3 x VelociRaptor WD5000HHTZ's: One of These Drives is Not Like the Others.

[wansurfer]

Hello,

I recently acquired three VelociRaptor WD5000HHTZ drives. Each one has firmware 04.06A00. I downloaded and installed a disk benchmarking program, CrystalDiskMark 3.0.2 x64. I'm sure there are other programs that do more things but I doubt I would know what I was looking at if I used them. I ran the program on all three drives and every time I found that one of the drives, always the same drive, for some reason has a lower sequential read and write. Why is this one drive like this and what, if anything, can I do about it?

The first set of benchmarks were completed with the SATA controller in AHCI mode.
MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]:

Test: 1000 MB [F: 0.0% (0.1/465.8 GB)] (x3)
==================================================
Sequential Read: 186.496 MB/s
Sequential Write: 182.695 MB/s
Random Read 512KB: 76.386 MB/s
Random Write 512KB: 127.599 MB/s
Random Read 4KB (QD=1): 1.100 MB/s [268.5 IOPS]
Random Write 4KB (QD=1): 3.488 MB/s [851.5 IOPS]
Random Read 4KB (QD=32): 3.447 MB/s [841.5 IOPS]
Random Write 4KB (QD=32): 3.710 MB/s [905.7 IOPS]

Test: 1000 MB [G: 0.0% (0.1/465.8 GB)] (x3)
==================================================
Sequential Read: 192.806 MB/s
Sequential Write: 188.542 MB/s
Random Read 512KB: 77.621 MB/s
Random Write 512KB: 130.108 MB/s
Random Read 4KB (QD=1): 1.075 MB/s [262.5 IOPS]
Random Write 4KB (QD=1): 3.540 MB/s [864.2 IOPS]
Random Read 4KB (QD=32): 3.432 MB/s [837.9 IOPS]
Random Write 4KB (QD=32): 3.938 MB/s [961.4 IOPS]

Test: 1000 MB [H: 0.0% (0.1/465.8 GB)] (x3)
==================================================
Sequential Read: 193.607 MB/s
Sequential Write: 189.564 MB/s
Random Read 512KB: 76.109 MB/s
Random Write 512KB: 125.328 MB/s
Random Read 4KB (QD=1): 1.071 MB/s [261.4 IOPS]
Random Write 4KB (QD=1): 3.246 MB/s [792.6 IOPS]
Random Read 4KB (QD=32): 3.328 MB/s [812.4 IOPS]
Random Write 4KB (QD=32): 3.446 MB/s [841.3 IOPS]

I wasn't expecting the benchmark program to show me exactly the same results for each drive. Of course one drive would be the slowest of the three. However, I expected each drive's performance to be fairly close to the other and drive F: seems to lagging a little bit more while the other two drives performed about the same.

All three drives together in RAID 0:

Test: 1000 MB [C: 41.5% (335.1/806.4 GB)] (x3)
==================================================
Sequential Read: 529.250 MB/s
Sequential Write: 512.250 MB/s
Random Read 512KB: 85.095 MB/s
Random Write 512KB: 180.961 MB/s
Random Read 4KB (QD=1): 1.169 MB/s [285.4 IOPS]
Random Write 4KB (QD=1): 6.699 MB/s [1635.5 IOPS]
Random Read 4KB (QD=32): 7.862 MB/s [1919.5 IOPS]
Random Write 4KB (QD=32): 11.255 MB/s [2747.9 IOPS]

In case you're wondering about the discrepancy of capacity: the C: drive above was actually one of two partitions on the physical drive and the original drive that was cloned to the array was only 1 TB and not 1.5 TB.

I wasn't sure what to try so I changed the order that the drives plug in to the SATA II ports. I thought maybe the motherboard has some kind of quirk which was preventing one of the drives from performing as well as it could. After I started Windows again I spent some time trying to figure out exactly which drive was assigned F, G, and H. I figured it out and gave the drives the same letters they had earlier.

Another set of benchmarks with the controller in RAID mode but no drives in an array:

Test: 1000 MB [F: 0.1% (0.7/465.8 GB)] (x5)
==================================================
Sequential Read: 187.917 MB/s
Sequential Write: 184.446 MB/s
Random Read 512KB: 77.672 MB/s
Random Write 512KB: 127.100 MB/s
Random Read 4KB (QD=1): 1.098 MB/s [268.0 IOPS]
Random Write 4KB (QD=1): 3.530 MB/s [861.7 IOPS]
Random Read 4KB (QD=32): 3.409 MB/s [832.4 IOPS]
Random Write 4KB (QD=32): 3.733 MB/s [911.5 IOPS]

Test: 1000 MB [G: 0.1% (0.7/465.8 GB)] (x5)
==================================================
Sequential Read: 192.788 MB/s
Sequential Write: 189.479 MB/s
Random Read 512KB: 76.210 MB/s
Random Write 512KB: 131.977 MB/s
Random Read 4KB (QD=1): 1.087 MB/s [265.4 IOPS]
Random Write 4KB (QD=1): 3.713 MB/s [906.4 IOPS]
Random Read 4KB (QD=32): 3.337 MB/s [814.8 IOPS]
Random Write 4KB (QD=32): 4.023 MB/s [982.2 IOPS]

Test: 1000 MB [H: 0.1% (0.7/465.8 GB)] (x5)
==================================================
Sequential Read: 194.722 MB/s
Sequential Write: 191.224 MB/s
Random Read 512KB: 79.073 MB/s
Random Write 512KB: 127.954 MB/s
Random Read 4KB (QD=1): 1.089 MB/s [265.8 IOPS]
Random Write 4KB (QD=1): 3.322 MB/s [811.2 IOPS]
Random Read 4KB (QD=32): 3.430 MB/s [837.5 IOPS]
Random Write 4KB (QD=32): 3.453 MB/s [842.9 IOPS]

Again drive F: has a slightly lower rate of sequential read and write. I have an ASUS P8Z68-V Pro (not Gen 3) motherboard with an Intel Z68 Express chipset. I am using the integrated controllers: Intel’s (4 x SATA II ports, 2 x SATA III ports) and Marvel’s (2 x SATA III ports). I have a 1 TB WD Black drive connected to an Intel Z68 SATA III port and the three VelociRaptor drives are connected to the Intel Z68 SATA II ports. The DVD drive is connected to a Marvel port.

I also tried doing a benchmark on the slower drive while only that one drive was connected to a SATA II ports. All other drives were disconnected from the SATA II ports. The results weren’t very different.

Test: 1000 MB [F: 0.1% (0.7/465.8 GB)] (x5)
==================================================
Sequential Read: 187.681 MB/s
Sequential Write: 184.657 MB/s
Random Read 512KB: 77.736 MB/s
Random Write 512KB: 127.737 MB/s
Random Read 4KB (QD=1): 1.095 MB/s [267.2 IOPS]
Random Write 4KB (QD=1): 3.484 MB/s [850.5 IOPS]
Random Read 4KB (QD=32): 3.439 MB/s [839.6 IOPS]
Random Write 4KB (QD=32): 3.765 MB/s [919.1 IOPS]

Any ideas or suggestions?

 

[BFG10K]

That’s about a 3% difference, which is within most noise margins.

It could also be simple product variance, or that the slower drive has a different firmware version to the other drives, which can impact performance.

 

[groberts101]

that man is right and is exactly why I buy all my HDD's in bulk lots and return the slowest of the bunch. Took me just over 30 Barracuda 500GB drives to find 8 of the best performers for my main storage array.. and 11 of the same in 1TB drives to get 4 good ones for the other array.

The wider the array.. the greater the speed loss potential and cherry picking is the ONLY way that I've found to maximize raids speed.. since just a few weaker drives can pull the whole works down by sizable margins.

 

[wansurfer]

Thanks for you replies.

I expect that WD will tell me something like what BFG10K stated. I sent their customer support a message about this, too. However, I posted about this here because sometimes companies aren't as helpful and quick at figuring out what's going on.

Yes, this one drive was probably like this from the start. A while ago I read about how people sometimes cherry pick the best drives and resell their unwanted drives. If I'm looking for a good deal for a drive I may be able to buy one of these drives for less but it may not be one of the better performing drives. I also read that some people were just RMA'ing their slower drives to WD but eventually WD quit letting people do that. I think WD simply increased their tolerance for variability of performance. If WD tests an RMA'ed drive and they find that it performs within tolerances then I expect they would bill the owner of the drive for the testing.

I was thinking about trying to cherry pick a couple more good VR drives but it sounds like that may be more trouble than it is worth for just a 3 - 4% difference in performance. Though, I've already had good luck for 2 out of 3. Hehe, or what if I get a drive that has a sequential read rate of something like 197 MB/s? Then I'll want to re-cherry pick the other drives. Maybe I should try getting a fourth drive. If its another slightly slower one maybe I'll just keep the two slower drives. If the fourth drive is a faster one maybe I'll sell off the slowpoke. At least the slowpoke drive's random read and write rates look the same as the other two drives. Calling it a slowpoke is kind of an exaggeration. It is still a lot quicker than my 1TB WD Black drive

All three drives were manufactured in 2012 and I think they have the same firmware.

 

[BrightCandle]

I had a similar experience with a replacement for a VelicoRaptor drive after one in a raid array failed. The new drive performed slightly better than the old one in some cases and slower in others and when combined in Raid did worse than the previous setup. I let WD know and they actually replaced both drives (the new one and the existing still working one) under their raid swap program and gave me two picked next to each other in the production line so their SN differed by just 1. They do perform marginally better than the previously slightly mismatched drives.

However I suspect if one fails the other wouldn't be far behind it as its likely any manufacturing fault that impacted one is likely to be quite similar on the other.

 

[wansurfer]

BrightCandle: "I let WD know and they actually replaced both drives ... under their raid swap program and gave me two picked next to each other in the production line ..."

That's interesting. I didn't know WD would do that. I think the slowpoke that I purchased wasn't a left over from cherry picking. It was one of two used drives from one person and they were used for about two weeks for a project. The third drive I bought was new and in a sealed antistatic bag with a little packet of absorbent material. I may keep the slower drive or see if I WD will exchange it for a speedier one. Though I'm guessing that WD won't benchmark the replacement(s) before sending it to me. They probably only get a set of drives that should be nearly identical so that the drives will work better together in a RAID configuration. Looks like the prices for the WD5000HHTZ are creeping back up to normal so now paying a lower price for a good drive will be more difficult. The slower VelociRapter that I bought cost $95. The best price I can find now is about $125. I haven't checked if they are used or not. Briefly in November Newegg was selling them for about $80.

 

[Insomniator]

Cherry picking mechanical hard drives.... gotta be kidding me.

 

[BrightCandle]

Yes I think cherry picking them is a bit much as well. Getting matching firmware on the other hand can make a surprising difference. I don't think the OP has too much of a problem looking at the numbers, not in the same way I did with my mismatch.

 

[wansurfer]

You can zoom-in on a lot of pictures in advertisements from sellers to read what's printed on the label of a drive. I've seen firmware versions printed on drive labels but unfortunately it looks like WD doesn't print the firmware version on the labels of WD5000HHTZ. If the pictures are important to you then, of course, be sure the picture is exactly what is being sold. The manufacture date is on the labels of WD5000HHTZ. I checked my drives' firmware versions again using WD's Data Lifeguard Diagnostics program. The version is the same for all three. A while ago at first when I learned how widely the performance of mechanical drives can vary I thought somebody was kidding me.