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

Second Anniversary!
3/13/2000 - SR`s IOMeter Tests
3/14/2000 - The New Database

  March 16, 2000 Author: Eugene Ra  

The debut of Intel's IOMeter here at StorageReview.com has stirred the pot a bit when it comes to the drive hierarchy with which we've all become a bit comfortable. Whether dealing with SCSI vs. ATA, contrasts between top SCSI drives, or the battle for ATA supremacy, IOMeter has something new to say. Let's take a look at some of these findings:

SCSI's Preeminence

IOMeter - Workstation Access Pattern - Total I/Os per second
Linear
Seagate Barracuda 9LP (9.1 GB U2W-SCSI) - 95.68 |
Western Digital Expert (27.3 GB ATA-66) - 86.24 |
Very LightLight
Seagate Barracuda 9LP (9.1 GB U2W-SCSI) - 112.60 |
Western Digital Expert (27.3 GB ATA-66) - 93.12 |
Western Digital Expert (27.3 GB ATA-66) - 118.06 |
Seagate Barracuda 9LP (9.1 GB U2W-SCSI) - 112.63 |
ModerateHeavy
Seagate Barracuda 9LP (9.1 GB U2W-SCSI) - 152.10 |
Western Digital Expert (27.3 GB ATA-66) - 136.58 |
Seagate Barracuda 9LP (9.1 GB U2W-SCSI) - 183.44 |
Western Digital Expert (27.3 GB ATA-66) - 151.22 |


A comparison between the slowest SCSI drive we've tested, the Seagate Barracuda 9LP, and the fastest ATA unit so far (more on this later), the Western Digital Expert WD273BA, yields a conclusion that may be startling to some: In four out of five Workstation Access Pattern cases, the SCSI drive comes out on top. Though the Expert owns the faster sequential transfer rate (23 MB/sec vs. 15 MB/sec), the Barracuda still manages to come out on top in the Linear Load (the lightest Load possible) scenario by a margin of 11%. When the Load reaches Heavy (pretty severe), the disparity increases to 21%. Remember, this is a 7200rpm SCSI drive two-generations removed from the current generation. The difference only increases when one considers top-tier 10k rpm drives. Speaking of which...

The 10,000rpm Grudge Match

IOMeter - Workstation Access Pattern - Total I/Os per second
Linear
Quantum Atlas 10k II (36.7 GB Ultra160/m SCSI) - 129.39 |
Seagate Cheetah 36LP (36.7 GB Ultra160/m SCSI) - 120.98 |
Very LightLight
Seagate Cheetah 36LP (36.7 GB Ultra160/m SCSI) - 152.57 |
Quantum Atlas 10k II (36.7 GB Ultra160/m SCSI) - 145.06 |
Seagate Cheetah 36LP (36.7 GB Ultra160/m SCSI) - 152.26 |
Quantum Atlas 10k II (36.7 GB Ultra160/m SCSI) - 145.16 |
ModerateHeavy
Quantum Atlas 10k II (36.7 GB Ultra160/m SCSI) - 211.60 |
Seagate Cheetah 36LP (36.7 GB Ultra160/m SCSI) - 204.66 |
Quantum Atlas 10k II (36.7 GB Ultra160/m SCSI) - 252.59 |
Seagate Cheetah 36LP (36.7 GB Ultra160/m SCSI) - 245.35 |

According to IOMeter, the Atlas 10k II is still king. The margin by which it bests the Cheetah 36LP, however, isn't nearly as great as that indicated by WinBench 99. Under the lightest possible Load (Linear), the Atlas comes out on top by a margin of 7%. Under a Heavy Load, however, the difference lessens to just 3%. Interestingly, in two of the lighter Load levels, the Cheetah 36LP actually pulls ahead by 5%. The trend generally holds for StorageReview.com's two other Access Patterns (File Server and Database): The Atlas' greatest victories arrive in under a Linear Load with margins shrinking considerably. This likely indicates that while the Atlas possesses the best sheer mechanical design, the Cheetah possesses more refined electronics/algorithms that allow it to shine under heavier Load. Our anticipation of the X15 only grows .

The Importance of Access Patterns

IOMeter - File Server Access Pattern - Total I/Os per second
Linear
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 88.66 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 81.15 |
Very LightLight
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 112.73 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 104.95 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 112.77 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 105.01 |
ModerateHeavy
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 162.02 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 154.80 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 204.88 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 185.60 |

IOMeter - Workstation Access Pattern - Total I/Os per second
Linear
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 97.85 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 92.91 |
Very LightLight
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 118.69 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 112.04 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 118.17 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 111.88 |
ModerateHeavy
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 170.68 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 149.72 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 207.75 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 176.61 |

IOMeter - Database Access Pattern - Total I/Os per second
Linear
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 89.01 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 82.50 |
Very LightLight
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 112.26 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 106.58 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 112.60 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 106.45 |
ModerateHeavy
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 162.42 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 156.44 |
Seagate Cheetah 4LP (4.5 GB UW-SCSI) - 202.20 |
Quantum Atlas V (36.7 GB Ultra160/m SCSI) - 188.88 |

When it comes to a straight random access test (Database Access Pattern, Linear Load), the 7200rpm Quantum Atlas V's superior seek time (6.3ms) allows it to overcome its rotational latency handicap to clock in faster than the 10,000rpm, 7.7ms Cheetah 4LP. The Atlas retains its advantage in the lighter Database levels but slips behind the Cheetah in heavier Loads, presumably due to the same refined algorithms that allow the Cheetah 36LP to approach the Atlas 10k II in performance. This same relationship holds true for the Atlas V vs. Cheetah 4LP under the File Server Pattern. When it comes to Workstation tests, however, the 4LP amazingly defeats the Atlas V in all tests, including Linear, Light, and Very Light Loads. This illustrates the importance of the three discrete Access Patterns that we've selected. At a given Load under a given Access Pattern, one drive may prevail. This does not mean that the same drive would come out on top at the same Load under a different Access Pattern.

Current-Generation 7200rpm ATA Drives

IOMeter - Workstation Access Pattern - Total I/Os per second
Linear
Seagate Barracuda ATA II (20.4 GB ATA-66) - 86.74 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 85.06 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 81.12 |
Very LightLight
Seagate Barracuda ATA II (20.4 GB ATA-66) - 91.01 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 88.91 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 85.02 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 108.19 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 106.50 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 102.06 |
ModerateHeavy
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 124.62 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 118.95 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 114.30 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 136.03 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 131.87 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 122.97 |


Except under the lightest Loads, the Maxtor DiamondMax Plus 40 retains its lead over the competition from Western Digital and Quantum in Workstation Loads. Interestingly, in the scenario of an ATA File Server (not shown here, but a comparison may easily be set up through our versatile benchmark database), the Barracuda ATA II manages to top the Maxtor.

WinBench 99 Access Times vs. IOMeter Access Times

Linear Response Time
Drives IOMeter WinBench 99
Maxtor DiamondMax Plus 6800 (27.2 GB ATA-66) 12.61 ms 13.6 ms
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) 12.95 ms 14.0 ms

In most cases, the access times reported by IOMeter (Database Access Pattern, Linear Load, Average I/O Response Time) and WinBench 99 are quite similar, with IOMeter scores being just a tad greater (perhaps by half a millisecond) due to the 8k transfer performed after each access. An interesting exception, however, occurs with Maxtor's two latest DiamondMax Plus drives, the +6800 and the +40. Here we find the access times reported by IOMeter to be a full millisecond less than the figures turned in by WinBench. Notice how the +6800 is actually faster than the +40 in this regard, which leads us to our next point...

The Fastest ATA Drives

IOMeter - Workstation Access Pattern - Total I/Os per second
Linear
Maxtor DiamondMax Plus 6800 (27.2 GB ATA-66) - 87.23 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 86.74 |
Western Digital Expert (27.3 GB ATA-66) - 86.24 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 85.06 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 81.12 |
Very LightLight
Western Digital Expert (27.3 GB ATA-66) - 93.12 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 91.01 |
Maxtor DiamondMax Plus 6800 (27.2 GB ATA-66) - 90.81 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 88.91 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 85.02 |
Western Digital Expert (27.3 GB ATA-66) - 118.06 |
Maxtor DiamondMax Plus 6800 (27.2 GB ATA-66) - 110.26 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 108.19 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 106.50 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 102.06 |
ModerateHeavy
Western Digital Expert (27.3 GB ATA-66) - 136.58 |
Maxtor DiamondMax Plus 6800 (27.2 GB ATA-66) - 126.41 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 124.62 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 118.95 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 114.30 |
Western Digital Expert (27.3 GB ATA-66) - 151.22 |
Maxtor DiamondMax Plus 6800 (27.2 GB ATA-66) - 137.69 |
Maxtor DiamondMax Plus 40 (41.0GB ATA-66) - 136.03 |
Seagate Barracuda ATA II (20.4 GB ATA-66) - 131.87 |
Western Digital Caviar 7200rpm (20.5 GB ATA-66) - 122.97 |


Here we find two older-generation ATA drives, the DiamondMax Plus 6800 and the Western Digital Expert WD273BA, actually topping today's premiere units in overall performance. The Expert, in particular, sets itself apart as the fastest ATA drive we've yet tested in IOMeter. This goes to show that IOMeter at varying Load levels is indeed more than a simple access time test. Though both DiamondMax Plus drives handily turn in better access times in IOMeter than the Expert, the latter prevails in most IOMeter results. The importance of finely tuned firmware, electronics, and algorithms should not be underestimated.

This concludes our look at some of the more interesting IOMeter cases. Next, we'll conclude our look at the new-testbed results by putting both WinBench 99 and IOMeter results into perspective.

 Concluding Thoughts...


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