[JLN]

Folks,

Some people have wanted to know the performance of the WD1500ADFD (enterprise version as tested by Storage Review) in a RAID 0 configuration, and understandably so. I was curious, too. So I bought two of them and hooked them up. After reading Eugene's excellent and thorough review, I was darn sure going to buy at least one to give me higher performance running professional flight-simulation and media-editing software. I built the array using my motherboard's built-in nvraid controller.

The platform is built around an Asus A8N-SLI Rev. 1.02 motherboard (the vanilla version) with an AMD 64 3800+ Venice core processor, overclocked to 215MHz X 12 (2.58 GHz over 2.4 standard, and very stable.)
2 gigs of Corsair XMS Twinx-3200C2PT DDR400 RAM @ CAS 2.5, 3-2-3-8, 1T
EVGA 7800GTX 512 card in the PCIe-X16 slot, no SLI, no tweaking
Two WD1500ADFDs in RAID 0 with 128KB data-block size and NCQ (formatted in NTFS to 279GB).
Single WD2500JS 250GB SATA II as backup drive
Enermax Noisetaker 600-watt power supply
Windows XP Home SP2 (yeah, I know ... but I'm waiting for Vista!)
All drivers (chipset, video, nvraid) are latest non-beta versions from nvidia.

I tested several times using Sisoft Sandra Lite 2005 SR.3 free version and got remarkably similar results each time. SiSoft Sandra uses a 2GB file for benchmarking. Block size is 1MB and I/O que depth is 4 requests. While this is nowhere near as good for benchmarking as Storage Review's tools, and the results are not comparable as they don't include I/O benchmark sets, it should give some idea what this drive is capable of. All results include NCQ. I don't yet know how to disable it. And I honestly don't know if TLER is enabled. I'm assuming not, based on Storage Review's article.

Here we go:
Overall score: 128 MB/Second
Buffered read: 260 MB/S
Sequential read: 170 MB/S
Random read: 59 MB/S
Buffered write: 176 MB/S
Sequential write: 168 MB/S
Random write: 89 MB/S
Average access time: 11ms

Other observations: The only SATA configuration faster for overall score in SiSoft's database is four 36GB Raptors in RAID 0.

These drives are, as reported, extremely quiet. I have a high-quality all-aluminum case, but it doesn't have any special noise-reduction features for the drives, such as rubber mount grommets, etc. The drives are mounted in the front drive cage, where they're cooled by a 120mm fan an inch away, using good ol' metal screws. I can barely hear them even when they're being benchmarked. In fact, that's about the only time I can hear them at all, and my case is about 2.5 feet away, next to my computer desk. They have to really crunch to be heard over the video card and processor cooling systems, and my computer overall is nearly silent.

They have been running constantly for two days without a hiccup.

IMPORTANT NOTE: Anyone who uses the nvraid controller might run into the same problem I did, namely that the nvraid service can conflict with Windows XP. I simply disabled nvraid service using msconfig. (Plenty of tools are still available, such as nvidia Raid Manager, etc.) I observed no deterioration in performance.

Honestly, doing what I've done is overkill for most people. Just one of these drives coupled with a large backup drive would be more than sufficient. But if you're using applications that load huge files frequently, they may just be worth your money. I found these readily available at my favorite online store for $289 each, OEM.

And let me add this: I would probably have been quite happy buying two additional WD2500JS SATA IIs for less than $200 total and configuring them in RAID 0 with the third as a backup. The performance would certainly have been good enough, and for $400 less! (For my money, the new JS is a fantastic drive. Highly recommended.) Still, I like to have the best performance possible within a certain price curve (I'm not shelling out $790 for an dual-core 4800+, much as I'd like one), so I'm deliriously happy with the two Raptors, and I don't find the price objectionable given the performance gains in massive applications.

Hope this post satisfies some curiosity. The bottom line is that the WD1500ADFD appears to be an excellent, surprisingly quiet and stable drive. If I have problems, I'll report them here.

Happy hunting!

John

[Shailesh]

Thanks JLN for the posting. We need some comparisons with different controllers, like the popular NVRAID, Sil3124, and Sil3112. I've read reviews of other drives where the controller made a big difference. While you're at it, why don't you post the Sandra benchmark of the single WD2500JS 250GB SATA II backup drive too?

[JLN]

You're quite welcome.

I'll be happy to benchmark and post the scores for the single WD2500. I can tell you, though, at $100, it's a steal.

John

Shailesh, on Jan 16 2006, 03:01 PM, said:

Thanks JLN for the posting. We need some comparisons with different controllers, like the popular NVRAID, Sil3124, and Sil3112. I've read reviews of other drives where the controller made a big difference. While you're at it, why don't you post the Sandra benchmark of the single WD2500JS 250GB SATA II backup drive too?

[TechNet]

I have 3 x Seagate 7200.9 160Gb drives in a RAID0 hanging off a standard Intel ICH7R controller in a Shuttle SD31P. Drives are full SATAII with 3.0Gb/s interfaces, 8Mb cache and NCQ. These are dirt cheap £60 drives which have 1 x 160Gb platter in each unit which I chose due to their low temperature characteristics, which suit a Shuttle PC.

Using the same SiSandra version as you have my benchmarks are in a different league I'm afraid which either indicates that SiSandra is rubbish for becnhmarks or platter size makes a difference.

SiSoftware Sandra

Benchmark Results
Drive Index : 159 MB/s
Results Interpretation : Higher index values are better.

Performance Test Status
Run ID : SHUTTLE on 16 January 2006 at 19:57:29
SMP Test : No
Total Test Threads : 1
SMT Test : No
Dynamic MP/MT Load Balance : No
Processor Affinity : No
Operating System Disk Cache Used : No
Use Overlapped I/O : Yes
IO Queue Depth : 4 request(s)
Test File Size : 2GB
File Fragments : 1
Block Size : 1MB
File Server Optimised : No

Benchmark Breakdown
Buffered Read : 428 MB/s
Sequential Read : 210 MB/s
Random Read : 88 MB/s
Buffered Write : 306 MB/s
Sequential Write : 196 MB/s
Random Write : 90 MB/s
Average Access Time : 7 ms (estimated)

[JLN]

Holy cow! I'm not sure what to make of that. We could have different options in the modules, or perhaps you just have a much faster system overall. Or MAYBE disabling the nvraid service made a huge difference, benchmark-wise, that I can't detect in real-world use.

It also could be that your RAID controller is faster/better than nvraid. I do believe that platter size makes a difference, btw.

At any rate, those are some impressive numbers.

TechNet, on Jan 16 2006, 03:06 PM, said:

I have 3 x Seagate 7200.9 160Gb drives in a RAID0 hanging off a standard Intel ICH7R controller in a Shuttle SD31P. Drives are full SATAII with 3.0Gb/s interfaces, 8Mb cache and NCQ. These are dirt cheap £60 drives which have 1 x 160Gb platter in each unit which I chose due to their low temperature characteristics, which suit a Shuttle PC.

Using the same SiSandra version as you have my benchmarks are in a different league I'm afraid which either indicates that SiSandra is rubbish for becnhmarks or platter size makes a difference.

SiSoftware Sandra

Benchmark Results
Drive Index : 159 MB/s
Results Interpretation : Higher index values are better.

Performance Test Status
Run ID : SHUTTLE on 16 January 2006 at 19:57:29
SMP Test : No
Total Test Threads : 1
SMT Test : No
Dynamic MP/MT Load Balance : No
Processor Affinity : No
Operating System Disk Cache Used : No
Use Overlapped I/O : Yes
IO Queue Depth : 4 request(s)
Test File Size : 2GB
File Fragments : 1
Block Size : 1MB
File Server Optimised : No

Benchmark Breakdown
Buffered Read : 428 MB/s
Sequential Read : 210 MB/s
Random Read : 88 MB/s
Buffered Write : 306 MB/s
Sequential Write : 196 MB/s
Random Write : 90 MB/s
Average Access Time : 7 ms (estimated)

[Eugene]

shoek, on Jan 16 2006, 01:19 PM, said:

Eugene,
How do you respond to GamePC.com's recent Raptor 150 review, where they show that a dual, triple, or quad RAID0 array of WD740GD's beats a single Raptor. They used the highly respected Areca 1220 PCIexpress card. I assume you used a Silicon Image SATA card? How do you think the nForce4 RAID would fare?

You've linked to IOMeter (a multi-user benchmark) results found at GamePC and imply that they somehow contradict the SR DriveMarks (a single-user benchmark).

I don't know how many times I have to say it, but IOMeter results are not representative of non-server performance. I highly suggest rereading the sidebar found at the bottom of this page.

[TechNet]

JLN

I reckon the extra drive in my RAID0 array makes a big difference also, which is apparent in Eugene's benchmarks when he tested the WD1500.

Can't work out why WDC restricted the controller interface to just 1.5Gb/s instead of 3.0Gb/s as it does make a difference when the drives are RAID'd

This post has been edited by TechNet: 16 January 2006 - 03:23 PM

[shoek]

Eugene, on Jan 16 2006, 03:16 PM, said:

You've linked to IOMeter (a multi-user benchmark) results found at GamePC and imply that they somehow contradict the SR DriveMarks (a single-user benchmark).

I don't know how many times I have to say it, but IOMeter results are not representative of non-server performance. I highly suggest rereading the sidebar found at the bottom of this page.

If IOMeter's workstation pattern doesn't float your boat, pick any of the other benchmarks in the GamePC review, such as DiskBench or HDTach.

Regardless of the benchmark, their conclusion is the same... 2+ Raptor 74's beat a single Raptor 150. The original question still stands... why do you think these benchmarks seem to contradict those of SR? Attribute it to the high-end controller? Blame it on the benchmarks or methodology not being as sound as those that SR uses?

Thanks in advance for answering my question.
shoek

PS: a tangent.. why do they call it the "workstation" pattern in IOMeter if it should only be used to measure "server" performance? A machine that heavily multitasks on, say, software development for the FEA application market, is not representative of the "workstation" that IOMeter attempts to model?

[Eugene]

shoek, on Jan 16 2006, 05:06 PM, said:

If IOMeter's workstation pattern doesn't float your boat, pick any of the other benchmarks in the GamePC review, such as DiskBench or HDTach.

These benchmarks are lower-level. As warned by the sidebar at the end of this page, one should not assume that low-level diagnostics equal application benchmarks.

Quote

PS: a tangent.. why do they call it the "workstation" pattern in IOMeter if it should only be used to measure "server" performance? A machine that heavily multitasks on, say, software development for the FEA application market, is not representative of the "workstation" that IOMeter attempts to model?

Have you any idea who "they" really is when it comes to the IOMeter "Workstation" pattern? As outlined here, I invented the IOMeter workstation pattern back in March of 2000. It was not and is not a formal pattern defined either by Intel or the current Sourceforge IOMeter project maintainers. It was, unfortunately, a flawed and ill-fated attempt to try and shoehorn IOMeter into measuring something that by definition it can not. I formally recanted and admitted this error in November of 2001. Unfortunately, many sites still use this flawed pattern despite my retraction and explanation.

Quote

Thanks in advance for answering my question.

So, once again:

1) The creator of the IOMeter Workstation pattern has publicly admitted that it does not measure single-user performance.

2) DiskBench and HDTach measure low-level performance, not application-level performance.

Does this answer your question? The sidebars on page three and page six were written to address readers with concerns such as yours.

[Spod]

TechNet, on Jan 16 2006, 08:22 PM, said:

I reckon the extra drive in my RAID0 array makes a big difference also, which is apparent in Eugene's benchmarks when he tested the WD1500.

Can't work out why WDC restricted the controller interface to just 1.5Gb/s instead of 3.0Gb/s as it does make a difference when the drives are RAID'd

1. For the things that RAID 0 helps with, i.e. STR heavy tasks, then yes, 4 drives are better than two (though often limited by the bandwidth to the SATA controller. As seen above, this STR advantage can easily be matched or exceeded by using a drive with better firmware or a bigger cache when it comes to real world applications.

Some benchmarks give too much importance to STR in the overall scheme of performance. STR benchmarks are the obvious example, giving no consideration to cache, access time or firmware optimisations. That's why SR benchmarks drives using recorded (and thus repeatable) traces of real applications.

As you can see from Eugene's results, if you increase STR while keeping everything else constant, you (usually) get an improvement. By using a single drive with lower STR, similar access times but with major improvements in firmware and cache, the improvement by increasing STR alone is surpassed. That should give you some perspective on the relative importance of STR - it helps, but not that much.

2. The maximum STR of an individual Raptor 150 GB is under 100 MB/s. Since each SATA drive has its own dedicated link to the controller, and that link has a bandwidth of 150 MB/s, the link to the drive will never be a bottleneck no matter how many drives you have.
(As long as you're not using port multipliers to run several drives off one connection to the controller. But you'd know if you were.)

With SATA, it's the bandwidth to the SATA controller that will limit STR from a RAID 0 array - anything not integrated into the chipset itself will be limited to the bandwidth of the PCI bus, usually about 120-130 MB/s.

You were probably thinking of PATA, where two drives might share the same cable (channel), so if you didn't have enough PATA channels free to give each drive its own cable (channel), then two drives had to share the bandwidth of one channel (up to 133 MB/s for ATA133).


So, basically, I stand with Eugene on this one. But I'm glad you challenged him - it's important to question any authority if you suspect that it is misleading you. I hope we've reinforced your trust in SR's benchmarks and opinions, but even if not, thanks for giving us the opportunity to try.*

* I should point out that I don't represent Eugene or SR at all. That last sentence was a bit ambiguous, and I don't want to give the wrong impression.