For raw computing power, super computers rely on massively parallel processors to break an enormous task into more finite elements that multiple slower processors can handle. Similarly, sophisticated high-performance data acquisition hardware is capable of producing data streams that exceed a Gbyte/sec, well beyond what any single collection system is currently capable of handling. One solution is to multiplex these massive data streams into multiple, slower data streams that existing computer technologies can handle. A key technology today that is enabling massively parallel data acquisition architectures is PCI expansion technology. This article provides an overview of PCI expansion technology and how it assists with real-time data acquisition.
Inside a typical server or desktop computer, the PCI bus serves as a major data highway and traffic cop for data flowing in and out of the computer memory. In many systems, the network, monitor, and storage subsystem connect to a slot on the PCI bus. High-performance systems offer segmented PCI buses to allow the system architect to isolate traffic between the various I/O and processing devices. PCI expansion technology allows the systems architect to add literally hundreds of additional PCI slots to any PCI-based computer.
The basic architecture of a PCI expansion system includes a host PCI card that plugs into an available PCI slot – a cable and a backplane offering multiple additional PCI slots and segments in a separate but connected (via cable) chassis. Each segment comprises a group of two or more PCI slots that connect via a bridge to subsequent segments in the overall system. A segment isolates I/O traffic between cards within that segment without affecting the performance of other segments whenever the host processor or other segments do not require services such as interrupt handling. Designers can configure PCI expansion systems as a Star, Daisy Chain, or Hybrid configuration (see Figure 1). A system suffers a performance penalty, as much as 30 percent or more, for each PCI bridge and each PCI Host card/cable or Hop in the architecture. Thus, the Star configuration offers the best architecture, fewest hops, when the application requires high-performance data transfer between the host system and the PCI card located in one of the expansion units. However, the Star configuration limits the number of PCI Host cards that a designer can install in the host computer. When host communications is not an issue, then the daisy chain or hybrid architectures may offer greater packaging flexibility; allowing up to 256 expansion units.
Modern radar, signal intelligence, and imaging systems are examples of systems capable of generating data rates beyond a Gbyte/sec. Both economic and physical restrictions prevent the system from processing this data in real-time for most applications. It is also undesirable to employ compression or real-time filtering, as these techniques may not be loss-free (i.e., you could lose data), and may result in additional performance bottlenecks. Ultimately, ultra-high-speed data acquisition systems require a recording solution that can record raw data for long durations at very high data rates allowing the system to process the data later (see Figure 2). When the data source can send data directly to the storage media without host interaction, it avoids the performance penalties of both the PCI expansion technology and the interaction with the host computer. Under these circumstances, it is possible to achieve real-time data rates and to construct a massively parallel, high-speed data acquisition system.
Figure 2
Figure 3 indicates how a high-speed camera (e.g., operating at 1 Gbyte/sec) is capable of outputting multiple channels of image data. Each of the recording chassis’ has a PCI expansion backplane, data acquisition board(s), suitable for the camera and a disk recording system. The master recorder, which is also the host computer, has visibility to all of the recorded data. With this architecture, additional performance and capacity is possible for faster cameras by simply adding additional expansion chassis’ to the configuration. During the data collection, the camera records each channel independently in a hardware-to-hardware data transfer directly to the PCI expansion systems. As the host is not involved in the actual recording, it is possible to architect a system that has very deterministic performance characteristics. Thus, if the sustained data rate of any single recorder is, say 200 Mbytes/sec, achieving 1 GByte/sec is simply a matter of configuring five recorders (5 x 200 Mbytes/sec = 1 Gbyte/sec) to a single host using PCI expansion technology. Once the system collects the data, the master recorder can read the recorded data for processing.
Figure 3
System designers must architect modern data acquisition systems on building blocks that provide scalability to address a wide range of data rates and capacities at the lowest possible cost. Although the scale continues to move from Mbytes/sec to Gbytes/sec and beyond, the fundamental need for scalability remains rooted in the need to break high speed data sources into manageable chunks or streams that a system can independently record. PCI expansion technology offers systems integrators a readily available solution for architecting systems that can both tackle today’s toughest data acquisition requirements and provide scalability for future requirements.
. . . . .
Thomas Skrobacz, vice president of business development for Conduant, develops business strategies and procedures, and coordinates implementation. Before joining the company, Thomas served at IBM during a 16-year period. While with IBM, he pioneered software strategies to deliver distributed mainframe systems software worldwide and developed security standards for migration to the Internet. Thomas developed and deployed technologies, policies, and procedures to conduct the single largest systems integration test ever performed at IBM. He is a certified project management professional and an award-winning entrepreneur. Thomas received his B.S. in computer science from the University of Maryland at College Park.
Conduant began in 1996 with a vision for creating a high-speed, long duration, disk recorder capable of capturing long duration “events” for instrumentation, development, and testing. Applications that use the recorder have expanded to include radio astronomy, radar/sonar gathering, signal intelligence (sigint), surveillance, signal processing, GPS downloads, and other instrumentation applications. Other applications such as high-resolution imaging, digital cinematography, and atmospheric research also use the Conduant recording system. The company’s founders Phil Brunelle and Ken Owens have collectively more than 40 years of experience in designing disk drive hardware and software for major disk manufacturers. The Denver Business Journal recently named Conduant as one of the fastest growing small companies in Colorado
For further information visit www.conduant.com.
. . . . .
Copyright © 2003 Embedded Computing Design. Reproduction in whole or part without permission is prohibited. All rights reserved. 7/2003 www.embedded-computing.com
|
