Instrument Bus Performance – Making Sense of Competing Bus Technologies for Instrument Control

Today, USB, PCI Express, and Ethernet/LAN have gained attention as attractive communication options for instrument control. Some test and measurement vendors and industry pundits have claimed that one of these buses, by itself, represents a solution for all instrumentation needs. In reality, it is most likely that two or more bus technologies will continue to coexist in future test and measurement systems because each bus has its own strengths. The challenge for today’s test engineer is not to choose a single bus or platform on which to standardize every single application, but to choose a bus or platform appropriate for a specific application or even a specific part of an application.

This paper presents a head-to-head comparison of the most popular instrumentation buses, so that test engineers can make informed decisions when choosing the bus and platform technologies to meet their application-specific needs. Specific bus technologies discussed below include GPIB, USB, PCI, PCI Express, and Ethernet/LAN/LXI.

First, it is important to outline the relevant performance criteria for instrument control buses, in order to set a baseline for evaluation and comparison.

Bandwidth

When considering the technical merits of alternative buses, bandwidth and latency are two of the most important bus characteristics. Bandwidth measures the rate at which data is sent across the bus, typically in MB/s (106 bytes per second). A bus with high bandwidth is able to transmit more data in a given period than a bus with low bandwidth. Most users recognize the importance of bandwidth because it affects whether their data can be sent across the bus to or from a shared host processor as fast as it is acquired or generated and how much onboard memory their instruments will need. Bandwidth is important in applications such as complex waveform generation and acquisition as well as RF and communications applications. High-speed data transfer is particularly important for virtual and synthetic instrumentation architectures. The functionality and personality of a virtual or synthetic instrument is defined by software; in most cases, this means data must be moved to a host PC for processing and analysis. Figure 1 charts the bandwidth (and latency) of all the instrumentation buses examined in this paper.

Latency

Latency measures the delay in data transmission across the bus. By analogy, if we were to compare an instrumentation bus to a highway, bandwidth would correspond to the number of lanes and the speed of travel, while latency would correspond to the delay introduced at the on and off-ramps. A bus with low (meaning good) latency would introduce less delay between the time data was transmitted on one end and processed on the other end. Latency, while less observable than bandwidth, has a direct impact on applications where a quick succession of short, choppy commands are sent across the bus, such as in handshaking between a digital multimeter (DMM) and switch, and in instrument configuration.

Message versus Register-Based Communication

Buses that use message-based communication are generally slower because this mode of communication adds overhead in the form of command interpretation and padding around the data. With register-based communication, data transfer occurs by directly writing and reading binary data to and from hardware registers on the device, resulting in a faster transfer. Register-based communication protocols are most common to internal PC buses, where interconnects are physically shorter and the highest throughput is required. Message-based communication protocols are useful for transmitting data over longer distances and where higher overhead costs are acceptable. It should be noted that the latency and bandwidth metrics are partially dependent whether the bus uses message or register-based communication and so this parameter is partially captured in those metrics.

Long-Range Performance

For remote monitoring applications and for systems that involve measurement over a large geographical area, range becomes important. Performance in this category can be viewed as a tradeoff with latency, because the error checking and message padding added to overcome physical limitations of sending data over longer cables can add delays to sending and receiving the data.

Instrument Setup and Software Performance

Ease of use in terms of instrument setup and software performance is the most subjective criterion examined here. Nonetheless, it is important to discuss. Instrument setup describes the “out of the box” experience and setup time. Software performance relates to how easily users can find interactive utilities or standard programming APIs such as VISA to communicate with and control the instrument.

Robustness of Connector

The physical connector for the bus affects whether it is suitable for industrial applications and whether additional effort will be required to “ruggedize” the connection between the instrument and the system controller. Figure 2 presents photos of several instrumentation bus connectors.

GPIB

The first bus we will look at is the IEEE 488 bus, familiarly known as GPIB (general-purpose interface bus). GPIB is a proven bus designed specifically for instrument control applications. GPIB has been a robust, reliable communication bus for 30 years and is still the most popular choice for instrument control because of its low latency and acceptable bandwidth. It currently enjoys the widest industry adoption, with a base of more than 10,000 instrument models with GPIB connectivity.

With a maximum bandwidth of about 1.8 MB/s, it is best suited for communicating with and controlling stand-alone instruments. The more recent, high-speed revision, HS488, increased bandwidth up to 8 MB/s. Transfers are message-based, often in the form of ASCII characters. Multiple GPIB instruments can be cabled together to a total distance of 20 m, and bandwidth is shared among all instruments on the bus. Despite relatively lower bandwidth, GPIB latency is significantly lower (better) than that of USB and especially Ethernet. GPIB instruments do not autodetect or autoconfigure when connected to the system; though GPIB software is among the best available, and the rugged cable and connector are suitable for the most demanding physical environments. GPIB is ideal for automating existing equipment or for systems requiring highly specialized instruments.

USB

USB (universal serial bus) has become popular in recent years for connecting computer peripherals. That popularity has spilled over into test and measurement, with an increasing number of instrument vendors adding USB device controller capabilities to their instruments.

Hi-Speed USB has a maximum transfer rate of 60 MB/s, making it an attractive alternative for instrument connectivity and control of stand-alone and some virtual instruments with data rates below 1 MS/s. Though most laptops, desktops, and servers may have several USB ports, those ports usually all connect to the same host controller, so the USB bandwidth is shared among all the ports. Latency for USB falls into the better category (between Ethernet at the slow end and PCI and PCI Express at the fast end), and cable length is limited to 5 m. USB devices benefit from autodetection, which means that unlike other technologies such as LAN or GPIB, USB devices are immediately recognized and configured by the PC when a user connects them. USB connectors are the least robust and least secure of the buses examined here. External cable ties may be needed to keep them in place.

USB devices are well-suited for applications with portable measurements, laptop or desktop data logging, and in-vehicle data acquisition. The bus has become a popular communication choice for stand-alone instruments due to its ubiquity on PCs and especially due to its plug-and-play ease of use. The USB Test and Measurement Class (USBTMC) specification addresses the communication requirements of a broad range of test and measurement devices.

PCI

PCI and PCI Express achieve the best bandwidth and latency specifications among all the instrumentation buses examined here. PCI bandwidth is 132 MB/s, with that bandwidth shared across all devices on the bus. PCI latency performance is outstanding; benchmarked at 700 ns, compared to 1 ms in Ethernet. PCI uses register-based communication. Unlike the other buses mentioned here, PCI does not cable to external instruments. Instead, it is an internal PC bus used for PC plug-in cards and in modular instrumentation systems such as PXI, so distance measures do not directly apply. Nonetheless, the PCI bus can be “extended” by up to 200 m by the use of NI fiber-optic MXI interfaces when connecting to a PXI system. Because the PCI connection is internal to the computer, it is probably fair to characterize the connector robustness as being constrained by the stability and ruggedness of the PC in which it resides. PXI modular instrumentation systems, which are built around PCI signaling, enhance this connectivity with a high-performance backplane connector and multiple screw terminals to keep connections in place. Once booted with PCI or PXI modules in place, Windows automatically detects and installs the drivers for modules.

An advantage that PCI (and PCI Express) share with Ethernet and USB is that they are universally available in PCs. PCI is one of the most widely adopted standards in the history of the PC industry. Today, every desktop PC has either PCI slots, PCI Express slots, or both. In general, PCI instruments can achieve lower costs, because these instruments rely on the power source, processor, display, and memory of the PC that hosts them, rather than incorporating that hardware in the instrument itself.

PCI Express

PCI Express is similar to PCI. It is the latest evolution of the PCI standard, much as Hi-Speed USB is to USB. Therefore, much of the above evaluation of PCI applies to PCI Express as well.

The main difference between PCI and PCI Express performance is that PCI Express is a higher bandwidth bus and gives dedicated bandwidth to each device. Of all the buses covered in this tutorial, only PCI express offers dedicated bandwidth to each peripheral on the bus. GPIB, USB, and LAN, divide bandwidth across the connected peripherals. Data is transmitted across point-to-point connections called lanes at 250 MB/s per direction. Each PCI Express link can be composed of a multiple lanes, so the bandwidth of the PCI Express bus depends on how it is implemented in the slot and device. A x1 (by 1) link provides 250 MB/s; a x4 link provides 1 GB/s; and a x16 link provides 4 GB/s dedicated bandwidth. It is important to note that PCI Express achieves software backward compatibility, meaning that users moving to the PCI Express standard can preserve their software investments in PCI. PCI Express also extensible by external cabling.

High-speed, internal PC buses were designed for rapid communication. Consequently PCI and PCI Express are ideal bus choices for high-performance, data-intensive systems where large bandwidth is required, and for integrating and synchronizing several types of instruments.

Ethernet/LAN/LXI

Ethernet has long been an instrument control option. It is a mature bus technology and has been widely used in many application areas outside of test and measurement. 100BaseT Ethernet has a theoretical max bandwidth of 12.5 MB/s. Gigabit Ethernet, or 1000BaseT, increases the max bandwidth to 125 MB/s. In all cases, Ethernet bandwidth is shared across the network. At 125 MB/s Gigabit Ethernet is theoretically faster than Hi-Speed USB, but this performance quickly declines when multiple instruments and other devices are sharing network bandwidth. Communication along the bus is message based, with communication packets adding significant overhead to data transmission. For this reason, Ethernet has the worst latency of the bus technologies featured in this tutorial.

Nonetheless, Ethernet remains a powerful option for creating a network of distributed systems. It can operate at distances up to 85 to 100 m without repeaters and with repeaters has no distance limits. No other bus has this range of separation from the controlling PC or platform. As with GPIB, autoconfiguration is not available on Ethernet/LAN. Users must manually assign an IP address and subnet configuration to their instrument. Like USB and PCI, Ethernet/LAN connections are ubiquitous in modern PCs. This makes Ethernet ideal for distributed systems and remote monitoring. It is often used in conjunction with other bus and platform technologies to connect measurement system nodes. These local nodes may themselves be composed of measurement systems relying on GPIB, USB, and PCI. Physical Ethernet connections are more robust than USB connections, but less so than GPIB or PXI.

LXI (LAN eXtenstions for Instrumentation) is an emerging LAN-based standard. The LXI standard defines a specification for stand-alone instruments with Ethernet connectivity that adds triggering and synchronization features.

Despite the conceptual convenience of designating a single bus or communication standard as the “ultimate” or “ideal” technology, history teaches us that several alternative standards are likely to continue to coexist, since each bus technology has unique strength and weaknesses. Table 1 compiles the performance criteria from the previous section. It should be clear that no single bus is superior across all measures of performance.

	Bandwidth (MB/s)	Latency (µs)	Range (m) (without extenders)	Setup and Installation	Connector Ruggedness
GPIB	1.8 (488.1) 8 (HS488)	30	20	Good	Best
USB	60 (Hi-Speed)	1000 (USB) 125 (Hi-Speed)	5	Best	Good
PCI	132	0.7	Internal PC bus	Better	Better Best (for PXI)
PCI Express	250 (x1) 4000 (x16)	0.7 (x1) 0.7 (x4)	Internal PC bus	Better	Better Best (for PXI)
Ethernet/LAN/LXI	12.5 (Fast) 125 (Gigabit)	1000 (Fast) 1000 (Gigabit)	100 m	Good	Good

Test system developers can exploit the strengths of several buses and platforms by creating hybrid systems. Hybrid test and measurement systems combine components from modular instrumentation platforms such as PXI and VXI and stand-alone instruments that connect across GPIB, USB, and Ethernet/LAN. One key to creating and maintaining a hybrid system is implementing a system architecture that transparently recognizes multiple bus technologies and takes advantage of an open, multivendor computing platform, such as PXI, to achieve I/O connectivity.

The other key to successfully developing a hybrid system is ensuring that the software you choose at the driver, application, and test system management levels is modular. Though some vendors may offer vertical software solutions for specific instruments, the most useful system architecture is one which breaks up the software functions into interchangeable modular layers so that your system is neither tied to a particular piece of hardware or to a particular vendor. This layered approach provides the best code reuse, modularity, and longevity. For example, VISA (Virtual Instrument Software Architecture) is a vendor-neutral software standard for configuring, programming, and troubleshooting instrumentation systems comprising GPIB, VXI, serial (RS232/485), Ethernet, USB, and/or IEEE 1394 interfaces. It is a useful tool because the API for programming VISA functions is similar for a variety of communication interfaces.

With hybrid systems, you can combine the strengths of many types of instruments, including legacy equipment and specialized devices. Despite the appeal of finding a one-size-fits-all solution for instrumentation, reality requires that test engineers fit the instruments and associated bus technologies to their specific application needs.

NI, a leader in automated test, is committed to providing the hardware and software products engineers need to create these next generation test systems.

• Download the instrument driver for your instrument (NI and 3rd Party) from our library of more than 10,000 drivers
• Download a free trial of NI LabVIEW to automate your measurements
• Learn about PXI, the industry-leading modular platform for automated test and measurement