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

Ultra SCSI vs Ultra2 SCSI: The Seagate Barracuda 9LP ST39173LW

  July 27, 1998 Author: Eugene Ra  
See also our Summer 1998 SCSI Drive Roundup
See also our Drive Cooler Roundup
See also our Seagate Barracuda 9LP ST39173W review.

Seagate Barracuda 9LP provided by Seagate Technology, Inc.
Adaptec ACK-W2W-5IT provided by Adaptec, Inc.

In the near future, the Storage Review plans to embark on a comprehensive survey of the available UltraWide and Ultra2 SCSI controllers. Though the differences may be rather small, we plan to evaluate variance in controllers in addition to presenting features of the controllers themselves. To do so, of course, we need to standardize on a single drive. Seagate has kindly loaned SR an Ultra2 SCSI Barracuda 9LP ST39173LW.

Seagate Barracuda 9LPSeveral readers have written regarding the use of Ultra2 drives on regular UW controllers. Does this impact performance? By how much? The educated guess is no, today's Ultra2 drives should not be hindered by normal use on a UW controller (at least not performance-wise but one does sacrifice U2's longer cable lengths). The controller may be a bottleneck only if a user has two or more drives operating at the same time both pumping out data at their maximum sequential transfer rates.

The reference controller used in our SCSI drive tests is the Adaptec AHA-2940U2W, a versatile SCSI host adapter that (with the assistance of a bridge chip) allows all four of its connections to be used simultaneously. There are two internal 68-pin connectors, one for Ultra2 (LVD, or Low-Voltage Differential) drives and another for UltraWide (SE, or Single-Ended) devices. This allows for easy booting from a UW hard disk while allowing the blank, freshly formatted test disk to operate in U2W mode if so desired.

We decided to take a look at the measured performance of the ST39173LW on the U2W and UW connectors to see if any difference at all could be detected by benchmarks. According to Seagate, the transfer rate of the Barracuda 9LP tops out at 17 megabytes per second, far lower than UW and U2W's 40 MB/sec and 80 MB/sec respectively. The Ultra2 specification calls for active termination on the cable rather than termination by the last device in the chain. Though this increases cable expense, it simplifies termination issues, especially when you find yourself swapping drives in and out as often as I do . The bottom of the Ultra2 9LP has the same row of jumpers as all Seagate SCSI drives, including two pins at the end labeled "TE." According to the manual, however, no active termination circuitry is built in to the drive. Users really must either put it in the middle of a SCSI chain or use an actively-terminated cable.

The 2940U2W ships with a rather ungainly-looking blue Ultra2 cable that features 5 positions and active termination. This cable was used to attach the 9LP to the LVD connector on the host adapter when performing U2W tests. For UW tests we used the Adaptec ACK-W2W-5IT, a wide SCSI cable with active termination built into the end. In each case, U2W or UW operation was confirmed by the host adapter's BIOS indicators, which reported "Ultra2/LVD" and "Ultra/SE" modes respectively during boot-up.

ZDBop's WinBench 98, along with Adaptec's ThreadMark 2.0, were both run on the unit in Windows 95 OSR 2.1. The drive was partitioned into a single volume of maximum size.

Ziff-Davis WinBench 98 under Windows 95 OSR 2.1 using FAT 32 *
BenchmarkSeagate Barracuda 9LP Ultra/SESeagate Barracuda 9LP Ultra2/LVD
Business Disk WinMark 98 (KB/sec) 1664 1660
Spread Sheet / Database 1500 1472
WordPerfect 2060 2066
Publishing 1546 1564
Browsers 1696 1690
Task Switching 2298 2288
High-End Disk WinMark 98 (KB/sec) 4682 4700
AVS/Express 3.1 2974 3052
Frontpage 97 3780 3732
MicroStation 95 7988 8070
Photoshop 4.0 3864 3796
Premier 4.2 10128 10380
PV-Wave 6.1 3194 3218
Visual C++ 5.0 10140 10180
Disk/Read Transfer Rate (KB/sec)
Beginning 15100 15100
End 9638 9640
Disk/Read Random Access (ms) 11.7 11.6
Disk/Read CPU Utilization 8.03 % 7.23 %
Transfer Rate (KB/sec) 15130 15137

Adaptec ThreadMark 2.0 under Windows 95 OSR 2.1 using FAT 32 *
BenchmarkSeagate Barracuda 9LP Ultra/SESeagate Barracuda 9LP Ultra2/LVD
Data Transfer Rate (MB/sec) 7.72 7.94
Average CPU Utilization 28.32 % 28.26 %

WinBench yielded results as expected; in either mode, the 9LP delivered the same results. Business and High-End WinMarks were practically identical. ThreadMark, however, consistently reported U2W results 3% greater than those of UW operation. Let's take a look at the same tests run under Windows NT:

Ziff-Davis WinBench 98 under Windows NT 4.0 using NTFS *
BenchmarkSeagate Barracuda 9LP Ultra/SESeagate Barracuda 9LP Ultra2/LVD
Business Disk WinMark 98 (KB/sec) 1948 1950
Spread Sheet / Database 1640 1636
WordPerfect 2206 2210
Publishing 1900 1908
Browsers 2782 2808
Task Switching 2170 2180
High-End Disk WinMark 98 (KB/sec) 4338 4414
AVS/Express 3.1 2852 2868
Frontpage 97 3066 3040
MicroStation 95 10264 11090
Photoshop 4.0 2878 2864
Premier 4.2 7920 8178
PV-Wave 6.1 2808 4002
Visual C++ 5.0 10016 10322
Disk/Read Transfer Rate (KB/sec)
Beginning 14860 14820
End 9634 9638
Disk/Read Random Access (ms) 9.7 9.7
Disk/Read CPU Utilization 5.92 % 5.97 %
Transfer Rate (KB/sec) 14954 14989

Adaptec ThreadMark 2.0 under Windows NT 4.0 using NTFS *
BenchmarkSeagate Barracuda 9LP Ultra/SESeagate Barracuda 9LP Ultra2/LVD
Data Transfer Rate (MB/sec) 9.66 9.96
Average CPU Utilization 14.70 % 15.41 %

Again, WinBench reports identical performance. ThreadMark remains stubborn, still reporting a 3% increase in U2W mode. What can we make of this? Ultra2 SCSI's maximum burst transfer rate is 80 MB/sec while standard UW's burst is 40 MB/sec. Could this be the difference? We went ahead and took a quick look at burst read transfer rates using TestaCD Lab's HD-Tach, v2.52:

TestaCD Lab's HD-Tach, v2.52 *
BenchmarkSeagate Barracuda 9LP Ultra/SESeagate Barracuda 9LP Ultra2/LVD
Burst Read Transfer Rate (MB/sec) 25.7 38.5

Yes, here's the one test where U2W operation truly out-raced UW mode, by a factor of 50%. U2W burst reads from the Barracuda 9LP topped out at 38.5 MB/sec, probably beyond the reach of UW operation with any drive due to overhead.

Is ThreadMark sensitive to the difference in UW and U2W burst speeds? Preliminary tests of the Cheetah 9LP's difference in U2W and UW chains also indicate a 3% difference in speed. This raises another question: How does the built-in buffer in a hard disk affect ThreadMark results? Does the 4MB Barracuda 9LP significantly outperform the 1MB version? Adaptec takes pride in having ThreadMark require a large amount of disk space during testing in an effort to defeat OS-level caching. To what extent does this affect the disk-controller caching subsystem? Yet another test added to our queue!

It should be clear from these tests that prospective purchasers of a U2W hard drive should not split hairs over putting the drive on a perfectly good UW controller. A 3% difference (especially in ThreadMark) is miniscule, and as expected, not noticeable at all in subjective usage. So, as long as you have means of terminating the SCSI chain (either on the cable or on another drive), go ahead and get the U2W version of the drive in question. You'll have the twin peace-of-mind benefits of knowing that you're getting maximum performance out of the drive today and knowing that the drive will be ready to be integrated into a pure U2W system in the future.

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


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