Robert Cocking and Kazuharu Shimamura

ArcInfo Performance across Different Platforms

ArcInfo has been supported across a number of different UNIX platforms for several decades. ArcInfo has been available on the Windows NT platform for only a few years and has become quite popular for a variety of reasons. Unfortunately, there is no good way of judging what NT PC might run ArcInfo equivalently to a particular UNIX computer. This is important for NT PC purchasers since there is such a wide choice of PC hardware. We have put together a benchmark to help judge the relative performance of particular hardware configurations.

THE ArcInfo BENCHMARK

The concept of the ArcInfo Benchmark is quite simple - to test the speed of commonly used ArcInfo processes. Each process was run three times and the average time was calculated.

Datasets

A number of coverages standard to geological mapping were used for this benchmark.

The Processes Tested

Test #1: IMPORT- converts the contours.e00 to a coverage

Test #2: DXFARC - converts the lakes.dxf and streams.dxf to coverages

Test #3: BUFFER - creates 100 metre stream buffers

Test #4: INTERSECT - intersects the stream buffers with the geology coverage

Test #5: DROPITEM - performs 52 dropitems from the geology coverage

Test #6: APPEND - appends the geology coverage with the stream buffers

Test #7: CLEAN - cleans the appended coverage

Test #8: PROJECT - projects the contours coverage from UTM into Transverse Mercator

Test #9: ArcPlot Drawing - draws all coverages in Arcplot

Test #10: ArcEdit Drawing - draws all coverages in ArcEdit

Test #11: EXPORT - converts the appended coverage to E00

Test #12: ARCDXF - converts the contours coverage to DXF format

Test #13: POSTSCRIPT - converts an ArcInfo Graphic file (.gra) to PostScript

The Methodology

The process times were calculated using the [date -time] function to store the process start and end times. The times were converted to seconds and the process start time was subtracted from the process end time. The average time was stored in the time.dat file as well as the total time and the date the benchmark was run. Persons participating in the benchmark were asked to answer a number of questions in the data.dat file relating to the type of system that they performed the benchmark on.

Participants were encouraged to run this benchmark on ArcInfo servers only. Running the benchmark on xterminals will not yield true results of a computer processor. Participants were also encouraged to run this benchmark at a time when there are no others system processes running; thereby ensuring that other programs do not slow down the ArcInfo benchmark.

The Results

Due to time constraints, the sample number for this survey was not large at the time of publishing. By July 1998, the time of the 1998 Esri Users Conference, the data should have grown significantly. Please visit the benchmark web site to review the more advanced findings or to take part in the survey of ArcInfo performance.

It is interesting to note that while the DEC Alpha 3000was slower than the Pentium 200 on most tests, it generally scored better on the import and dropitem tests. 

The Sparc 2 had largely similar performance to the DEC Alpha 3000, but performed significantly worse on the buffer, intersect and clean tests, all processes concerned with topology.

The numbers for this benchmark can be viewed in a Microsoft Excel spreadsheet downloadable from the Geological Survey of Canada website http://www.nrcan.gc.ca/ess/carto/english/benchmark/index.html.

ACKNOWLEDGMENTS

Many thanks to all the folks who ran the benchmark on their own systems: Sonia Talwar, Dwight McCullough, Steve Williams, Gary Labelle, Mike Sigouin, and the many who ran the test after the publication date of this paper.

Robert Cocking
GIS Cartographer, Geological Survey of Canada
Kazuharu Shimamura
GIS Specialist - Surficial Geology, Geological Survey of Canada