Topic 'Has HDD seek times increased?' in forum 'PC Hardware Discussion/Reviews' - dslreports.com

Re: Has HDD seek times increased?

Wed, 06 Feb 2013 18:59:43 EDT

koitsu posted : {deleted -- I don't know why the forum took a good 120 seconds for my previous reply to show up. Grrr...}

Re: Has HDD seek times increased?

Wed, 06 Feb 2013 18:55:47 EDT

koitsu posted :

said by trparky:

The question is... are they doing something special in the Caviar Black drives to make them run better or are the manufacturers purposely gimping their low end drives to make people buy the high end drives?

It's not that simple. Trying to make some generalised statement about the state of mechanical/magnetic storage isn't the right way to go about this, mate.

As I said, the question should be "Why do my ST31000528AS and HTS725050A9A364 perform so significantly worse compared to other MHDDs?"

There are an almost infinite number of reasons that can explain what's going on, but the important part to take away is that every situation is unique/different and must be treated as such. I can't stress this enough. Otherwise you end up with generalised statements like this which amount to absolutely nothing.
--
Making life hard for others since 1977.
I speak for myself and not my employer/affiliates of my employer.

Re: Has HDD seek times increased?

Wed, 06 Feb 2013 18:36:03 EDT

trparky posted : Granted, that WD2002FAEX is a Western Digital Caviar Black Edition drive. That drives performs well for a traditional HDD.

The other drivers are a different story. They seem to be drives that are the standard run of the mill consumer drives that people buy just because they need an HDD. Yeah... the low end drives.

OK, so we know that HDDs, at least from the desktop benchmarks, that some HDDs can perform very well but you're going to pay for that performance. I get that. The question is... are they doing something special in the Caviar Black drives to make them run better or are the manufacturers purposely gimping their low end drives to make people buy the high end drives?
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Wed, 06 Feb 2013 18:09:41 EDT

koitsu posted :

said by trparky:

I don't ever remember benchmarks for hard drives being this piss poor. What has happened to hard drives lately?

Did you notice that your screenshots show a benchmark for a WD2002FAEX which refutes your own statement? :-)

So really, the question shouldn't be "Why are MHDDs getting worse" (they aren't), but instead "Why do my ST31000528AS and HTS725050A9A364 perform so significantly worse compared to other MHDDs?"

There isn't enough data provided by you at this time to answer the latter question.

Also, be sure to keep something in mind: while I can believe the read benchmarks for all of these drives, I cannot necessarily believe the write benchmarks. The reason is that you have active filesystems on the drives, so doing sequential LBA writes would completely stomp all over the filesystems. This means the benchmarking software in question is probably doing something like creating a sparse file of an extremely large size (say 25GBytes), does an lseek() to offset 0 in the file, proceeds to do benchmarked fwrite()s, takes note of the response time, then when finished deletes the sparse file. If that sparse file starts near the end of the drive (meaning around the inner portions of the platters), it's going to show worse results than if it starts LBA 0 (the outer portion of the platters). In turn this means the filesystem layer plays a large role with regards to the performance data shown.

ATTO is another benchmarking utility that does this (which is fine, I use ATTO all the time for this exact purpose), while HD Tune Pro does it right -- the "Benchmark" tab does actual LBA reads/writes (which is why you can't have a filesystem on the drive you're write benchmarking), while the "File benchmark" tab does exactly what I described above.

Welcome to benchmarking software 101, and why I avoid benchmarks.
--
Making life hard for others since 1977.
I speak for myself and not my employer/affiliates of my employer.

Re: Has HDD seek times increased?

Tue, 05 Feb 2013 14:38:36 EDT

trparky posted : Here are some comparisons between the drives in two of my systems. Each benchmark was done by the same software, the Samsung SSD Magician software.

Desktop Drive Comparisons
[att=1]

Notebook Drive Comparisons
[att=2]

Look at the notebook drive comparison. It's garbage... utter garbage!

I don't ever remember benchmarks for hard drives being this piss poor. What has happened to hard drives lately?
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Desktop Drive Comparisons
Notebook Drive Comparisons

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 14:20:53 EDT

aurgathor posted :

said by trparky:

I understand that there is more than seek times when it comes to HDD performance but it seems that the performance of today's HDDs have gone down.

You're wrong.

I recommend you to try drives of various vintage in the same system under identical conditions. (make a drive image of any reasonable recent Windoze install, and put that into various drives)

Now, I'm not talking about those synthetic benchmarks like what HDTune performs, I'm talking about real world performance numbers which are those numbers that come up when you actually use the drive for everyday operation.

Due mostly due to bloat and increased resolution, HDs need to read a lot more data to complete the equivalent operations.

Now, this is a theory of course.

Well, an incorrect assumption on your part.

The most important parameters that determine HDs performance are:
a) areal density
b) spindle speed (RPM)
c) actuator speed
d) size of platters
e) number of platters

These are the main parameters designers can play with, although there are many other things they can do like adding big caches, or having two sets of heads.

Capacity depends mostly on areal density, size, and number of platters.
Seek time depends on spindle speed, actuator speed, and the size of the platters.
Sustained DTR (Data Transfer Rate) depends on areal density, spindle speed, and the data's location on the platter. (assuming the electronics is fast enough)

Having said all this, most of the improvements lately have been in capacity and DTR -- seek times didn't go up as much because of their dependency on spindle speed, but if someone needs good seek times, SSDs provide a much better solution.
--
Wacky Races 2012!

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 13:07:11 EDT

trparky posted : I understand that there is more than seek times when it comes to HDD performance but it seems that the performance of today's HDDs have gone down.

Now, I'm not talking about those synthetic benchmarks like what HDTune performs, I'm talking about real world performance numbers which are those numbers that come up when you actually use the drive for everyday operation.

Now, this is a theory of course. I know that defragmentation software tries to put all of the EXE files in a contiguous pattern on the drive and some even go further and move them to the fastest parts of the drive. I have to wonder if the defragmentation software that does move the EXE files to a "faster" (I know, marketing term here) part of the drive also moves dependencies as well, such as DLLs, data files, images, and other resources.
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 13:02:56 EDT

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 12:55:46 EDT

trparky posted :

said by HarryH3:

While transfer speeds have increased (drive rotation speed has remained constant (assuming we're talking about 7200 RPM drives) but the data bits are closer together, thus more bits pass the head in a given amount of time) there are limits to decreasing seek times.

That's why I would think that seek times would decrease with greater data densities. Since the data is packed tighter, the drive's head assembly wouldn't have to move as much to go to the target LBA.
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 12:52:37 EDT

trparky posted :

said by HarryH3:

The Windows 8 .iso file is nearly 3 GB. :o And the first thing it did after installation was inform me that it needed to download some critical updates, around 450 MB worth.

But most of these files that it downloads ends up being extracted and replacing existing system files. You don't have these files laying about for the system to load at boot each time it boots, those files have replaced existing system files like kernel32.dll and ntdll.dll with newer versions.
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 11:08:23 EDT

HarryH3 posted :

said by trparky:

Obviously, I know that software has become bloated over the years but has it really gotten that bad as of late? I know it was bad but is it really that bad?

The Windows 8 .iso file is nearly 3 GB. :o And the first thing it did after installation was inform me that it needed to download some critical updates, around 450 MB worth. There are a lot more lines of code running today than in the past. The drive has to read them before it can pass them along to RAM. Just the AMD video driver package download is around 140 MB these days.

While transfer speeds have increased (drive rotation speed has remained constant (assuming we're talking about 7200 RPM drives) but the data bits are closer together, thus more bits pass the head in a given amount of time) there are limits to decreasing seek times. It takes time for the head to physically move from point A to point B. The head and arm assembly have mass that has to move VERY quickly and yet stop EXTREMELY accurately to find the correct track. The laws of physics make it very difficult to make both of those happen any faster, at a reasonable cost.

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 10:50:28 EDT

trparky posted : It's coming to the point where you need an SSD to even have a decent boot time let alone programs that load fast.

I remember when clicking on a program icon meant having to wait a few seconds for the program to load. On my SSD the program loads like... damn, was the program already running? They load so fast that I could've sworn the process was already running.
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Sat, 02 Feb 2013 10:48:09 EDT

trparky posted : Obviously, I know that software has become bloated over the years but has it really gotten that bad as of late? I know it was bad but is it really that bad?

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 23:55:44 EDT

HarryH3 posted : You can attribute part of it to code bloat. The sheer amount of data that has to be read just to open a file today is orders of magnitude higher than it used to be. Windows 3.1 used to come on a few floppy disks, then Windows 95 came on several floppy disks (or one of those new fangled CD's!). These days Windows is so huge that it comes on a DVD.

The amount of RAM in many systems today exceeds the hard drive sizes of not too long ago. Somewhere around here I have a laptop from around 1998 that has a whopping 6 GB hard drive. It was amazing in its day. But after I loaded Windows 2000 on it and a few Service Packs came out to bloat the code even more, it now takes around 5 minutes just to boot. :(

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 23:05:46 EDT

trparky posted : What I'm getting at is that it seems to me that HDDs performed better in the past than they did today. I have no idea why that would be the case since the underlining technology of HDDs hasn't changed in decades.

I understand fragmentation but even when properly defragged today's HDDs seem slower than in years past. Is it because we are demanding more from our drives and that they can't keep up with the raw I/O requests that we are making to them?

Windows operating systems haven't changed much when you look at how it reads and sends requests to the HDD. Linux however has changed many times and you can even change it yourself by changing the I/O scheduler which effects how the system is going to send I/O requests to the HDD. There are about a dozen different I/O schedulers that you can choose from to compile into your kernel. Some are made to perform better on HDD and some are better when performing on solid state flash memory.
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 20:01:13 EDT

koitsu posted : No, that isn't how the heads work at all. It's 1:1 as far as I know -- otherwise when reading LBA 12345, physically the drive would try to read LBA 12345 + 12346 + 12347 + 12348 (say, a 4-sector read-ahead), and if one of those subsequent LBAs was bad, you'd have no real way of knowing. My point is that there is no read-ahead at the head level (that I've seen anyway). You should be able to find some really well-written documentation (PDFs) online explaining how heads work in combination with things like perpendicular recording (which is what paved the way for drives to greatly increase in capacity while keeping the same form factor; this would have been in roughly 2006).

The issue you're complaining about has more to do with filesystems than it does with the actual hard disk. You're assuming that all of the data stored on the drive is stored in a 100% contiguous/linear model -- it isn't. FAT, FAT32, and NTFS are, if I remember right, the worst filesystems when it comes to fragmentation. Other filesystems like ext2, ext3, UFS/FFS, ZFS, and many others (which tend to be used by UNIX or UNIX-like operating systems) have very, very small fragmentation percentages (2-3% at most).

Drives today easily outperform drives from 6-7 years ago. I can get 150-160MBytes/second sequential reads from a 7200rpm 3.5" drive along the outer edge of the platter; 6-7 years ago that was more along the lines of 110-120MBytes/s. Write speeds tend to be around the same, maybe a tiny bit slower. On some models of drives today -- such as "low-power" drives (sometimes called "GreenPower" or "GP") -- performance is much more along the lines of what it was 6-7 years ago. There are a couple reasons that explain why they perform worse, but the one I've been trying to determine for years now isn't really explained/documented anywhere (I have a feeling it's purely a drive firmware design/decision driven by marketing, but that's speculation on my part). In fact, in the above Wikipedia article, look under the "Implementations" section citing speeds of a drive in 2006 (75-125MByte/sec). Drives today easily outperform that.

There is one aspect to the increased data density that does affect performance, however -- as we cram more and more bits into the same physical space, the chance of error does increase. You need to keep in mind how tightly packed these bits are, and things like how close the heads are to the platters. I'm not talking about actual I/O errors (as in a "bad sector") -- I'm talking about the drive having to auto-correct bit errors when reading a sector using the ECC portion of the sector. This happens automatically within the drive and is not an I/O error. With magnetic media, the more writes you issue over time the "worse" that area of the platter actually performs -- meaning there is a degree of quality loss over time that ECC makes up for. In some cases, heavily used drives (say, 10000 to 15000 power-on hours or more which have done lots and lots of I/O) can perform significantly worse compared to when they were purchased yet show no signs of degradation from a SMART or host controller perspective.

This is one of the downsides of magnetic media -- it doesn't last forever, and how long it lasts is affected by density. Everyone wants more space, more space, more space, but doesn't want to give up the form factor. This is why I've advocated bringing back 5.25" form factor MHDDs. But it probably won't matter with the introduction of SSDs and so on -- seek time is no longer an issue there, nor is the downsides to magnetic media.

Otherwise you're making some general vague/wild statements about hard disks today without any actual hard technical data to back them up. If you give some actual data (models of drives, SMART attributes of each drive, and benchmark comparisons between them done on the same machine) I'd be happy to analyse and discuss the results.
--
Making life hard for others since 1977.
I speak for myself and not my employer/affiliates of my employer.

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 19:12:02 EDT

trparky posted : I would think that with the higher capacities the seek times would drop because, in theory of course, the data is closer together the read/write head would not have to move as much to get to where the data is.

Think of it this way. It takes you less time to get somewhere in the city because everything is closer together and because everything is closer together it takes you less time to get from home to the store. Crude way to look at it, yes, but it gets there job done.

So what is happening in these drives? Are they missing the target? Is the read/write head not getting to where it needs to be in time for the operation to complete thus it has to wait for the sector to come back around on the next rotation?

Hard drives used to not be this slow. I have no idea why these drives perform so poorly these days.
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 14:45:52 EDT

koitsu posted :

said by trparky:

I've noted that with the recent capacity increases with HDDs, speeds haven't improved at all, if anything, it's gotten worse. And when I say speeds, I mean seek time. You'd think that with the data being packed tighter on the platters the seek times wouldn't(sic) be this bad.

As pnjunction said, platter density has absolutely nothing to do with seek time. Seek time quite literally is the amount of time for the actuator arm to move from point X to point Y on the platter(s). Seek times advertised in specifications are an average; real-world seek times are both less than and higher than the average, greatly dependent upon multiple factors (the common one is distance between X and Y while taking into consideration actuator arm speed combined with rotational speed). This is where NCQ comes into play (without NCQ, the drive more regularly has to wait an entire rotation, since the queue is not optimised; the Wikipedia diagram should suffice).

The number of heads to platters is 2:1 (one head for each side of the platter), but there is only one actuator arm. I imagine the trade-offs of adding a 2nd actuator arm have been weighed many times over the years (cost vs. reliability vs. noise vs. power usage vs. availability of key metals and natural resources).

said by trparky:

I know, for the head to go from one part of the drive to another, it has to stop off at Track 0 Sector 0 first then go to the new position.

This is incorrect. The only time the drive will sit at LBA 0 is when its asked to read/write there. During or periods where the drive needs/wants to park its heads, the heads will rest over an area on the platters which is unused (I forget if it's along the outer edge or the inner edge). Full off-platter head parking I don't think is used any longer, given the risks.

Also, nobody (aside from possibly legacy PC BIOS INT 0x10) uses CHS addressing any longer. Referring to things by "track and sector" is long deprecated -- people still use the term to describe "areas of a platter" but that actual model isn't particularly used any longer. LBA is what's used now. Are you still using 8" floppy disks and writing code to manually adjust the stepper motor yourself? :-) For internal referencing of platter locations, I honestly have no idea what terminology drive manufacturers use internally -- as I've said in past posts, I would love to get to know some folks at WD, Seagate, etc. (engineers, not PR people) to learn more.

said by trparky:

And don't me started on notebook drives, those were bad before and now they're even worse in the performance tests. They have consistently scored at the bottom of the performance charts when compared to that of traditional desktop drives, even those notebook drives that run at 7200 RPM (I know, few of them do).

There are a lot of reasons for this (I can expand on them if you want, it has to do with a lot of things unrelated to the technology and more about the usability difference between a laptop and a dekstop), however I'll point you to a counterexample: Western Digital's VelociRaptor drives. 10krpm, 2.5" form factor. Capacity is lower because it has to be (look up the term "short stroke").

In general laptops today should really be using SSDs solely for the lower power usage and increased reliability (WRT no moving parts -- this matters much more in laptops than in desktops).
--
Making life hard for others since 1977.
I speak for myself and not my employer/affiliates of my employer.

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 14:25:22 EDT

aurgathor posted : You're probably not looking at the right drives. There are many variables and sometimes the effects are inversely proportional when adjusting one. There are ways to increase capacity, but some can have a detrimental effect on seek time, and consumer oriented drives usually go for capacity.

There are enterprise and other high performance drives (i.e.: WD RE4s and Raptors, and most of the SAS drives) that will be much faster than the usual consumer HDs, be it 5400/5900 or 7200 RPM.

You just need to choose your HD accordingly.
--
Wacky Races 2012!

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 13:34:49 EDT

trparky posted : I have an SSD for my system boot drive, it really makes the system take off like a rocket.

Re: Has HDD seek times increased?

Fri, 01 Feb 2013 13:31:47 EDT

pnjunction posted : Platter density doesn't improve seek time at all, to my knowledge it is dictated mostly by the physical size of the platter and drive mechanics, including rotation speed. Without knowing details I'd guess it is not easy to improve those mechanics beyond where we have been for a while, at least not cheaply.

If you care about seek time scrape together the cash for SSDs. They are well under $1/GB now and there are deals to be had at around $0.75/GB.

Has HDD seek times increased?

Fri, 01 Feb 2013 13:24:40 EDT

trparky posted : I've noted that with the recent capacity increases with HDDs, speeds haven't improved at all, if anything, it's gotten worse. And when I say speeds, I mean seek time. You'd think that with the data being packed tighter on the platters the seek times would be this bad.

I know, for the head to go from one part of the drive to another, it has to stop off at Track 0 Sector 0 first then go to the new position.

And don't me started on notebook drives, those were bad before and now they're even worse in the performance tests. They have consistently scored at the bottom of the performance charts when compared to that of traditional desktop drives, even those notebook drives that run at 7200 RPM (I know, few of them do).
--
Tom
Boycott AT&T uVerse! | Tom's Android Blog | AOKP (The Android Open Kang Project)