Hitachi Deskstar 7K1000 in RAID 0: Is Two Terabytes really better than One?
by Gary Key on April 19, 2007 12:15 AM EST- Posted in
- Storage
Hardware Setup
Our current test bed reflects a common level system that has been in use for a year now. Based upon the continuing proliferation of dual core processors and future roadmaps from AMD and Intel signifying new chipset and processor designs, we will revise our test bed shortly to reflect these changes in the marketplace. This change will also allow us to expand our real world multitasking benchmarks in the near future while providing a stable platform for the next few months. We are currently conducting preliminary benchmark testing under Vista with both 2GB and 4GB memory configurations. We will offer real-world Vista benchmarks once the driver situation matures but iPeak results will continue to be in XP as the application is not compatible with Vista.
Test Setup - Software
With the variety of disk drive benchmarks available, we needed a means of comparing the true performance of the hard drives in real world applications. While we will continue to utilize HDTach and PCMark05 for comparative benchmarks our logical choice for application benchmarking is the Intel iPeak Storage Performance Toolkit version 3. We originally started using this storage benchmark application in our Q2 2004 Desktop Hard Drive Comparison. The iPeak test can be designed to measure "pure" hard disk performance, and in this case we kept the host adapter consistent while varying the hard drive models. The idea is to measure the performance of individual hard drives with a consistent host adapter.
We utilize the iPeak WinTrace32 program to record precise I/O operations when running real world benchmarks. We then utilize the iPeak AnalyzeTrace program to review the disk trace file for integrity and ensure our trace files have properly captured the activities we required. Intel's RankDisk utility is used to play back the workload of all I/O operations that took place during the recording. RankDisk generates results in a mean service time in milliseconds format; in other words, it gives the average time that each drive took to fulfill each I/O operation. In order to make the data more understandable, we report the scores as an average number of I/O operations per second so that higher scores translate into better performance in all of our iPeak results. While these measurements will provide a score representing "pure" hard drive performance, the actual impact on the real world applications can and will be different.
The drive is formatted before each test run and three tests are completed on each drive in order to ensure consistency in the benchmark results. The high and low scores are removed with the remaining score representing our reported result. We utilize the NVIDIA nF4 SATA ports along with the NVIDIA IDE-SW driver to ensure consistency in our playback results when utilizing NCQ, TCQ, or RAID settings. We will test with AAM and NCQ turned on with our Deskstar 7K1000 unit as AAM does not noticeably impact performance and this drive generally performs better with NCQ on in the majority of tests.
Our iPeak tests represent a fairly extensive cross section of applications and usage patterns for both the general and enthusiast user. We will continually tailor these benchmarks with an eye towards the drive's intended usage and feature set when compared to similar drives. In essence, although we will report results from our test suite for all drives, it is important to realize a drive designed for PVR duty will generate significantly different scores in our gaming benchmarks than a drive designed with gaming in mind such as the WD Raptor. This does not necessarily make the PVR drive a bad choice for those who capture and manipulate video while also gaming. Hopefully our comments in the results sections will offer proper guidance for making a purchasing decision in these situations. Our iPeak Test Suite consists of the following benchmarks.
VeriTest Business Winstone 2004: trace file of the entire test suite that includes applications such as Microsoft Office XP, WinZip 8.1, and Norton Antivirus 2003.
VeriTest Multimedia Content Creation 2004: trace file of the entire test suite that includes applications such as Adobe Photoshop 7.01, Macromedia Director MX 9.0, Microsoft Windows Media Encoder 9.0, Newtek Lightwave 3D 7.5b, and others.
AVG Antivirus 7.1.392: trace file of a complete antivirus scan on our test bed hard drive.
Microsoft Disk Defragmenter: trace file of the complete defragmentation process after the operating system and all applications were installed on our test bed hard drive.
WinRAR 3.51: trace file of creating a single compressed file consisting of 444 files in 10 different folders totaling 602MB. The test is split into the time it takes to compress the files and the time it takes to decompress the files.
File Transfer: individual trace files of transferring the Office Space DVD files to our source drive and transferring the files back to our test drive. The source and test drives are the same make and model for these tests. The content being transferred consists of 29 files with a content size of 7.55GB.
AnyDVD 5.9.6: trace file of the time it takes to "rip" the Office Space DVD. We first copy the entire DVD over to our source drives, defragment the drive, and then measure the time it takes for AnyDVD to "rip" the contents to our test drive. While this is not ideal, it does remove the optical drive as a potential bottleneck during the extraction process and allows us to track the write performance of the drive.
Nero Recode 2: trace file of the time it takes to shrink the entire Office Space DVD that was extracted in the AnyDVD process into a single 4.5GB DVD image.
Game Installation: individual trace files of the time it takes to install Sims 2 and Battlefield 2. We copy each DVD to our secondary test drives, defragment the drive, and then install each game to our source drive.
Game Play: individual trace files that capture the startup and about 15 minutes of game play in each game. The Sims 2 trace file consists of the time it takes to select a pre-configured character, setup a university, downtown, business from each expansion pack (pre-loaded), and then visit each section before returning home. Our final trace file utilizes Battlefield 2 and we play the Daqing Oilfield map in both single and multiplayer mode.
| Standard Test Bed Playback of iPeak Trace Files and Test Application Results |
|
| Processor: | AMD Opteron 175 utilized for all tests. |
| RAM: | 2 x 1GB Corsair 3500LL PRO Settings: DDR-400 at (2.5-3-3-7, 1T) |
| OS Hard Drive: | 1 x Western Digital WD1500 - SATA - 16MB Buffer - 10,000RPM |
| System Platform Drivers: | NVIDIA Platform Driver - 6.85 |
| Video Card: | 1 x ASUS 7600GS (PCI Express) for all tests. |
| Video Drivers: | NVIDIA nForce 84.21 WHQL |
| Optical Drive: | BenQ DW1640 |
| Cooling: | Zalman CNPS9500 |
| Power Supply: | OCZ GameXStream 700W |
| Case: | Gigabyte 3D Aurora |
| Operating System: | Windows XP Professional SP2 |
| Motherboard: | MSI K8N Diamond Plus |
Our current test bed reflects a common level system that has been in use for a year now. Based upon the continuing proliferation of dual core processors and future roadmaps from AMD and Intel signifying new chipset and processor designs, we will revise our test bed shortly to reflect these changes in the marketplace. This change will also allow us to expand our real world multitasking benchmarks in the near future while providing a stable platform for the next few months. We are currently conducting preliminary benchmark testing under Vista with both 2GB and 4GB memory configurations. We will offer real-world Vista benchmarks once the driver situation matures but iPeak results will continue to be in XP as the application is not compatible with Vista.
Test Setup - Software
With the variety of disk drive benchmarks available, we needed a means of comparing the true performance of the hard drives in real world applications. While we will continue to utilize HDTach and PCMark05 for comparative benchmarks our logical choice for application benchmarking is the Intel iPeak Storage Performance Toolkit version 3. We originally started using this storage benchmark application in our Q2 2004 Desktop Hard Drive Comparison. The iPeak test can be designed to measure "pure" hard disk performance, and in this case we kept the host adapter consistent while varying the hard drive models. The idea is to measure the performance of individual hard drives with a consistent host adapter.
We utilize the iPeak WinTrace32 program to record precise I/O operations when running real world benchmarks. We then utilize the iPeak AnalyzeTrace program to review the disk trace file for integrity and ensure our trace files have properly captured the activities we required. Intel's RankDisk utility is used to play back the workload of all I/O operations that took place during the recording. RankDisk generates results in a mean service time in milliseconds format; in other words, it gives the average time that each drive took to fulfill each I/O operation. In order to make the data more understandable, we report the scores as an average number of I/O operations per second so that higher scores translate into better performance in all of our iPeak results. While these measurements will provide a score representing "pure" hard drive performance, the actual impact on the real world applications can and will be different.
The drive is formatted before each test run and three tests are completed on each drive in order to ensure consistency in the benchmark results. The high and low scores are removed with the remaining score representing our reported result. We utilize the NVIDIA nF4 SATA ports along with the NVIDIA IDE-SW driver to ensure consistency in our playback results when utilizing NCQ, TCQ, or RAID settings. We will test with AAM and NCQ turned on with our Deskstar 7K1000 unit as AAM does not noticeably impact performance and this drive generally performs better with NCQ on in the majority of tests.
Our iPeak tests represent a fairly extensive cross section of applications and usage patterns for both the general and enthusiast user. We will continually tailor these benchmarks with an eye towards the drive's intended usage and feature set when compared to similar drives. In essence, although we will report results from our test suite for all drives, it is important to realize a drive designed for PVR duty will generate significantly different scores in our gaming benchmarks than a drive designed with gaming in mind such as the WD Raptor. This does not necessarily make the PVR drive a bad choice for those who capture and manipulate video while also gaming. Hopefully our comments in the results sections will offer proper guidance for making a purchasing decision in these situations. Our iPeak Test Suite consists of the following benchmarks.
VeriTest Business Winstone 2004: trace file of the entire test suite that includes applications such as Microsoft Office XP, WinZip 8.1, and Norton Antivirus 2003.
VeriTest Multimedia Content Creation 2004: trace file of the entire test suite that includes applications such as Adobe Photoshop 7.01, Macromedia Director MX 9.0, Microsoft Windows Media Encoder 9.0, Newtek Lightwave 3D 7.5b, and others.
AVG Antivirus 7.1.392: trace file of a complete antivirus scan on our test bed hard drive.
Microsoft Disk Defragmenter: trace file of the complete defragmentation process after the operating system and all applications were installed on our test bed hard drive.
WinRAR 3.51: trace file of creating a single compressed file consisting of 444 files in 10 different folders totaling 602MB. The test is split into the time it takes to compress the files and the time it takes to decompress the files.
File Transfer: individual trace files of transferring the Office Space DVD files to our source drive and transferring the files back to our test drive. The source and test drives are the same make and model for these tests. The content being transferred consists of 29 files with a content size of 7.55GB.
AnyDVD 5.9.6: trace file of the time it takes to "rip" the Office Space DVD. We first copy the entire DVD over to our source drives, defragment the drive, and then measure the time it takes for AnyDVD to "rip" the contents to our test drive. While this is not ideal, it does remove the optical drive as a potential bottleneck during the extraction process and allows us to track the write performance of the drive.
Nero Recode 2: trace file of the time it takes to shrink the entire Office Space DVD that was extracted in the AnyDVD process into a single 4.5GB DVD image.
Game Installation: individual trace files of the time it takes to install Sims 2 and Battlefield 2. We copy each DVD to our secondary test drives, defragment the drive, and then install each game to our source drive.
Game Play: individual trace files that capture the startup and about 15 minutes of game play in each game. The Sims 2 trace file consists of the time it takes to select a pre-configured character, setup a university, downtown, business from each expansion pack (pre-loaded), and then visit each section before returning home. Our final trace file utilizes Battlefield 2 and we play the Daqing Oilfield map in both single and multiplayer mode.
Facebook
Twitter
RSS
48 Comments
View All Comments
userexists - Thursday, April 19, 2007 - link
As I understand it, from previous articles, the limiting factor in gaming tests seems to be the CPU. I understand why you'd want to use an Opteron or Xeon system for benchmarking the access patterns -- the only people who care about those results are probably going to be running servers. But most people playing games aren't using server components. I'd love to see how the QX6800, for example, and some fast RAM affects gaming benchmarks under RAID-0 -- i.e. answer the question of whether the CPU bottleneck has been relieved. Probably not, but who knows until you test it, right?Gary Key - Sunday, April 22, 2007 - link
I will have some Intel benchmarks with a QX6700 up this week although I doubt the results will be that surprising. ;-)cbuchach - Thursday, April 19, 2007 - link
For all the arguing, NO ONE can say that RAID0 is overall slower. In most situations it is faster by varying degrees, maybe a percent or two or maybe more. For enthusiasts, the percent counts. Look at heatsinks or overclocking. Someone may spend an extra $50 for a better heatsink, for what, maybe an increased overlcock from 3300 MHz to 3400 MHz or spend lots of cash for a water cooling setup, or spend an extra $100 for slightly better RAM; the list goes on and on. For real enthusiasts, the extra 1-2% counts.(And as far as data loss goes, everyone should backing up all their nonrecoverable data data anyways, so that point is moot.)
tshen83 - Thursday, April 19, 2007 - link
and don't even talk about performance, people who want performance will buy Raptors X or SCSI drives. this drive is for storage. RAID storage for cheaptshen83 - Thursday, April 19, 2007 - link
yes, backing up 2TB of data, with what? probably a RAID1(takes 4 drives to have 2TB) or RAID5(3 drives) of the same drives. so why not just use RAID1 or RAID5 in the beginning?ncage - Thursday, April 19, 2007 - link
I just wish hitachi would make this drive in something other than 1TB. I love hitachi hard drives and just which this awesome thing would come in like a 500GB or something like that.Ncage
Gary Key - Thursday, April 19, 2007 - link
750GB drives will be available in May, the smaller capacities later this summer.TomWomack - Thursday, April 19, 2007 - link
What I'd be much more interested by is a review of performance for a pair of these drives in RAID1. RAID1 read speed ought to be the same as RAID0, and most disc-limited tasks are read-limited, whilst running drives in RAID1 seems a sensible reaction to the combined unreliability and cheapness of modern HDDs.[also you can break a RAID1 mirrored pair and grovel for deleted data on one of the drives while running the computer happily on the other, which I've found useful in the past]
Watson - Thursday, April 19, 2007 - link
I would love to see if useful speed increases are actually available over multiple drives when splitting OS and cache files from applications vs. Raid 0. I have a Raid 0 on 10k Raptors in my machine, and they are very fast (obviously), but I have often wondered in a reinstall if I would be better off splitting the drives and what is put on them. Any thoughts?yyrkoon - Thursday, April 19, 2007 - link
If you are wondering if booting from this array would be slower, or faster, the chances are with RAID0 if anything, the array will boot slower vs a single Raptor. The reason behind this is simple: booting windows, HDDs benifit more from faster access times, and RAID0 will increase random access times. RAID1 on the other hand, could help some here, but it really depends on the controller(RAID1 paired with the right controller can actually decrease access times, but it will not be a huge difference).Now, all that being said, there is a reason why systems, where speed, and redundancy is crucial, people opt for RAID10. Obviously, there is the redundancy factor, but you can get the from RAID5 as well as speed if enough drives are used. Pretty much, you get the best of both worlds having a RAID10 array, faster access times, and throughput. This performance of course comes at a cost, you need a minimum of 4 HDDs, so for instance, using 4x 1TB Hitachi drives, we are talking in the balpark of $1600 for a bare minimum + controller capable of handling RAID10.
When it is all said and done, unless your system is serving thousands of people data every hour of the day, heavily editing video, or some other similar task you do not need this kind of disk performance. Also, for the life of me, I can not see how making one large disk array, for your OS, and putting all your data on this array is going to help things either. Personally, I think it is much smarter, to use a single fast disk for the OS, and perhaps multiple drives for data, keeping everything seperate. As for using RAID, well, I can see an application for it, even in the home, but not for the OS.
Think about it, what is so important about the OS that you need redundancy for it ? Nothing, plain and simple. Need a RAID array for video editing, or something else ? fine, get a third HDD for the OS, and keep the RAID0 array seperate. Same goes for RAID1, or RAID5, keep it seperate from the OS, and if something catostrophic does happen, chances are, it wont be on the data disks(however, nothing is ever set in stone). I have been using this technique since the mid 90's, and have had very little problems, and have lost next to zero data, that I needed. Not only this, but it does help to organize your data, so it is more easily found later on, but not as important.
As for splitting OS/swap accross multiple drives, this is debatable. First, if all of your SATA channels are saturating what your system is capable of handeling, then no, but in theory it should help. I personally have noticed the bigest differences when transfering file locally, if I am transfering files from a PATA -> SATA drive, or vice versa. Two different interfaces, using two difference I/O channels.
</two cents>