Home | Polski Polski
Attached
Go back

A sidebar to the article “Spreadsheets” published in Byte Summer 1987 issue, pp. 72.

Benchmarking Spreadsheets

The Savage benchmark, as described in “The 8087/80287 Performance Curve” by Stephen S. Fried in BYTE’s Inside the IBM PCs (Fall 1985, page 70), is ideal for testing a program’s ability to handle floating-point calculations.

The Savage benchmark involves three function pairs: tangent and arctangent, exponent and natural log, and square root and square, all of which should combine with each other to form an identity function. In addition, the operand of the function is varied from 1 to 1000. The Savage benchmark used in these tests consisted of 1000 rows of a single cell containing the following:

tan(arctan(exp(ln(sqrt((prev__cell)^2)))))+1

Prev__cell refers to the cell immediately above. The initial value of prev__cell is 1. After 1000 iterations, the value of 1001 is subtracted from the resulting output to obtain the error, which is compounded during each iteration.

I also used three other tests based on an expanded version of the simple worksheet used in previous BYTE spreadsheet reviews. The worksheet consists of 100 columns and 25 rows, where each cell is the product of 1.001 times the cell to its left. The cells in the first column are the products of the last cells of the previous row times 1.001. The value of the cell in the first row of the first column is 1.

The first test on this worksheet was a simple recalculation. The second was the Scroll Right test, which measured how long it would take to scroll from the upper-left cell to the top cell of the rightmost column (column 100). The third test measured how long it takes to insert a row at the top of the 100-column by 25-row worksheet with the automatic recalculation turned off.

I performed each speed test three times on a plain 512K-byte IBM PC and a 512K-byte Macintosh and averaged the results. No special programs were resident in the background of each system, and I ran each spreadsheet in its standard configuration. Note that I did not test the programs on a PC AT-style IBM PC; based on past experience, I’d estimate that users of those machines can expect times to be approximately one-third of those for the standard PC.

These tests were designed to measure the performance of the programs in some common spreadsheet operations. Like all benchmark tests, they should be interpreted relative to other data.



 
Page added on 22nd September 2004.

Copyright © 2002-2005 Marcin Wichary
Printable version | Contact | Site map