice: (865) 584-0449 x220 Knoxville, TN 37950 fax: (865) 584-5720
My guess is that the network metrics overwhelm the device metrics--you're not using TCP for a device net, you're using it more for an inexpensive, shared-infrastructure, well-understood MMI/fieldbus. For example, you may toss lots of traffic into the bit bucket before the stack discovers a missing RIP update (30 to 90 seconds) or missing route. And since TCP does it for you, you probably ought to flow control your data throughput by giving the stack only the receive buffers you can deal with and let the stack report that to the sender in the Window Size field of the TCP header. Once you get someone half way trained to tune this sort of stuff for the applications, comm. infrastructure, and number of devices on the network, the IS department hires him away at twice the salary. I wonder if there's a way to make this network of devices configure themselves by selecting the right values of the properties you enumerate. Perhaps there are different sets which optimize for throughput, or response time, or cost (in byte/packet counts). Best, B.O. Mar. 13, 2001 -- Robert Old, System Architecture, [email protected] Siemens Building Technologies, Inc., HVAC Division 1000 Deerfield Pkwy., Buffalo Grove, IL 60089-4513 USA Phone: +1(847)941-5623, Fax: +1(847)419-2401
This sounds like the specifications that are used for many of the existing industrial networks (AB RIO, Profibus, Interbus-S, DeviceNet, ControlNet, AS-I, etc.). Any of the CAN derived industrial busses have these types of specs. They use the basic CAN protocol and add system and device requirements (baud rates, Max. # of nodes, data (single or multiple packet) size, voltage, media, connectors, short circuit protection, standard device profiles (starters, general input/output devices, motor drives, HMIs, etc.). I once read about a study done that showed that if ethernet traffic was less than 10% used, it was as deterministic as any of the "Industrial" busses. Since there is no real Max. node count to ethernet, maybe a simple calculator can be devised to determine when you will be over the 10% for a given network. -- C. Thomas Wiesen [email protected]