TR Forums

Hard drive manufacturers over the years have improved the access time however they are still mechanical devices and as Scottie used to say, "Ye canna change the laws of physics".

A lot of cases come with convenient "screwless" mounting arrangements for hard drives and some even come with noise dampening cushioning material. So what's not to like?

When the heads of a hard drive go to data on that hard drive they do so very quickly and then come to a very abrupt stop when they think they are over the data to be either read or written. They will never be directly above the track to be read/written however. They will "wobble" a bit and it is only when the drive is certain that the heads are where they should be that the actual read/write operation begins.

So you have installed the new hard drive into your screwless 3.5" inch bay, hooked it up, formatted it and now you want to test it. You run a benchmark test and you find that your scores are lower than the ones you see in reviews of the drive online. OK, so you got a crappy drive, or the benchmarks in the review are fake - right?

That is not necessarily so. For the hard drive to work to its full potential it has to be mounted as rigidly as possible to the chassis. If this is not the case then, when the heads come to an abrupt full stop, not only will they wobble but so will the hard drive itself. Now on a brand new, empty, hard drive this may not be so noticeable; but as the hard drive gets filled and things get deleted and re-written the heads will move a lot more (until you defrag for instance) and the performance drop will be more noticeable.

Every time the heads move, if the hard drive is not rigidly connected to the chassis the vibration of the drive itself will throw the heads off and cause them to take longer to be precisely where they should be.

So much for theory what does that look like in practice?

Here are the results of my test on my Seagate ST4000DM000 HDD.15 5900 RPM 4 TB drive. I have it mounted in the hotswap bay of my CoolerMaster HAF XM case. I tested it just fitted into the caddy with the four padded mounts and the result on the program DiskSpeed was:

Access Time: 14.96 ms
Overall score: 1080.5

I then took the drive out and just added two screws (one on each side) to affix the drive more rigidly to the caddy and the result was:

Access Time: 14.55 ms
Overall score: 1112.7

I ran the test through three times in each configuration and I have noted the best result I got out of three for each.

As you can see, just by making sure the drive itself cannot vibrate so much it improved the "performance" of my hard drive pretty significantly. I think that's not bad for the inconvenience of just affixing the drive to the caddy more rigidly with the addition of two screws.

You have to remember that my drive only rotates at 5900 RPM and I would expect the differences in performance to be larger on a 7200 RPM drive.

Nec_V20 wrote:

I ran the test through three times in each configuration and I have noted the best result I got out of three for each.

Why did you pick the best of three in each case? I would be a lot more interested in seeing the average of each set of three. That's probably a better indication of true performance since it filters out more of the normal variation you'll see from run to run due to measurement slop. I'm not questioning your original hypothesis (it makes sense to me); I just think your measurement methodology leaves something to be desired.

In systems with more than one drive you've got another variable: vibrations transmitted between the drives. While this will be smaller than the vibrations from a drive's own head actuator, it could still be significant since no case is perfectly rigid. As you add more drives, there may even be a point at which the "squishy" mounts become a net win because they help isolate a drive from the vibrations of its neighbors.

Edit: If you want convenience *and* a mechanically rigid drive mount, maybe what you want is trayless/caddyless hot swap bays. These typically have a door mechanism that holds the drive pretty firmly. since they need to ensure that it stays fully seated in the SATA connector at the back of the drive bay.

just brew it! wrote:

Nec_V20 wrote:

I ran the test through three times in each configuration and I have noted the best result I got out of three for each.

Why did you pick the best of three in each case? I would be a lot more interested in seeing the average of each set of three. That's probably a better indication of true performance since it filters out more of the normal variation you'll see from run to run due to measurement slop. I'm not questioning your original hypothesis (it makes sense to me); I just think your measurement methodology leaves something to be desired.

In systems with more than one drive you've got another variable: vibrations transmitted between the drives. While this will be smaller than the vibrations from a drive's own head actuator, it could still be significant since no case is perfectly rigid. As you add more drives, there may even be a point at which the "squishy" mounts become a net win because they help isolate a drive from the vibrations of its neighbors.

Why not ten runs each? Boredom would be the answer to that.

That was plain and simply because if I had taken the average then the results would be even further apart. The tests done when the drive was in the caddy without the screws were a lot further apart than the ones I did when the drive was affixed with the two screws, so the result would have been crasser. The two other tests I did with the screws out had the access time at over 15.0 ms (15.05 and 15.02) and the tests I did with the screws in were all in the same (14.58 and 14.57) ballpark. That would be 15.01 vs 14.56. As it was, just using the best results of each the difference is 0.41ms. If I compared the averages the difference would be 0.45ms.

0.41ms or 0.45ms apart puts the results in two entirely different worlds. I would credit your criticism if the results were in the region of 0.1ms apart but not at over 0.4ms.

If my methodology were to be rigid then I would have to run at least ten tests each to average out (one bad outlier at three tests would skew the result a lot more than one outlier in ten tests). So with the limited runs of three each I thought it would be fairer to use the best value I got from each scenario.

As you can see however under the two different conditions, with the same drive, same casing, same caddy and the only difference being two screws the results were not even close.

I doubt very much that any hypothetical harmonics transferred to the casing from another drive would change the results in any significant way.

Given the negligible performance gains, plus the fact that pretty much everyone would rather be using solid-state drives for their OS these days, I see no reason to ever make your entire case resonate as a giant sounding box by rigidly attaching it to a disk.

Decoupling/damping the drives has been the more desirable feature for over a decade and when you think about rigidly fixing two drives to a cage, the vibrations of one drive will actually throw off the heads on the second drive more if they are both rigidly mounted. If you were interested in testing this, I'd be curious to see how a pair of drives softmounted/hardmounted perform when benchmarked simultaneously.

Chrispy_ wrote:

Given the negligible performance gains, plus the fact that pretty much everyone would rather be using solid-state drives for their OS these days, I see no reason to ever make your entire case resonate as a giant sounding box by rigidly attaching it to a disk.

Decoupling/damping the drives has been the more desirable feature for over a decade and when you think about rigidly fixing two drives to a cage, the vibrations of one drive will actually throw off the heads on the second drive more if they are both rigidly mounted. If you were interested in testing this, I'd be curious to see how a pair of drives softmounted/hardmounted perform when benchmarked simultaneously.

Who said I was using the 4 TB drive for my OS?

The only reason why they do that is because people were whining about the hard drives being too noisy and they didn't know "lefty loosey, righty tighty" and nothing else. And if you don't know that then you haven't been in the game all that long.

The effect of one HD on another would be negligible unless you have a REALLY flimsy HD cage and even then the effect would not be of the order of over 0.4ms. You know what, you can believe the other posters old wives tale if you want, it's no skin off my nose.

Just sit back and actually THINK about what you are trying to tell me here!

Also any harmonics caused by other HDs is MINUSCULE compared to the vibration of the HD caused by the heads within it coming to an abrupt stop in an HD which is not rigidly attached to the case. And the movements of those drives are as likely to cancel each other out as they are to augmenting each other.

Now I did the work and gave you and just brew it! the results. If you think that more drives would interfere THAT MUCH then do the experiment and get back to me with the results.

I'd like to see the results of more of these tests, out of morbid curiosity.

Having taken apart several hard drives, the voice coil and arm seem to be pretty well balanced. The only force applied to the drive chassis itself is by the voice coil, and is thus parallel to the spinning moment of the discs. The discs are FAR heavier, and even at 5900RPM are moving INCREDIBLY fast (ever lift a non-mounted hard drive? you can balance it on your finger due to the gyroscopic physics at work). I just don't see the voice coil putting enough energy in to the drive to add enough mechanical noise to be a problem.

Also, newer hard drives that pack extreme densities have secondary actuator arms for fine seeking. If I recall, they're guided by piezoelectric materials.

Definitely would like to see more data. Maybe Damage Labs can plug in a few various drives.

Nec_V20 wrote:

That was plain and simply because if I had taken the average then the results would be even further apart. The tests done when the drive was in the caddy without the screws were a lot further apart than the ones I did when the drive was affixed with the two screws, so the result would have been crasser.

Erm... wait a sec. So you intentionally cherry-picked results that show only a 3% difference (i.e. nowhere near statistically significant, given that you've just indicated that the results are varying all over the place), thereby undermining the very point you're trying to make?

Nec_V20 wrote:

I doubt very much that any hypothetical harmonics transferred to the casing from another drive would change the results in any significant way.

So RAID/enterprise class drives that implement vibration compensation (also known as RAFF) are snake oil?

Western Digital wrote:

Rotary Acceleration Feed Forward (RAFF)
Information Sheet
Rotational Vibration Cancellation Technology in WD Enterprise Hard Drives
...
RV is induced in a hard drive as a result of other drives' spinning and seeking in the same chassis.

(Reference: http://www.wdc.com/wdproducts/library/o ... 001079.pdf)

mac_h8r1 wrote:

Having taken apart several hard drives, the voice coil and arm seem to be pretty well balanced. The only force applied to the drive chassis itself is by the voice coil, and is thus parallel to the spinning moment of the discs. The discs are FAR heavier, and even at 5900RPM are moving INCREDIBLY fast (ever lift a non-mounted hard drive? you can balance it on your finger due to the gyroscopic physics at work). I just don't see the voice coil putting enough energy in to the drive to add enough mechanical noise to be a problem.

Yes, the head arm is very well balanced, but that doesn't prevent it from exerting forces on the drive. It has a non-trivial amount of mass, and accelerates/decelerates extremely rapidly. So long seeks *will* cause significant rotational forces about the pivot of the head arm, which will in turn cause the drive to vibrate if it is not mounted firmly.

Yes, the platters are much more massive (and will help damp out the vibrations); however track pitch on current drives is measured in tens of nanometers, so even small amounts of vibration can have a measurable effect.

The main issue I have with the OP is that the data he presented doesn't support his point; I think the overall point is valid.

As far as compliant drive mounts go, some people will actually choose to accept a small reduction in performance if it makes the drive's seek noise less audible. Others will choose performance over quiet. Knowing the tradeoffs involved (if any) is a good thing!

I have to post http://www.youtube.com/watch?v=tDacjrSCeq4

Man, I don't know where you've sprung up from but a lot of your replies to other people seem to be argumentative, malicious or irrationally-defensive based on your assumptions of what you think other person is implying rather than the hard facts of what they have actually written.

Nec_V20 wrote:

Who said I was using the 4 TB drive for my OS?

Not me. The reason I mentioned it at all is because the OS itself is typically the most seek-intensive thing a mechanical hard drive needs to do these days, and generalising, the trend is towards the OS being on an SSD and mechanical being used for low-IOPS media storage with minimal seeking.

Nec_V20 wrote:

The only reason why they do that is because people were whining about the hard drives being too noisy

Well, yeah. There's a multi-million dollar industry sector dedicated to making everything quieter; typically if you look at independent professional reviews, as well as user reviews you'll typically see "too loud" and "noisy" listed as negative points that reduces the value of the item in question. Loud is bad, mmmkay?

Nec_V20 wrote:

and they didn't know "lefty loosey, righty tighty"

I guarantee you that the tightening the hard drive mounting screws will cause the setup to generate more noise than leaving them loose (which is effectively partial-decoupling). If you don't understand that then you need to read up on the theory of acoustic resonance, I'm not going to re-explain what is easily Googled.

Nec_V20 wrote:

And if you don't know that then you haven't been in the game all that long.

See above, and my opening sentence. There's no need to take this kind of stance and it just makes you look bad for veiling an insult with patronism. If you were unquestionably right (which you definitely are not), being patronising is still considered an abrasive attitude trait that polite society frowns upon; Fortunately I'm not particularly polite (I prefer blunt, brutal honesty) so anything I say is mildly hypocritical and I'm not going to waste effort frowning upon you.

Nec_V20 wrote:

The effect of one HD on another would be negligible unless you have a REALLY flimsy HD cage and even then the effect would not be of the order of over 0.4ms. You know what, you can believe the other posters old wives tale if you want, it's no skin off my nose.

Just sit back and actually THINK about what you are trying to tell me here!

Uh, really? Well for a start, I'm not telling you anything you haven't already suggested yourself - which is that drive arms seeking cause significant vibration and failing to secure this vibration slows down the seeks. I'm suggesting that if a single drive is improved by reducing its own vibrations, then it may also be hindered when introduced to vibrations transmitted through a drive cage by other drives. This is why I say "I'd be curious to see them benchmarked simultaneously" because I don't know for sure and have no empirical data oooh, now I do - notfred has given me some!

Anecdotally, enterprise drives incorporate accelerometers to help counteract the foreign vibrations caused by other drives in the same rack, the fact these are expensive and complex upgrades seems to indicate that these vibrations are more than trivial and it is worth significant investment to counter them.

Nec_V20 wrote:

Now I did the work and gave you and just brew it! the results. If you think that more drives would interfere THAT MUCH then do the experiment and get back to me with the results.

Well, it's your experiment and you publicly published the results which are free for criticism and comment.

• JBI queried your rather unusual method of cherry picking the best results
• I implied that the experiment is only valid for one drive (with reasons) and that further testing could show if the results apply to multiple drives or not.

If you can't take constructive criticism well, then you probably shouldn't publicly publish your work. This is the internet, after all - if you don't know that then you haven't been in the game all that long

just brew it! wrote:

Nec_V20 wrote:

I ran the test through three times in each configuration and I have noted the best result I got out of three for each.

Why did you pick the best of three in each case? I would be a lot more interested in seeing the average of each set of three. That's probably a better indication of true performance since it filters out more of the normal variation you'll see from run to run due to measurement slop. I'm not questioning your original hypothesis (it makes sense to me); I just think your measurement methodology leaves something to be desired.

Exactly. Shoot, picking the fastest times mean you picked the outliers. Getting an average of those numbers is far better. But, its interesting stuff. I would love to see the results where there is some statistical significance. At this point, it looks interesting, but isn't influential.

Erm... wait a sec. So you intentionally cherry-picked results that show only a 3% difference (i.e. nowhere near statistically significant, given that you've just indicated that the results are varying all over the place), thereby undermining the very point you're trying to make?

hypothesis: screws make a difference
null hypothesis: screws make no difference

I picked the two results I thought would be fairest to compare (the lowest value of each run) given the limited amount of runs and did not want to accidentally overstate the difference in my original post.

Test runs without screws:
15.05
14.96
15.02
sigma = 0.0374

Test runs with screws
14.57
14.58
14.55
sigma = 0.005

Results from all test runs
15.05
14.96
15.02
14.57
14.58
14.55
sigma = 0.223

I don't see those results as "varying all over the place". There is a clear and persistent similarity of results within a test run as there are a complete and persistent difference between the runs. The standard deviation (sigma) within the runs is very narrow (0.037 and 0.005) i.e homogeneous whereas the standard deviation of the entire population of measurements is very wide (0.223) i.e. heterogeneous.

What are the chances of any values from a test run without screws overlapping (intersecting) a test run with screws? Even given the limited number of runs under each scenario I would still say that the chances of this are very slim indeed (about 0.1%) - well within the 5% confidence level needed for the differences between the two sets of measurements to be considered statistically significant.

Although I didn't like it (the lecturers were boring - like ketamine on legs boring) statistics was a part of my Psychology degree at Bonn University.

So RAID/enterprise class drives that implement vibration compensation (also known as RAFF) are snake oil?

RAFF was introduced to:

Rotary Acceleration Feed Forward (RAFF) overcomes the
effects of RV [Rotational Vibration] on a hard drive by sensing RV disturbance and
controlling head position to keep the drive heads within the safe
operating region during read and write operations

http://www.wdc.com/EN/LIBRARY/SATA/2579-001079.PDF

What is Rotational Vibration?

RV is induced in a hard drive as a result of other drives'
spinning and seeking in the same chassis. RV can also be
induced by external forces on the rack or chassis containing the
hard drives. Even linear vibrations applied on a chassis may get
converted to RV at the drive level if a chassis is not designed
appropriately. An example of this would be a drive bay structure
that is not rigidly attached to a chassis.

http://www.wdc.com/EN/LIBRARY/SATA/2579-001079.PDF

Now as you can clearly see the concept of RAFF has nothing to do with the access time I alluded to in my original post, but rather to keep the heads over the track whilst reading/writing is taking place. So you are comparing apples and oranges and declaring them equal.

Nec_V20 wrote:

Now as you can clearly see the concept of RAFF has nothing to do with the access time I alluded to in my original post, but rather to keep the heads over the track whilst reading/writing is taking place. So you are comparing apples and oranges and declaring them equal.

Average the times of several test runs for both conditions and we'll talk. If you read the "how did we do this" section in any review posted by Damage, Dissonance, or Cyril you'll see that all test results are averages of multiple runs. Do the work, work to the standards of TR reviewers, and no one will give you crap for your testing methods.

Chrispy_ wrote:

Man, I don't know where you've sprung up from but a lot of your replies to other people seem to be argumentative, malicious or irrationally-defensive based on your assumptions of what you think other person is implying rather than the hard facts of what they have actually written.

Nec_V20 wrote:

Who said I was using the 4 TB drive for my OS?

Not me. The reason I mentioned it at all is because the OS itself is typically the most seek-intensive thing a mechanical hard drive needs to do these days, and generalising, the trend is towards the OS being on an SSD and mechanical being used for low-IOPS media storage with minimal seeking.

Nec_V20 wrote:

The only reason why they do that is because people were whining about the hard drives being too noisy

Well, yeah. There's a multi-million dollar industry sector dedicated to making everything quieter; typically if you look at independent professional reviews, as well as user reviews you'll typically see "too loud" and "noisy" listed as negative points that reduces the value of the item in question. Loud is bad, mmmkay?

Nec_V20 wrote:

and they didn't know "lefty loosey, righty tighty"

I guarantee you that the tightening the hard drive mounting screws will cause the setup to generate more noise than leaving them loose (which is effectively partial-decoupling). If you don't understand that then you need to read up on the theory of acoustic resonance, I'm not going to re-explain what is easily Googled.

Nec_V20 wrote:

And if you don't know that then you haven't been in the game all that long.

See above, and my opening sentence. There's no need to take this kind of stance and it just makes you look bad for veiling an insult with patronism. If you were unquestionably right (which you definitely are not), being patronising is still considered an abrasive attitude trait that polite society frowns upon; Fortunately I'm not particularly polite (I prefer blunt, brutal honesty) so anything I say is mildly hypocritical and I'm not going to waste effort frowning upon you.

Nec_V20 wrote:

The effect of one HD on another would be negligible unless you have a REALLY flimsy HD cage and even then the effect would not be of the order of over 0.4ms. You know what, you can believe the other posters old wives tale if you want, it's no skin off my nose.

Just sit back and actually THINK about what you are trying to tell me here!

Uh, really? Well for a start, I'm not telling you anything you haven't already suggested yourself - which is that drive arms seeking cause significant vibration and failing to secure this vibration slows down the seeks. I'm suggesting that if a single drive is improved by reducing its own vibrations, then it may also be hindered when introduced to vibrations transmitted through a drive cage by other drives. This is why I say "I'd be curious to see them benchmarked simultaneously" because I don't know for sure and have no empirical data oooh, now I do - notfred has given me some!

Anecdotally, enterprise drives incorporate accelerometers to help counteract the foreign vibrations caused by other drives in the same rack, the fact these are expensive and complex upgrades seems to indicate that these vibrations are more than trivial and it is worth significant investment to counter them.

Nec_V20 wrote:

Now I did the work and gave you and just brew it! the results. If you think that more drives would interfere THAT MUCH then do the experiment and get back to me with the results.

Well, it's your experiment and you publicly published the results which are free for criticism and comment.

• JBI queried your rather unusual method of cherry picking the best results
• I implied that the experiment is only valid for one drive (with reasons) and that further testing could show if the results apply to multiple drives or not.

If you can't take constructive criticism well, then you probably shouldn't publicly publish your work. This is the internet, after all - if you don't know that then you haven't been in the game all that long

Accepting data and learning from it are part and parcel of being a computer techie.

By inserting the screws and tightening up the drive's contact with the chassis the heads coming to an abrupt stop above the track to be written/read will be absorbed by the mass of the chassis instead of just the drive.

JBI's "criticism" was both statistically and factually unfounded and I replied to both of those above. You jumped in assuming that JBI had his facts straight and proceeded to use that as the basis for criticising my original post.

It was your blind premise "JBI must be right therefore Nec_V20 must be wrong" attitude which irked me.

This is not by any stretch of the imagination "constructive criticism", but rather, given the amount of time and posts you have both have shared here, a case of "incestuous amplification".

Running the test again but this time with a second drive in the system I got the following results:

Disk 1 = Seagate ST4000DM000 5900 RPM 4 TB (1 TB platters)
Disk 2 = Seagate ST3000DM001 7200 RPM 3 TB (1 TB platters)

Both drives no screws attached
Disk 1 running Disk 2 idle
15.44
15.40
15.12
Disk 2 running disk 1 idle
14.88
14.88
14.87
Both drives running
Disk 1
15.34
15.27
15.39
Disk 2
14.99
14.74
14.81

Both drives screws attached
Disk 1 running Disk 2 idle
14.63
14.57
14.76
Disk 2 running Disk 1 idle
14.33
14.56
14.40
Both drives running
Disk 1
14.55
14.41
14.69
Disk 2
14.54
14.57
14.50

Even with two drives the difference between the drives just being in the caddy and the drives being screwed in is eminently noticeable.

Nec_V20 wrote:

JBI's "criticism" was both statistically and factually unfounded and I replied to both of those above. You jumped in assuming that JBI had his facts straight and proceeded to use that as the basis for criticising my original post.

It was your blind premise "JBI must be right therefore Nec_V20 must be wrong" attitude which irked me.

This is not by any stretch of the imagination "constructive criticism", but rather, given the amount of time and posts you have both have shared here, a case of "incestuous amplification".

Do you really want to double down on being wrong? Follow the TR testing procedures stated in every review and get back to us.

Captain Ned wrote:

Nec_V20 wrote:

Now as you can clearly see the concept of RAFF has nothing to do with the access time I alluded to in my original post, but rather to keep the heads over the track whilst reading/writing is taking place. So you are comparing apples and oranges and declaring them equal.

Average the times of several test runs for both conditions and we'll talk. If you read the "how did we do this" section in any review posted by Damage, Dissonance, or Cyril you'll see that all test results are averages of multiple runs. Do the work, work to the standards of TR reviewers, and no one will give you crap for your testing methods.

I did the experiment for myself because I have built up a new NAS in a CoolerMaster HAF XM. Now the fact that access times for hard drives rises if the drives are not rigidly attached has been known for decades and as a matter of course I screw all my hard drives snugly to the chassis. So it is not something I am putting out there off my own bat.

I was however interested to see if the new caddy system of the HAF XM would mitigate that. So I took the screws out and tested it.

As it turns out, even the new caddy system doesn't preclude the necessity for securing the drives additionally with screws.

I was not testing some new concept but rather seeing if, in the light of changes with regard to the convenience of adding drives without screws, it has cured the known issue and the answer is no.

Captain Ned wrote:

Nec_V20 wrote:

JBI's "criticism" was both statistically and factually unfounded and I replied to both of those above. You jumped in assuming that JBI had his facts straight and proceeded to use that as the basis for criticising my original post.

It was your blind premise "JBI must be right therefore Nec_V20 must be wrong" attitude which irked me.

This is not by any stretch of the imagination "constructive criticism", but rather, given the amount of time and posts you have both have shared here, a case of "incestuous amplification".

Do you really want to double down on being wrong? Follow the TR testing procedures stated in every review and get back to us.

Excuse me, the data I collected supports my original statement. If you think I am wrong then show your own results.

What started off as something which I considered trivial and did to assuage my own curiosity and then documented, is being trolled.

Captain Ned wrote:

Nec_V20 wrote:

JBI's "criticism" was both statistically and factually unfounded and I replied to both of those above. You jumped in assuming that JBI had his facts straight and proceeded to use that as the basis for criticising my original post.

It was your blind premise "JBI must be right therefore Nec_V20 must be wrong" attitude which irked me.

This is not by any stretch of the imagination "constructive criticism", but rather, given the amount of time and posts you have both have shared here, a case of "incestuous amplification".

Do you really want to double down on being wrong? Follow the TR testing procedures stated in every review and get back to us.

With the numbers I put up, are you really telling me that you are too stupid to do arithmetic?

You guys might want to do the whole pecking order thing and you may be big fish in this small pond. However although a small fish in this pond I do swim in the ocean and have not yet come adrift.

You might - with regard to my lack of posts here - do well to consider an iceberg.

This thread upset me enough to pull me out of the lurker shadows.

I see only one guy posting test results about a hardware mechanic that seems miss the attention of the PC building masses. The conclusion of these test results would steer readers in a direction counter of the direction of mainstream products. These results aren't 100.00% conclusive, but still strongly suggests something meaningful.

What I don't see is a bunch of other people posting their own (hopefully more statistically conclusive) results using different products, test methods, and environmental conditions. Followed by more replies making conclusions about the results. I don't even see another poster addressing the topic directly in any way.

Instead, something much less productive happened here. While Nec_V20 contributed to this garbage pile, he also provided the only posts on this thread of the slightest worth reading: the ones with numbers. And I appreciate it.

Without accusing him of anything, please note that the bar for 'statistics' here is set rather high.

If someone wants to do some independent testing and be taken seriously, they'll need to submit to intense scrutiny. All of the other extraneous BS that Nec_V20 has brought to this thread and this forum has done enough to discredit his results, and I think that forum members that have asked him to follow standard TR testing procedures are being completely fair. It's not personal, even though he's gone out of his way to deserve it.

Airmantharp wrote:

Without accusing him of anything, please note that the bar for 'statistics' here is set rather high.

If someone wants to do some independent testing and be taken seriously, they'll need to submit to intense scrutiny. All of the other extraneous BS that Nec_V20 has brought to this thread and this forum has done enough to discredit his results, and I think that forum members that have asked him to follow standard TR testing procedures are being completely fair. It's not personal, even though he's gone out of his way to deserve it.

I spoke originally of "results" not "statistics" in my original post.

And personally I know techies when I meet them and so far what has confronted me here on this thread is wannabe's with a ton of probably less rather than more valuable posts on the MB.

So don't try to pile on, you just make yourself look bad by association.

Nec_V20 wrote:

I spoke originally of "results" not "statistics" in my original post.

And personally I know techies when I meet them and so far what has confronted me here on this thread is wannabe's with a ton of probably less rather than more valuable posts on the MB.

So don't try to pile on, you just make yourself look bad by association.

Wow, lots of attitude. Sad, really. Way to make an impression.

Nec_V20 wrote:

... Now as you can clearly see the concept of RAFF has nothing to do with the access time I alluded to in my original post, but rather to keep the heads over the track whilst reading/writing is taking place. So you are comparing apples and oranges and declaring them equal.

I was responding to your implication that vibrations transmitted from other drives in the array would have no effect on drive performance. Clearly, keeping the heads over the proper track is an aspect of that, since if the heads stray from the track the operation needs to be interrupted and retried, which will hurt performance.

In case you missed it, I actually agree with your original hypothesis, and (at least at first) merely noted that I would've liked to have seen the complete data set. Things quickly went downhill from there.

You're posting on a forum with a lot of OCD computer geeks on it. Technical posts will get picked apart; sometimes justifiably, sometimes not. I'm generally a pretty patient guy here on the forums, but I have less patience for rude people. If this thread is an indication of how you're going to react every time anyone disagrees with you, perhaps this forum is not a good place for you.

just brew it! wrote:

I have less patience for rude people. If this thread is an indication of how you're going to react every time anyone disagrees with you, perhaps this forum is not a good place for you.

This, basically.

Your initial post was a good, interesting topic with some useful information based on your own testing.

• JBI queried the validity of the cherry-picking and you did then post the full results, rather than the cherry-picked results - also good.
• JBI queried the validity of your single-drive results in a multi-drive system through transferred vibrations, as did I, but rather than doing what you did the first time and providing further results, or providing a counter theory, you just dismissed them with a sentence that could be summarised as "I doubt that matters".

Man, that is an awful attitude in any discussion. At least back that up with a counter-argument or reasoning.
Anyway, then there's this:

Nec_V20 wrote:

You jumped in assuming that JBI had his facts straight and proceeded to use that as the basis for criticising my original post

I very much doubt I even read JBI's post and no - I did not use his post as the basis for criticising yours: I talked about the hardmounting causing increased noise and then proposed that two drives would interfere, but that I would be curious to see the results for a pair of drives since we didn't have any evidence either for or against our arguments at that time.

• You can't just provide incomplete results and then dismiss other people's theories; that's rude and unscientific.
• You can't misquote people, insinuate lies and dismiss criticism; that's just misleading and it clogs up the thread with argumentative replies like this.

There's a lot of sh*t in this thread now and it's primarily caused because your attitude towards the (constructive) comments of others is just plain awful.

If it seems like I have undue negativity towards you in this thread when treated as a single thread, it's because you attitude is simliarly abrasive and rude in several other threads I've seen you post in since you joined. I've just bitten my tongue in those instances. There is nothing to ever be gained from being rude, dismissive or personally attacking people. If you must do that, at least back up your argument with some hard evidence or an accurately quote what you think they said wrong. Accusing someone of saying something they didn't and then going off on a rant about it really is a pretty dumb way to behave, which is why I think there is now a lot of negativity towards you.

Chrispy_ wrote:

If it seems like I have undue negativity towards you in this thread when treated as a single thread, it's because you attitude is simliarly abrasive and rude in several other threads I've seen you post in since you joined. I've just bitten my tongue in those instances. There is nothing to ever be gained from being rude, dismissive or personally attacking people. If you must do that, at least back up your argument with some hard evidence or an accurately quote what you think they said wrong. Accusing someone of saying something they didn't and then going off on a rant about it really is a pretty dumb way to behave, which is why I think there is now a lot of negativity towards you.

What he said.

Nec_V20 wrote:

Erm... wait a sec. So you intentionally cherry-picked results that show only a 3% difference (i.e. nowhere near statistically significant, given that you've just indicated that the results are varying all over the place), thereby undermining the very point you're trying to make?

hypothesis: screws make a difference
null hypothesis: screws make no difference

I picked the two results I thought would be fairest to compare (the lowest value of each run) given the limited amount of runs and did not want to accidentally overstate the difference in my original post.

Test runs without screws:
15.05
14.96
15.02
sigma = 0.0374

Test runs with screws
14.57
14.58
14.55
sigma = 0.005

Results from all test runs
15.05
14.96
15.02
14.57
14.58
14.55
sigma = 0.223

I don't see those results as "varying all over the place". There is a clear and persistent similarity of results within a test run as there are a complete and persistent difference between the runs. The standard deviation (sigma) within the runs is very narrow (0.037 and 0.005) i.e homogeneous whereas the standard deviation of the entire population of measurements is very wide (0.223) i.e. heterogeneous.

What are the chances of any values from a test run without screws overlapping (intersecting) a test run with screws? Even given the limited number of runs under each scenario I would still say that the chances of this are very slim indeed (about 0.1%) - well within the 5% confidence level needed for the differences between the two sets of measurements to be considered statistically significant.

Although I didn't like it (the lecturers were boring - like ketamine on legs boring) statistics was a part of my Psychology degree at Bonn University.

When you actually show the times for each test run like you do here, given that they are fairly consistent, your decision to pick the fastest times is as good as anything. However, without any indication of the variance within each test group (and more to the point, how far away from the mean - the best estimate you have of the "true" time - these fastest run times were), taking the mean time of however many test runs you do is the more statistically rigorous method. So yes. JBI (and the others who mentioned it) was justified in pointing this out.

The most interesting thing here is actually not which time is the truest time (the fastest times or mean time of each group) but that the variance in the non-screw fastened drives is lower (as well as them having faster times). This would suggest that screw mountings are in deed more secure and help reduce deficits in access time due to interfering vibrations. So either way, your hypothesis is vindicated by your data.

EDIT: BTW, A 2-sided t-test for two samples with unequal variances results in a p-value of 0.0016. So this would suggest that the screwed-down drives have statistically significantly faster access times. Though you are correct that more samples would be better just to be more certain that these results weren't artifactual.