McCarthy, GPIB Product Manager, National Instruments
Most test systems are hybrids – they comprise instruments from multiple vendors and use multiple instrument control buses to connect the instruments to the PC core. GPIB continues to grow at or above the overall electronics industry, almost all of the latest instrument introductions include a GPIB port, and, while mature, GPIB still offers some unique advantages (like performance, ease of setup, connector locking, etc.). Because most instruments offer multiple bus options, the challenge is deciding which bus provides the best performance for your application. While theoretical latency and bandwidth specifications offer an indication of expected performance, actual benchmark results from an industry-leading scope demonstrate that GPIB performs better in systems with small data transfers and Ethernet is better for large transfers. Because the optimal bus may change depending on application, it is important to select software that works well with all the major instrument control buses.
Two factors affecting bus performance are theoretical bandwidth and latency. Most peripheral buses, such as Ethernet, USB, and GPIB, trade off higher bandwidth for lower latency. Figure 1 shows how higher-bandwidth buses such as Fast Ethernet (100 Mb/s) and Gigabit Ethernet (1000 Mb/s) are better suited to larger data transfers. Lower-latency buses such as GPIB and USB 2.0 are better suited to small data transfers. System buses such as PCI and PCI Express are designed to maximize both bandwidth and latency because they must aggregate multiple peripheral buses and thus work very well as the core of a hybrid system.
Figure 1. Lower latency and higher bandwidth provide the best performance.
The actual bus performance of an instrument depends on many factors including the processor, chipset, RAM, operating system, protocol stacks, and buffers. To obtain actual results, National Instruments performed a series of benchmark tests on a Tektronix TDS5104B oscilloscope. These tests incorporated a high-precision counter to measure the transfer time for different size data transfers on GPIB, high-speed 488 (HS488), and Ethernet. All tests used the same hardware to remove any host hardware variances. The tests were written in the National Instruments LabVIEW graphical development environment using Virtual Instrument Software Architecture (VISA) direct I/O software calls to minimize software driver inconsistencies. The results are shown in Figure 2.
Figure 2. HS488 offers the
best performance up to 8 KB transfers.
With the lower latency, HS488 provides the higher throughput for data transfer sizes up to 8 KB. Above 8 KB, Ethernet delivers the higher throughput due to higher bandwidth. Without these benchmark results, it is hard to know where the trade-off between bandwidth and latency will negatively affect system performance for a given transfer size.
It is impossible to determine which bus provides the better system performance for your application when you do not know the data transfer sizes. Figure 3 provides a mix of transfer sizes for two hypothetical test systems.
Figure 3. Ethernet is 1.7 times faster for the validation profile and HS488 is 3.3 times faster for the production profile.
The first system is a hypothetical validation test system for characterizing a device under test (DUT). In characterizing a DUT, you must take many large waveform measurements at different operating conditions and instrument settings. From these measurements, you can determine the limits and ranges of acceptable values for production test. Hence in production test, the mix of transfers shifts to smaller instrument set commands and smaller measurement results being returned. Using the transfer mix for the two cases in Figure 3 shows that Ethernet has an advantage in validation and GPIB is more than 3 times faster in production. With VISA, you can obtain a common interface so you can use the same test software in both cases.
For optimal performance, choose the bus with the lowest latency and highest bandwidth. GPIB has a clear performance advantage in systems with smaller transfers while Ethernet has an advantage with larger transfers. Benchmarking the performance of a particular instrument helps you determine the crossover point between the available buses. Using a bus-agnostic software interface such as VISA provides an easy way to benchmark and migrate between multiple buses in hybrid systems. Finally, the optimal bus for your application is only apparent when real results are applied to a particular use case.