Access time question

[januszian]

Hello.

So a couple of general questions about HDD access time.

Access time seems to be the magic word when you think of speed. Is low access time always = lots of I/O ?

WD74 raptor is a legend. But I can't afford it. So I'm going to buy WD3200AAKS and partition it to use only 70GB. This should get me better performance than a 74GB raptor?
According to this article:
http://blogs.zdnet.com/Ou/?p=322

Also, will the remaining 250GB be usable as a storage? Or will it slow things down? If some program (virus scanner, spyware scanner or Diskeeper) accesses the partition for some reason?

Another question:
does RAID 0 decrease access time? I don't need RAID 0 for my uses but I'm just curious...

My usage for these harddrives will be system disc and editing disc for video editing. The drives have a good enough transfer rate for DV and HDV editing. So it's just a matter of access time as I understand it?

From strictly access/seek time point of view: Will I get similar performance if I partition a WD3200AAKS to use 70GB? Compared to a 74GB raptor.
Or are there other factors that affect video editing usage in which the raptor dominates?

 

[jsc]

Oh, my. Apples, oranges, and grapefruits here.

Speed with respect to hard drives has two components. Access time is determined by physical characteristics of the drive - how fast the mechanical parts can move the r/w heads to where you want to start the r/w operation. The raptors are good at this.

The other part of speed is how quickly you can read or write the data (transfer rate) once the heads get there. Because the data density on the platters is so high, the 320 GB, 500 GB, 750 GB, and 1 TB drives are good at this. Add SATA II into the mix, and there's not much practical use for drives like raptors in consumer systems.

(Look at the signature systems of a lot of the regular posters here. Many of us could install raptors, but choose not to. We choose to spend the money elsewhere.)

"WD74 raptor is a legend. But I can't afford it. So I'm going to buy WD3200AAKS and partition it to use only 70GB. This should get me better performance than a 74GB raptor?" NO. A sizeable part of the improved performance is coupled to the mechanical drive system of the raptor. A WD320 or 500 just doesn't spin that fast. All you are doing is throwing away 250 GB of storage. Oh, and raptors generate a lot of heat.

"Also, will the remaining 250GB be usable as a storage? Or will it slow things down? If some program (virus scanner, spyware scanner or Diskeeper) accesses the partition for some reason? " If you choose to use the space in a second partition, yes, it will slow things down. Having said that, you will need a benchmark program to measure the difference. You probably will not notice anything.

"Another question: does RAID 0 decrease access time?" No, but it will increase transfer rate. And a lot us think that the risks of Raid 0 are not worth the speed increase.

"My usage for these harddrives will be system disc and editing disc for video editing." In that case, depending on your budget, get a 250 - 320 GB drive for your boot drive and a 500 GB to 1 TB drive for your data. And take the money you were going to spend on raptors and buy an Intel C2Q, a decent HSF to OC with, and 4 GB of memory.

"Or are there other factors that affect video editing usage in which the raptor dominates?" Raptors do not dominate in video editing. What raptors are good for is moving lots of little files around quickly. Video editing requires moving one or a few really, really, large files around. Their namesakes are excellent at swooping down out of the sky and carrying a rabbit away. I have never heard of a hawk flying away with a cow.

 

[januszian]

So I said I'm NOT going to buy a raptor...

And Raptors didn't fly? They were land dinosaurs?

Anyway, I'm getting an Intel 9450 Core 2 Quad with Noctua NH-U12P so that a "check."

You said:

"A sizeable part of the improved performance is coupled to the mechanical drive system of the raptor. A WD320 or 500 just doesn't spin that fast. All you are doing is throwing away 250 GB of storage. Oh, and raptors generate a lot of heat."

But according to the article:

"But this is a comparison on server drive components and we can actually see a more dramatic effect when we’re talking about the desktop storage market. In that market, you will actually pay DOUBLE for 1/4th the capacity on 73 GB 10K SATA RPM drives than typical 300 GB 7200 RPM SATA hard drives. Now the speed difference is more significant since the 7200 RPM drives have typical average seek times in the 8.9 millisecond range and you have to add 4.17 milliseconds average rotational latency for a relatively pathetic access time of 13.07 milliseconds. The 10K RPM SATA drive designed for the enthusiast performance desktop market has an average access time of 7.7 milliseconds. But since the 300 GB 7200 RPM drive is 4 times bigger than the 73 GB 10K drive, we can actually use quarter stroking and end up with a high-performance 75 GB partition along with a 225 GB partition we can use for large file archival such as a DVD collection.

By quarter stroking the 300 GB drive, we can actually shave 6.68 ms off the seek time which means we’ll actually end up with an average access time of 6.4 milliseconds which is significantly faster than the 10K RPM "performance" drive. This means that PC enthusiasts are paying twice the money for a slower hard drive with a quarter of the storage capacity!"

So according to him, I can cut down the seek time enough that there will be something like 2 ms and latency somewhere like 4 ms. These are averages...

Does the raptor have something else that matters? Or is the average 6.4 ms access time on the 70GB partition better than a 74GB raptor?

And the reason why I was looking at the WD3200AAKS is that it costs about 60 euros. Compared to the 120 euros price tag that the raptor has.
I really don't mind that I'm "throwing away" the rest of the disc space.
The partition trick will be used for the system disc only...

 

[SomeJoe7777]

First, let's be clear on the labeling of these parameters. It's easy to get confused and therefore become inaccurate.

Seek time: The time it takes for a hard drive head to move from one track to another. Manufacturer's quoted value for seek is generally the average time it takes the head to move 1/3 of the way across the platter, hence the terk 1/3 average seek time. This is generally in the 8.5-9.0 msec range for 3.5" platters, 4.5-5.0 msec for 2.5" platters.

Rotational latency: After a hard drive head has landed at the destination track after a seek, it now has to wait until the desired sector rotates underneath it. It could be up to a full rotation of the platter away. The time spent waiting is a function of the rotational speed of the hard drive. Manufacturer's quoted values for rotational latency are the average rotational latency, which is 1/2 of a platter rotation. This is fixed at 4.167 msec for 7200 RPM drives, 3 msec for 10,000 RPM drives.

Access time: This is the total time required to read a sector, and is simply the sum of the seek time and the rotational latency time. This is in the range of 12.5-13.5 msec for 7200 RPM drives (which typically have 3.5" platters), and 8.0 msec for 10,000 RPM drives (which have 2.5" platters).

STR: Sequential transfer rate. The maximum data rate (MB/sec) that a hard drive can deliver. This varies depending on where on the platter you are reading data from. The outer edges of the platter have the highest data rates because there is a longer linear distance on the outer edge, and therefore more bits pass under the head at the same data density.

IOPs: I/O Operations per second. This is the maximum number of data read/write transactions per second that the hard drive as whole can do. Sophisticated caching algorithms, re-ordering of reads and writes, and delayed grouping of writes are important to increase this figure.


Now, knowing this, let me throw out a few numbers from some benchmarks and specs:

WD3200AAKS: 320GB, 3.5", 7200RPM, 8.9 msec seek, 4.167 msec rotational latency, 13.067 access to data. 78 MB/sec STR on platter outer edge. Probably no more than 100 IOPs.

WD74ADFD: 74GB, 2.5" platters in a 3.5" case, 10,000 RPM, 4.6 msec seek, 3 msec rotational latency, 7.6 msec access to data. 83 MB/sec STR on platter outer edge. 172 IOPs.

What could you expect if you short-stroked the WD3200AAKS? Well, the problem is that there is some overhead for seeks. The WD website specifies that the track-to-track (i.e. adjacent track) seek time is 2.0 msec. So if you 1/4 stroke the drive, you don't get 1/4 of the 8.9 msec seek, you only get to have 1/4 of the non-overhead portion, which is only 6.9 msec. So, expected 1/3 seek when 1/4 short-stroked should be around 2.0 + 1/4 * 6.9 = 3.7 msec.

So now, you're looking at the following numbers:

WD3200AAKS: 80GB, 3.5", 7200RPM, 3.7 msec seek, 4.167 rotational latency, 7.9 msec access to data. Still 78 MB/sec STR on platter outer edge, still no more than 100 IOPs.

The Raptor will still edge you out here.

Further, you cannot use the remainder of the drive for storage and expect the average seek to remain low. If the heads ever leave the 1/4 short-stroked section, the average seek time plummets back to 8.9 msec. If you have a storage partition on the rest of the drive, your virus scanner, Windows search, other background tasks (to say nothing of Vista's caching) will periodically move to that storage partition even if you're not doing anything. To realize any benefits of short-stroking, the remaining space must be left unpartitioned and blank.

Some other notes:

STR is specifically helpful in tasks that read large files sequentially. i.e. Video capture, audio processing, DVD authoring, Photoshop work.

Access time is specifically helpful in tasks that read lots of small files in a random order, like Windows startup.

IOPs is specifically helpful in multi-process, multi-user scenarios where many outstanding requests for the drive to read/write data exist. This is primarily the number most closely looked at when selecting drives for servers.

RAID 0 is designed for one sole purpose - increase the STR. It does nothing for IOPs, and in fact causes average access times to go up.

People seem to be looking at the outer platter edge STR of some of these new drives and proclaiming them to "beat" the Raptor, when in actuality they have no idea what they're talking about. (I'm not referring to you, this is a general comment). There is far more that goes into hard drive performance than a few numbers from the benchmarks.

 

[geofelt]

It's not so simple. The article is true, as far as it goes, but it is based on random reads, and not sequential operations. There will be relatively few random reads in most cases. Many times, the read/write head will be positioned to a folder which is random, but then there will be no arm motion while some number of sectors are read in sequentially. In the real world, we will have some random, some sequential, and some degree of locality of reference.

What to do?

If you will do lots of video editing, then sequential data transfer is relatively important. There is a valid case for raid-0 here, but you will pay for it with longer random access times, and in general is not worth the bother. Read this for some considerations on raid: http://faq.storagereview.com/tiki-index.php?page=SingleDriveVsRaid0

My suggestion is not to worry too much about this, and get the capacity you need. For video editing, two smaller drives would be good if you could arrange the source on one drive, and the output on a separate drive. That way, there could be minimum arm stealing and faster throughput.

As to the raptors, the raptor150 will generally be the fastest device if you can afford it . Read this for a comparison of devices using one typical single user desktop workload: http://www.storagereview.com/php/benchmark/bench_sort.php

 

[januszian]

First, let's be clear on the labeling of these parameters. It's easy to get confused and therefore become inaccurate.

Seek time: The time it takes for a hard drive head to move from one track to another. Manufacturer's quoted value for seek is generally the average time it takes the head to move 1/3 of the way across the platter, hence the terk 1/3 average seek time. This is generally in the 8.5-9.0 msec range for 3.5" platters, 4.5-5.0 msec for 2.5" platters.

Rotational latency: After a hard drive head has landed at the destination track after a seek, it now has to wait until the desired sector rotates underneath it. It could be up to a full rotation of the platter away. The time spent waiting is a function of the rotational speed of the hard drive. Manufacturer's quoted values for rotational latency are the average rotational latency, which is 1/2 of a platter rotation. This is fixed at 4.167 msec for 7200 RPM drives, 3 msec for 10,000 RPM drives.

Access time: This is the total time required to read a sector, and is simply the sum of the seek time and the rotational latency time. This is in the range of 12.5-13.5 msec for 7200 RPM drives (which typically have 3.5" platters), and 8.0 msec for 10,000 RPM drives (which have 2.5" platters).

STR: Sequential transfer rate. The maximum data rate (MB/sec) that a hard drive can deliver. This varies depending on where on the platter you are reading data from. The outer edges of the platter have the highest data rates because there is a longer linear distance on the outer edge, and therefore more bits pass under the head at the same data density.

IOPs: I/O Operations per second. This is the maximum number of data read/write transactions per second that the hard drive as whole can do. Sophisticated caching algorithms, re-ordering of reads and writes, and delayed grouping of writes are important to increase this figure.


Now, knowing this, let me throw out a few numbers from some benchmarks and specs:

WD3200AAKS: 320GB, 3.5", 7200RPM, 8.9 msec seek, 4.167 msec rotational latency, 13.067 access to data. 78 MB/sec STR on platter outer edge. Probably no more than 100 IOPs.

WD74ADFD: 74GB, 2.5" platters in a 3.5" case, 10,000 RPM, 4.6 msec seek, 3 msec rotational latency, 7.6 msec access to data. 83 MB/sec STR on platter outer edge. 172 IOPs.

What could you expect if you short-stroked the WD3200AAKS? Well, the problem is that there is some overhead for seeks. The WD website specifies that the track-to-track (i.e. adjacent track) seek time is 2.0 msec. So if you 1/4 stroke the drive, you don't get 1/4 of the 8.9 msec seek, you only get to have 1/4 of the non-overhead portion, which is only 6.9 msec. So, expected 1/3 seek when 1/4 short-stroked should be around 2.0 + 1/4 * 6.9 = 3.7 msec.

So now, you're looking at the following numbers:

WD3200AAKS: 80GB, 3.5", 7200RPM, 3.7 msec seek, 4.167 rotational latency, 7.9 msec access to data. Still 78 MB/sec STR on platter outer edge, still no more than 100 IOPs.

The Raptor will still edge you out here.

Further, you cannot use the remainder of the drive for storage and expect the average seek to remain low. If the heads ever leave the 1/4 short-stroked section, the average seek time plummets back to 8.9 msec. If you have a storage partition on the rest of the drive, your virus scanner, Windows search, other background tasks (to say nothing of Vista's caching) will periodically move to that storage partition even if you're not doing anything. To realize any benefits of short-stroking, the remaining space must be left unpartitioned and blank.

Some other notes:

STR is specifically helpful in tasks that read large files sequentially. i.e. Video capture, audio processing, DVD authoring, Photoshop work.

Access time is specifically helpful in tasks that read lots of small files in a random order, like Windows startup.

IOPs is specifically helpful in multi-process, multi-user scenarios where many outstanding requests for the drive to read/write data exist. This is primarily the number most closely looked at when selecting drives for servers.

RAID 0 is designed for one sole purpose - increase the STR. It does nothing for IOPs, and in fact causes average access times to go up.

People seem to be looking at the outer platter edge STR of some of these new drives and proclaiming them to "beat" the Raptor, when in actuality they have no idea what they're talking about. (I'm not referring to you, this is a general comment). There is far more that goes into hard drive performance than a few numbers from the benchmarks.

Thank you SO MUCH! This was incredibly helpful!!!!

Just one more question:

According to this:

http://www.storagereview.com/php/benchmark/suite_v4.php?typeID=10&testbedID=4&osID=6&raidconfigID=1&numDrives=1&devID_0=350&devID_1=306&devID_2=259&devCnt=3

The WD7500AAKS has almost as much I/O's as the 74GB raptor.
(I don't know which Raptor is more close to current versions...)

And according to this:

http://kitami.sakura.ne.jp/mako_s/pc/bench/hdtune.htm

The WD3200AAKS has similar results as a WD7500AAKS. I won't need the more STR than what the WD3200AAKS can offer. And those HD TuNE numbers have been posted by owners in other forums too.

So wouldn't it still be worth buying a WD3200AAKS for half the price of the raptor? Or will the sequential read/write capability be too limiting?

I realize that this isn't about making my system work anymore. It's about small optimizations.

And I don't mean to be a pain in the a*s. Just wan't to get info about things that I'm interested in and to select the best components for my system, which I will be using for video, music (I'm in a band and we record our own songs) and photo editing.

 

[SomeJoe7777]

The version of the Raptor they were testing there didn't have NCQ (much older). Current NCQ version does 172 IOPs.

WD3200AAKS showed 11 MB/sec slower STR than WD7500AAKS (79 vs. 90).

For your applications, it's unlikely that IOPs or access times will really affect you much. For video, music, and photo editing, the ideal drive is one with enough storage space, with STR as a secondary consideration.

The WD3200AAKS is a fine choice, the WD7500AAKS would be even slightly faster due to the higher data density.

 

[Turas]

Wow that was some wonderful information. Thanks for that great write up.

 

[jsc]

Great writeup, SomeJoe. I was trying to keep it simple.

So I said I'm NOT going to buy a raptor...

And Raptors didn't fly? They were land dinosaurs?

Raptors describe a class of birds including eagles, hawks, and falcons. Velociraptors were land dinosaurs.

 

[januszian]

Pardon my bad language knowledge... I'm Scandinavian...

But yeah! GREAT INFO! Thanks!