Reviews Leaderboard Database Reference Search StorageReview Discussion Reliability Survey Search About StorageReview.com Contents

Windows 95 vs Windows 98: A Perspective From Disk Performance

  July 21, 1998 Author: Eugene Ra  
See also our Summer 1998 ATA Drive Roundup
And also our Summer 1998 SCSI Drive Roundup
Maxtor DiamondMax Plus 2500 provided by Maxtor Corp.

Seagate Cheetah 9LP provided by Dirt Cheap Drives.

On June 25th, Microsoft released Windows 98 to the general public. Though it promises better ease-of-use through Internet Explorer's shell integration and better internet connectivity, the core of the OS remains the same as Windows 95 OSR 2.1. But the question is, has disk performance been impacted?

Windows 98 includes an Application Start Tracker that logs the order in which various files are loaded by the system during bootup and the launching of programs. When Win98 users defragment their hard drives, the start log is consulted and some fragmentation is purposefully induced to minimize seeking during bootup and program launches. While this may be quite noticeable in actual use, such a performance increase shouldn't be reflected in WinBench 98 or Threadmark since test files are created and trials are run all in one swoop without any defragmentation/optimization being performed in the interim.

Also included in the new release are some fixes and patches that were previously required by Windows 95 to ensure maximum DMA mode compatibility, specifically REMIDEUP.EXE and SETUP_EX.EXE (the 82371 update). Primarily intended to address ATA removable storage and newer Intel Chipsets, these files also allowed smoother implementation of DMA access using OSR2's native drivers. Windows 98, of course, has all this built-in already. ATA busmastering continues to mature.

Thus, the hypothesis is clear: using the same video driver (a driver for the testbed's Millennium II is not "built-in" to the Win98 installation procedure, but there's one in the driver folder) as before, a basic installation of Windows 98 should yield the same disk performance as a basic installation of Windows 95. To conduct these comparisons, we chose the two top performers in the ATA and SCSI fields, the Maxtor DiamondMax Plus 2500 and Seagate Cheetah 9LP. All tests were performed five times and averaged to yield the results below. Windows NT results are presented for reference purposes.

Ziff Davis WinBench 98 Benchmarks for the Maxtor DiamondMax 2500 *
Benchmark Windows 95 OSR 2.1, FAT32 Windows 98, FAT32 Windows NT 4.0, NTFS
Business Disk WinMark 98 (KB/sec) 1686 1714 2110
Spread Sheet / Database 1602 1630 1918
WordPerfect 2028 2048 2284
Publishing 1510 1538 2014
Browsers 1748 1738 2738
Task Switching 3148 2996 3092
High-End Disk WinMark 98 (KB/sec) 4740 4734 5104
AVS/Express 3.1 2916 2802 2952
Frontpage 97 3698 3744 4332
MicroStation 95 8080 7794 10146
Photoshop 4.0 4452 4892 4254
Premier 4.2 9076 9320 8502
PV-Wave 6.1 3246 3246 3950
Visual C++ 5.0 10420 9362 10440
Disk/Read Transfer Rate (Kb/sec)
Beginning 13800 13780 14900
End 9290 9290 9220
Disk/Read Random Access (ms) 15.0 14.9 14.1
Disk/Read CPU Utilization 8.64 % 9.58 % 5.88 %
Transfer Rate (KB/sec) 12724 12719 14881

Adaptec ThreadMark 2.0 Benchmarks for the Maxtor DiamondMax 2500 *
Benchmark Windows 95 OSR 2.1, FAT32 Windows 98, FAT32 Windows NT 4.0, NTFS
Data Transfer Rate (MB/sec) 7.46 7.58 9.45
Average CPU Utilization 25.42 % 25.53 % 16.97 %

As expected, Win95 and Win98 perform similarly with an ATA drive. There are slight variations here and there, but as a whole the results are close enough to be considered identical. Now let's take a look at the same experiment conducted with the SCSI bus.

Ziff Davis WinBench 98 Benchmarks for the Seagate Cheetah 9LP *
Benchmark Windows 95 OSR 2.1, FAT32 Windows 98, FAT32 Windows NT 4.0, NTFS
Business Disk WinMark 98 (KB/sec) 2112 2174 2498
Spread Sheet / Database 1964 2022 2090
WordPerfect 2620 2712 2776
Publishing 1894 1976 2514
Browsers 2180 2130 3456
Task Switching 2910 2996 2672
High-End Disk WinMark 98 (KB/sec) 5650 5618 5450
AVS/Express 3.1 3666 3582 3736
Frontpage 97 4262 4318 2983
MicroStation 95 9504 9308 12120
Photoshop 4.0 4676 4854 3748
Premier 4.2 12080 12200 9860
PV-Wave 6.1 4052 3946 4742
Visual C++ 5.0 12080 11540 12640
Disk/Read Transfer Rate (Kb/sec)
Beginning 19000 19080 18960
End 12560 12580 12520
Disk/Read Random Access (ms) 8.9 9.0 7.2
Disk/Read CPU Utilization 9.16 % 9.78 % 7.54 %
Transfer Rate (KB/sec) 19053 19065 18902

Adaptec ThreadMark 2.0 Benchmarks for the Seagate Cheetah 9LP *
Benchmark Windows 95 OSR 2.1, FAT32 Windows 98, FAT32 Windows NT 4.0, NTFS
Data Transfer Rate (MB/sec) 10.18 11.53 12.51
Average CPU Utilization 33.91 % 35.51 % 19.38 %

So close! Here, however, we find a fly in the ointment. Threadmark 2.0 (what else did you expect? ) is the culprit here, yielding consistently higher results under Windows 98 than under Windows 95. WinBench 98, on the other hand, presents virtually identical figures for both operating systems. Why? My only guess is that Win98 includes a new driver for the Adaptec 2940U2W used in our test bed, dated in May, while with Win95 I had to install the U2W drivers included with the card, dated in January. Perhaps the new driver includes optimizations for Threadmark. I'll take a look at performance with the Cheetah 9LP running in Win98 with the older SCSI driver to either confirm or rule out this possibility.

At any rate, we've conducted these tests for two reasons. First, we've received several inquiries saying "Ok, I see that the IBM is faster with Windows 95 and the Maxtor is faster with NT, but which would be faster in Win98?" These trials show that the disk performance you can expect from a drive under Win98 will be much closer to results obtained under Win95 than under WinNT. Secondly, the Storage Review has amassed a database of no less than 40 drives using our venerable Windows 95 OSR 2.1 software configuration. Demographically speaking, surfers who frequent computer hardware enthusiast sites such as this one are probably moving to Windows 98 much more quickly than the average user. Naturally, we were concerned with the applicability of our Win95 data to such users and were contemplating moving to Win98 as the standard test configuration. Doing so, however, would have introduce a schism in the database. The figures above, however, show that such a transition is unnecessary; we'll stick with using Win95 as a baseline, and switch to Win98 only when we change other hardware variables (processor, motherboard, ram, etc).

* Note: All reported test results are the average of five trials.


HOME | ARTICLES | LEADERBOARD | PERFORMANCE DATABASE | REFERENCE GUIDE
COMMUNITY | RELIABILITY SURVEY | SUPPORT SR! | ABOUT SR |

Copyright © 1998-2005 StorageReview.com, Inc. All rights reserved.
Write: Webmaster