New: Oct. 14, 2000

Windows Me Timing Results

The following timing results augment Appendix C—they add an 8th column to Tables C-1 through C-4 and contrast Windows 2000 and Windows Me on the exact same hardware (a 500 MHz Pentium III). The left column shows the Windows 2000 results; this data is exactly the same as the rightmost column of Tables C-1 through C4 and allow a direct comparison of W2000 and Me.

By and large, the timings are comparable and the comparisons to NT/2000 are about the same as for W95/98.

Some comments:

• atouMM (Table C-2) has the same performance advantages (relative to normal sequential file I/O) on Me as on NT/2000, so Microsoft must have done some re-engineering of the Me kernel to improve memory mapped file I/O. I have not repeated the full set of experiments shown in File Processing Performance.
• Also, note the the file copy (Table C-1) and ASCII to Unicode (Table C-2) copy times are slightly longer as I made the tests a bit more rigorous in the may in which I assured that files were not cached from one test run to the next. These increased times are consistent with what is now occurring under Windows 2000.
• I found Me to be less reliable in certain cases than W9x and NT/2000. In particular, the sortMT tests would crash on about 50% of the runs. I had to repeat the test again in order to collect the data. There were also some inexplicable situations in which the file deletes in the timing batch files would cause a C run-time library error message box; I've never seen this happen before. I have not had time to investigate these anomalies; reader experiences would be very helpful.
• Table C-5 is not repeated here; there results are very similar to the Windows 2000 results.

Windows Me Extension to Table C-1. File Copy Performance 

Note: The cpW time is a typo. 18.29 is the correct number is 18.20.

Windows Me Extension to Table C-2. ASCII to Unicode Performance 

Note: The atouMM time is very good relative to sequential file processing. Also, the Windows Me times are considerably higher due to the changes in the timing methodology.

Windows Me Extension to Table C-3. Pattern Searching Performance

Windows Me Extension to Table C-4. File Sorting Performance

Note: I'm at a loss to explain these sorting times; file processing could not explain it. The bulk of the time should be CPU time. However, I repeated the tests and the data is correct.