William Graham, Michael Adams

Employees of NASA's Commercial Remote Sensing Program (CRSP) Office at the John C. Stennis Space Center in Mississippi recently compared 64-bit and 32-bit computing for digital image processing (DIP) and geographic information systems (GISs). Using ERDAS Imagine 8.2 and ArcInfo, operation times for frequently used DIP procedures were recorded and compared to the operation times achievable on a typical 32-bit platform. This evaluation was application- oriented, examining the speed of operation of the two platforms rather than the differences in hardware. The authors and technical assistants made every effort to standardize operating conditions for both platforms. DIP/GIS operations were timed using session log times or by explicit command-line coding. Upon examination of the results, the 64-bit platform required 49% less time to complete a sample of DIP/GIS procedures than the 32-bit platform. As a capstone to this examination, three-dimensional fly-throughs were obtained by capturing successive images from slightly different viewing perspectives. These images were then viewed in succession using a movie emulation tool.

The computer industry is currently undergoing an important transition from 32-bit architectures to 64-bit architectures. This is a new area of computing for many GIS users and software vendors. These systems incorporate a new generation of computer chips, new versions of the UNIX operating system, and new software tools and applications. Ultimately, a new class of user applications will be developed that take advantage of the enhanced features offered by 64-bit systems. Digital Equipment Corporation (DEC), which first introduced 64-bit workstations into the marketplace in 1992, was particularly interested in exploring the implications of this new 64-bit architecture for GIS users, both in terms of what could be done faster and what could now be done that was not possible in a 32-bit world. NASA's Commercial Remote Sensing Program at the John C. Stennis Space Center in southern Mississippi was identified as a GIS user organization meeting DEC's evaluation criteria: the group is regularly engaged in particularly complex and compute-intensive applications, is recognized as being unbiased, and is an active user of the industry-leading GIS and remote sensing software applications.

The CRSP agreed to evaluate a DEC 64-bit UNIX workstation in a controlled, production-type environment. The general context of the evaluation was to identify the possible benefits of 64-bit processing related to typical DIP/GIS production flow. From a CRSP perspective, two benefits were quickly determined. A primary benefit to DIP/GIS production managers would be the increased productivity and subsequent cost efficiency provided by the reduction in processing times. A benefit to the marketing managers of DIP/GIS organizations would be increased product attractiveness, and ultimate professional savvy, of displays or demonstrations on the 64- bit platform. With its smooth transitions and almost imperceptible flicker, 64-bit processing would enable truly effective virtual reality demonstrations.

For the purposes of the evaluation, the CRSP system administrators configured two identical working environments to compare the two processors independently of network or server traffic. Both the 32-bit computer (a.k.a. "GATOR") and the 64-bit computer (a.k.a. "NEPTUNE") had network and server connections only to facilitate data transfer during the initial setup of this evaluation, not for data or software access during processing. Both systems were configured as stand-alone systems, with data and software mounted and accessed from the two systems' local hard drives. Table 1 provides a brief listing of the equipment and data used during this evaluation.

The raster data used were collected by a NASA sensor called the Airborne Terrestrial Applications Sensor (ATLAS). The file size was roughly 40 megabytes: 1100 (columns) x 1800 (rows) x 12 spectral channels stored as double-precision floating point data. The raster digital elevation model (DEM) corresponds to the same area as the ATLAS data; however, the DEM is a single-channel file and is stored as an unsigned 16-bit integer. The DEM and the other ancillary vector data consisted of land and cultural features digitized or otherwise developed from USGS 1:24,000 topographic maps.

A panel of CRSP remote sensing/geographic information system (RS/GIS) scientists developed a representative, albeit less than comprehensive, list of frequently used and time-consuming DIP/GIS procedures. The evaluation focused on the speed of operation of those procedures. Operation times for raster procedures were obtained from the Imagine Session Log. The vector operations were written into Arc Macro Language (AML) programs with time-reporting commands coded immediately before and after execution of the various Arc commands. Each operation was repeated three times on each machine to confirm proper hardware performance and to ensure consistent results. In addition to this concentrated study, NEPTUNE was also used by several RS/GIS scientists to perform regular production-type operations (e.g., image classifications, spatial and radiometric enhancements, map compositions, ArcEdit and ArcPlot sessions).

Overall performance of the two platforms confirmed expectations: NEPTUNE out-performed GATOR. However, this outcome was not without exception. Table 2 lists the results of ten DIP/GIS procedures.

The operation times listed in Table 2 represent the average speed of operation for three iterations of each operation. The greatest variation for any three iterations on either platform was 3 seconds. The percent difference column represents the percentage of GATOR time in which NEPTUNE was able to complete the same operation. In the case of Operations #1 and #2, the image classification and grid-to-polygon conversion operations required more time on NEPTUNE than on GATOR; no explanation is given for this result.

NEPTUNE was also used in a production-type environment. This portion of the evaluation was not subject to the rigid timing of operations, but rather judged the ease of operation and the individual perceptions of 64-bit computing. In three months of nearly continuous operation, NEPTUNE was well-received and, in most cases, preferred over the other 32-bit workstations. During demonstrations to visiting officials, NEPTUNE's exceptional speed of operation provided a time-efficient platform for displaying soft-copy and virtual reality illustrations.

NEPTUNE's primary impediment to full production capabilities was the lack of disk space. The demonstration unit provided to the CRSP was equipped with a total of six gigabytes of storage, which was quickly consumed by the software (Imagine and ArcInfo) and the high-resolution data. Due to the high spatial and/or spectral resolution of commonly used CRSP data sets, individual file sizes frequently exceed 500 megabytes. Consequently, standard CRSP working environments require four to eight gigabytes of data storage for each project. NEPTUNE also suffered from decreased speed of operation and occasional "crashes" resulting from several operations being processed simultaneously.

NEPTUNE's 64-bit processing out-performed GATOR's 32-bit processing, reducing the overall processing time by 49% and therefore increasing productivity and subsequent cost efficiency. However, two operations related to raster environments required more time on the 64-bit platform than on the 32-bit platform. From a production perspective, quick file handling and rapid display capabilities made 64-bit processing the preferred choice for general production and display demonstrations over the 32-bit platform. Database processing in the Arc environment seemed sluggish in comparison to other operations. Access to larger amounts of disk storage would have been preferred based on the standard operating requirements of CRSP projects. Additional disk storage arrays would have supplemented NEPTUNE's already high production viability. Applications software providers, such as ERDAS and Esri, will need to re-tailor their processing environments in order to fully utilize 64-bit computing.

In evaluating the benefits of the 64-bit computing architecture, we tried to identify tasks and applications that are particularly time-consuming. For example, building movie sequences from images derived from the TINS module in ArcInfo has relevance for local and regional government environments where a 4-D GIS "movie" can help non-technical officials and the public understand complex spatial trade-offs. These "movies," which traditionally can take many hours to create, are well suited to the higher performance of 64-bit systems. Another application that we did not test but feel may have significant benefits for 64-bit computing is shortening turn-around time for processing ad-hoc emergency imagery to less than 24 hours from initial request to acquisition to final delivery.

Portions of this work were supported by the NASA Office of Space Access and Technology, Commercial Remote Sensing Program Office, John C. Stennis Space Center, Mississippi.