[mediahound]

I'm in the market for a new hard drive or two and my friend said when you get up over 1TB, a 5400rpm drive is not much slower than a 7200rpm drive due to the data density. Is this true or a load of bull?

I'm asking because the quiet drives are all 5400rpm but I do do some video encodes and things like that and I've used 7200rpm drives for the past several years.

I'll probably go with a Samsung drive. Their Spinpoint 1TB is 7200 but I would rather have 1.5TB of storage, which only comes in 5400rpm...

[Uisce]

As I understand it, faster drives are faster, but consume more power, so if you're working unplugged, the faster drive will consume battery power quicker. No clue on data density.

[mediahound]

I'm not super concerned about power as this is for a MacPro. I'm more concerned about speed and noise level.

Originally Posted by Uisce 

As I understand it, faster drives are faster, but consume more power, so if you're working unplugged, the faster drive will consume battery power quicker. No clue on data density.

[reader50]

1.5 TB and 2.0 TB drives are available in 5400 and 5900 RPM. They are around 90 MB/s sequential performance (start of drive), vs 110-120 MB/s sequential for 1 TB 7200 drives. That's fairly good, especially if you RAID them. There's good reason to believe 7200 versions will be notably faster when they arrive.

- Hitachi has announced a 2.0 TB 7200 RPM drive. Not shipping yet, and it is a 5-platter (400 GB per platter).
- Western Digital had some marketing material leak. They're planning to announce a 2.0 TB Caviar Black at 7200 RPM. Not yet announced though, much less shipping. Platter count/density unknown.
- Samsung has announced the 7200 RPM Spinpoint F3 series, at 500 GB per platter. A 3 or 4 platter version would have 1.5 TB and 2.0 TB respectively. However, Samsung has not announced anything past the 2 platter 1.0 TB version.

Here is a performance chart of the Samsung F3 7200 RPM. It's for the 500 GB single platter, but performance should be much the same for any 7200 RPM drive with 500 GB per platter. It starts at 140 MB/s sequential.

[Simon]

Faster rpm drives are faster provided the data density per platter is equal and the capacity and available capacity are equal.

IOW a single 250 GB platter at 5400 rpm (as in some 500 GB drives) can be faster than a 160 GB platter at 7200 rpm (as in some 320 GB drives). A 320 GB 5400 rpm drive with 20 GB used will be faster than a 320 GB 7200 rpm drive with 300 GB used.

Bottom line, when buying, look for high capacity platters and high capacity drives first. Then, look for high rpm.

If you're seriously worried about noise/heat you should be looking at drives like the WD Caviar Green or Seagate Barracuda LP series. These drives are below 7200 rpm. Most are 5400 or 5900 rpm.

[Spheric Harlot]

Also to keep in mind:

A high-RPM but full drive can easily be SLOWER than a lower-RPM but empty drive.

IOW: Keep a Terabyte 7.200 drive loaded with 950GB of data, and it will probably be slower than a 2 Terabyte 5.400 drive loaded with those same 950 GB of data.

Number will vary with models, of course.

For further reading, Tom's Hardware generally appears to have pretty good analysis, as well.

[mediahound]

Thanks for the input everyone.

So, do you think rpm is not a very important thing to consider when making a buying decision? Should I just go with a quiet drive in the capacity I want (1.5tb even if it's not 7200rpm) since the speed of data transfer will be fairly variable anyways?

[reader50]

If the other variables are the same, higher RPM will give you greater speed. At this moment, higher RPM is not readily available above 1.0 TB. Unless you want to take your chances with Seagate's 1.5 TB 7200.11 model. That drive's reliability seems to have improved, but it's hard to tell from customer reviews if it's fully cleared up.

If you need the storage space now, pick up the drives you need. Especially for backups, buying slightly slower drives is better than losing data. If you can wait, the 7200 drives > 1 TB are coming. They will be faster. If you can wait ~5 years, the solid state drives will be cost effective for mass storage, and they will blow all normal hard drives away.

So I'd give the standard advice for electronics purchases. Buy what you need today. If it isn't urgent, waiting will produce better options in the future.

[mediahound]

Thanks. Actually right now there are 1TB 7200 drives but not a lot of 1.5tb 7200rpm ones.

Maybe I should just go with 1TB drives now.

Originally Posted by reader50 

If the other variables are the same, higher RPM will give you greater speed. At this moment, higher RPM is not readily available above 1.0 TB. Unless you want to take your chances with Seagate's 1.5 TB 7200.11 model. That drive's reliability seems to have improved, but it's hard to tell from customer reviews if it's fully cleared up.

If you need the storage space now, pick up the drives you need. Especially for backups, buying slightly slower drives is better than losing data. If you can wait, the 7200 drives > 1 TB are coming. They will be faster. If you can wait ~5 years, the solid state drives will be cost effective for mass storage, and they will blow all normal hard drives away.

So I'd give the standard advice for electronics purchases. Buy what you need today. If it isn't urgent, waiting will produce better options in the future.

[Simon]

Well so tell us then, how much data are you planning on putting on the drive?

900 GB on a 5400 rpm 1.5TB drive will show better performance than those same 900 GB on a 7200 rpm 1TB drive.

That's what I was trying to say in my post above. Decide on the capacity requirement first. Then, compare anticipated performance with cost/availability. Never the other way around.

[Atheist]

The OP says he does "some video encodes and things like that". Surely the encoding will be significantly slower than slowest hard drive. From a real world standpoint, does the hard drive speed really matter that much?

[mediahound]

I'll have about 500GB initially and it will gradually grow over time...

Originally Posted by Simon 

Well so tell us then, how much data are you planning on putting on the drive?

900 GB on a 5400 rpm 1.5TB drive will show better performance than those same 900 GB on a 7200 rpm 1TB drive.

That's what I was trying to say in my post above. Decide on the capacity requirement first. Then, compare anticipated performance with cost/availability. Never the other way around.

[Simon]

In that case I'd get either a 1.5TB at 5400 rpm or a 1TB drive at 7200 rpm.

Seagate Barracuda LP, 1.5 TB, 5900 rpm, $109 shipped
Samsung Spinpoint F1, 1 TB, 7200 rpm, $85 shipped
Seagate Barracuda 7200.12, 1 TB, 7200 rpm, $90 shipped
And as a reference
Seagate Barracuda 7200.11, 1.5 TB, 7200 rpm, $119 shipped

[OreoCookie]

I second Simon's suggestion, capacity > RPM, especially since you need a lot of storage.

[Tomchu]

How much data is on the hard drive won't really affect performance at all, unless fragmentation is an issue. I don't know where this kind of advice comes from. There's also no guarantee that the file system will choose to allocate space sequentially from sector 0, so the whole discussion about the outside of the platter being faster than the inside is moot unless we actually investigate how HFS+ does its thing.

Go with the 7200 RPM drives -- it's not just about sequential transfer rates, but also about the seek time. The difference between 90 MB/s and 110-120 MB/s is significant (and well worth it IMO), but the difference in seek time is what makes the difference between Snappy(™) and not-so-Snappy.

They're not going to be noticeably louder or hotter than 5400 RPM drives, either. In my experience, 5400 RPM drives are harsher-sounding than their 7200 RPM counterparts, anyway. Other people's experiences may differ.

[Spheric Harlot]

Originally Posted by Tomchu 

How much data is on the hard drive won't really affect performance at all, unless fragmentation is an issue. I don't know where this kind of advice comes from. There's also no guarantee that the file system will choose to allocate space sequentially from sector 0, so the whole discussion about the outside of the platter being faster than the inside is moot unless we actually investigate how HFS+ does its thing.

Am I misinterpreting this graph?

All hard drives have performance curves like this:



I thought it was "fullness of drive" along the X axis.

Or is it actually graphing the surface towards the outside of the platters along the X axis?

How would the benchmarking tool go about this?

Isn't that a function of HD firmware? How would software override? (Especially since the graph doesn't read "distance from center over all platters", but "XX GB".)

[Tomchu]

It's graphing the sequential transfer rate across all sectors, as they go from the outside (sector 0) to the inside (sector N). The decreasing performance doesn't have anything to do with the amount of data on the drive, but rather with physics -- over a single platter rotation, more surface area flies under the read/write head on the outer edge than on the inner edge.

Also due to physics is the fact that the latter 50% of the platter (going by distance from the centre) contains >50% of the storage capacity of the hard drive. I'm sure there's an equation for this somewhere. Notice how the slope of the graph is gradual at first and then starts dropping more and more rapidly?

SSDs, because they don't have a spinning platter, maintain consistent read/write performance all across the board -- no matter the sector location.

[Spheric Harlot]

wouldn't any drive manufacturer worth his salt make sure that the fastest areas are used first - resulting in a *direct* correlation between speed and amount of data on the disk?

[Tomchu]

That's not up to the drive manufacturer, it's up to the OS and file system. The hard drive is a dumb storage device -- it doesn't understand file systems.

Get something like iDefrag (even the demo), run it, and let it show you visually how the data you have is stored across the hard drive.

[Spheric Harlot]

So it's a software thing, not a firmware thing.

I see.

[mediahound]

Not wanting to change the subject, but as an aside, is iDefrag something that is worth using/paying for?

[OreoCookie]

Originally Posted by Tomchu 

It's graphing the sequential transfer rate across all sectors, as they go from the outside (sector 0) to the inside (sector N). The decreasing performance doesn't have anything to do with the amount of data on the drive, but rather with physics -- over a single platter rotation, more surface area flies under the read/write head on the outer edge than on the inner edge.

You misunderstand Simon's argument: if you fill a 1 TB drive with 900 GB, the last bits of data reside on the inside of the platters, the slowest part. If you fill a 1.5 TB drive with 900 GB of data, the last bits of data reside within about the first half (more data fits on the outside than on the inside). Thus, on the larger 1.5 TB drive, the throughput may be faster on average, because your data lives mostly on the outside of the platters (which means they can be read and written faster) while on the smaller 1 TB drive, you need almost all of the space available and part of your data will be on the slowest part.

Hence, larger capacity drives can be faster.

Originally Posted by Tomchu 

That's not up to the drive manufacturer, it's up to the OS and file system. The hard drive is a dumb storage device -- it doesn't understand file systems.

That's not quite true: if you create partitions, you always start from the `fast part' and then go to the `slow part.' This is before filesystems take over. Obviously, modern operating systems know about this for a long time and they will try to keep data in the faster parts. But before any OS sees the drive, it needs to be partitioned and formatted.

[mediahound]

Originally Posted by OreoCookie 

if you fill a 1 TB drive with 900 GB, the last bits of data reside on the inside of the platters, the slowest part.

I really don't believe that operating systems fill up a hard drive radially like that though. Data is typically scattered all over the drive even when it's not full, right?

[Tomchu]

Originally Posted by OreoCookie 

You misunderstand Simon's argument: if you fill a 1 TB drive with 900 GB, the last bits of data reside on the inside of the platters, the slowest part. If you fill a 1.5 TB drive with 900 GB of data, the last bits of data reside within about the first half (more data fits on the outside than on the inside). Thus, on the larger 1.5 TB drive, the throughput may be faster on average, because your data lives mostly on the outside of the platters (which means they can be read and written faster) while on the smaller 1 TB drive, you need almost all of the space available and part of your data will be on the slowest part.

Hence, larger capacity drives can be faster.

Can be -- if storing files was as easy as "put it as close to the outside as possible". In reality it doesn't happen like that.

Originally Posted by OreoCookie 

That's not quite true: if you create partitions, you always start from the `fast part' and then go to the `slow part.' This is before filesystems take over. Obviously, modern operating systems know about this for a long time and they will try to keep data in the faster parts. But before any OS sees the drive, it needs to be partitioned and formatted.

Yeah, but since we're on a Mac forum, and since the OP is a Mac user, I'll bet you five Internet dollars that there will be no fancy partition schemes happening. I think the percentage of Mac users who actually create more than one partition per drive is in the single digits -- therefore, the chances are that we're talking about one single partition covering the whole drive.

Originally Posted by mediahound 

I really don't believe that operating systems fill up a hard drive radially like that though. Data is typically scattered all over the drive even when it's not full, right?

Correct.

To illustrate my point, here's an iDefrag screenshot -- take a look at the bottom bar. That bar represents my hard drive (80 GB Intel SSD) as a sequential storage device, with the left edge being sector 0. I only have 17 GB of data on this drive, but do you notice how a) there's a moderate amount of data fragmentation, b) there's a LOT of free space fragmentation (the gaps between data are too fine and too plentiful for the resolution of the bar to illustrate, therefore it looks like I have more than 17/80 GB used), and c) data is all over the place physically, and definitely not sequentially stored.

I recently reformatted this drive (about 2 months ago), and installed a fresh Snow Leopard build. This state of things is only after moderate use.

Note that there's generally no difference (not yet, anyway) in how the file system behaves, regardless of the medium (magnetic, or solid-state). This demonstration applies equally to regular hard drives.

[OreoCookie]

Originally Posted by Tomchu 

Can be -- if storing files was as easy as "put it as close to the outside as possible". In reality it doesn't happen like that.

The data has to be put on slower parts of the drive if you use the smaller one.

Originally Posted by Tomchu 

Yeah, but since we're on a Mac forum, and since the OP is a Mac user, I'll bet you five Internet dollars that there will be no fancy partition schemes happening. I think the percentage of Mac users who actually create more than one partition per drive is in the single digits -- therefore, the chances are that we're talking about one single partition covering the whole drive.

It's not about whether the OP uses a particular partition scheme. I'm merely pointing out that the harddrive is aware of which blocks are faster and thus should be filled sooner than the slower ones. This is put into the firmware which translates between the logical geometry (number of heads, cylinders, etc.) and the physical geometry. The OS only sees the logical geometry.

[Tomchu]

"The data has to be put on slower parts of the drive if you use the smaller one."

No. See my screenshot. Also, what happens when you write some files, make changes, delete a few here, create a few bigger ones there, etc.? The file system will use multiple criteria in determining where to allocate the space for a brand-new file. It may have the option of putting some of it near the beginning of the drive, with the disadvantage that the file will have to be fragmented. It may also have the option of putting it near the end of the drive, and not having to split it up. Or perhaps the file you're writing is tiny, and can fit into virtually any sliver of free space. The criteria change yet again in such a case.

Simply put, none of what you are describing works the way you're describing it. The OS controls which logical block a file fragment gets written to. The hard drives maps physical blocks pretty much linearly to the logical blocks specified in the ATA commands, except when it has to use bad block reallocation (which makes use of a lookup table, as far as I understand).

Again, the hard drive *does not* entirely control where bits are written. For the hard drive to be able to do what you claim, it would have to be aware of the file system in use, it would have to understand it, and it would have to reference the file system information structures in order to figure out whether a particular block is in use by a file or not. That is an IMMENSE amount of overhead and complexity to add to the hard drive firmware, and it just doesn't happen.

The hard drive is a dumb storage device. It presents block storage capabilities. It's the job of the file system to keep things organized, to optimize where things are written, and to deal with data/space fragmentation. That's why it's the FILE SYSTEM.

I suggest you read up on the SCSI spec. SCSI is simpler than ATA, and it illustrates the relative ignorant nature of the hard drive.

[Tomchu]

I see that you edited your post.

The OS does indeed only see logical geometry -- but the mapping between physical/logical is still very much linear and 1:1 for all intents and purposes. How else do you imagine the hard drive does its stuff?

[Simon]

Tomchu, it's a purely statistical argument and has nothing to do with the actual technical implementation.

Case A: 2 TB drive, 950 GB used
Case B: 1 TB drive, the same 950 GB used

We will assume this is not just static storage (like a clone once made and then left as a read-only device), but that the user will move, delete, and copy files every one in a while.

Would you agree with me that in case B you will be forced to use the slowest parts of the drive, while in case A there is a roughly equal chance (every other bit is occupied) of using the fastest and the slowest parts of the drive? This last part obviously assumes allocation is random. This is basically the worst case. If the OS/FS are smart, they will skew those odds to achieve better performance.

Would you then also agree with me that in case A fragmentation will on average be higher than in case B?

Finally, would you also agree that assuming the same number of platters, the data density of disk A is twice that of disk B? Would you then further agree that in case B the rotational speed would have to make up for the higher data density of drive A, just to be on par?

These points are where the argument for larger drives comes from. If you end up facing the choice of a 7200 rpm drive which has just barely enough space and an equally expensive 5400 rpm drive which has much more space (and quite possibly significantly higher density), it's quite possible the same money will buy you better performance even with the "slower" drive.

[Simon]

But before we get carried away with theoretical arguments, let me point out that I also posted actual offers above as a reference. And they paint a different picture.

Compare the Seagate LP (1.5 TB, 5900 rpm) with the Segate 7200.12 (1.5 TB, 7200 rpm). There's only a $10 difference! So would I sugest the LP over the 7200 rpm drive just because of those $10. No, of course not. IMHO you buy the LP because you want a low noise/heat/power drive and that's what you end up paying for. But in terms of raw performance it doesn't stand a chance against a drive only $10 more expensive.

There's a similar situation at 1 TB. You can get Samsung Spinpoint (1TB, 7200 rpm) for $85. The cheapest 5400 rpm 1TB drive I'd buy is the $80 EcoGreen F2. Again, ask yourself, would you sacrifice performance because of $5? No, never. Again the argument for getting the slower disk is noise/heat/power, but in terms of performance the 7200 rpm Spinpoint is clearly the better choice.

Bottom line, the theoretical argument isn't wrong per se, but the economics are different (especially with 3.5" drives, notebook drives are another story). The truth is that at 1.5 TB or 1 TB the difference in cost between 7200 rpm drives and 5400 rpm drives is just simply not significant enough. If you're interested in performance you should pay the $5-$10 extra and get the faster drive at the same capacity. My 2¢.

[Tomchu]

Simon: I agree that the OS will more often have to use the slower parts of the drive. Fragmentation will also become higher due to the higher % of capacity used. At the same time, remember that the 7200 RPM drive has faster seek times and better sequential transfer rates. Data density is also usually increased by increasing track density, not sector density -- therefore the 7200 RPM drive will have better sequential transfer rates if the number of platters is the same simply due to rotational speed.

However, if the choice is a drive that just has barely enough space for what you need, then you're not looking at the right choices.

I would get the +$5/$10 more expensive drive every time, because the heat/noise output of a 7200 RPM drive is often the same or less than a 5400 RPM drive. There's simply no argument for 5400 RPM drives if they're the same size and only a few bucks cheaper.

[OreoCookie]

The faster seek time is going to do nothing to improve the situation, the throughput will be slower.

Simon and I are not making an argument that rpm doesn't matter if you consider drives of the same capacity, we're saying capacity is in this case more important than rpm. And yes, there is an argument for very large capacity 5.4k drives: they are readily available while 7.2k large capacity drives are not.

[OreoCookie]

Originally Posted by mediahound 

I really don't believe that operating systems fill up a hard drive radially like that though. Data is typically scattered all over the drive even when it's not full, right?

Usually the OS tries to write files in successive blocks so as to avoid seeks. That's what defragging is all about: you don't want your data scattered across the drive, any drive that has to seek a lot is a lot, lot, lot slower than the continuous throughput even at the slowest parts.

In any case, what you're doing is not necessarily sensitive to speed, so I'd advise you to get the larger drive. If a 7.2k version is available now and the difference is only a few bucks, then by all means, get the faster drive.

[mediahound]

Semi-related to this: if I clone a fragmented drive, will the clone also be fragmented? Or, will the clone be all nice and non-fragmented? And by clone I mean using Carbon Copy Cloner.

[Atheist]

Originally Posted by mediahound 

Semi-related to this: if I clone a fragmented drive, will the clone also be fragmented? Or, will the clone be all nice and non-fragmented? And by clone I mean using Carbon Copy Cloner.

It depends on whether you do a block copy (exact copy of the disk) or file-copy (copies one file at a time thus reorganizing the data).

[mediahound]

Originally Posted by Atheist 

It depends on whether you do a block copy (exact copy of the disk) or file-copy (copies one file at a time thus reorganizing the data).

I didn't realize CCC could do each. Which one is better? Seems like the file copy would be best? This would be for my new system drive (getting rid of the old drive).

[Atheist]

To be honest, I don't know if CCC does both. I presume it does. Block copy is faster. I don't have enough knowledge to make a reliable recommendation. Presumably OS X attempts to optimize how the drive is written to so if you do a file-copy, it may not result in an optimally arranged drive. This is purely guesswork as I really don't know all the details. Maybe you could do a block copy and then use the iDefrag tool mentioned earlier in the thread to optimize the new drive?

[OreoCookie]

You don't need CCC to clone your disk, you can do that quite comfortably with Disk Utility. CCC is really not necessary.

[Simon]

Originally Posted by mediahound 

Semi-related to this: if I clone a fragmented drive, will the clone also be fragmented? Or, will the clone be all nice and non-fragmented? And by clone I mean using Carbon Copy Cloner.

Block copy will clone fragmentation. File copy won't. That said, block copy is faster, less error-prone, and usually the better choice.

Under OS X the system will remove the worst fragmentation all by itself. The rest you get rid if by making sure you never come close to filling up your drive. When you do, you buy a new larger disk.

One last piece of advice. Get rid of CCC. It has simply caused too many people too much trouble in the past. Instead use Apple's built-in cloning tool in OS X. It's rock solid, it's very fast and it's of course free. You'll find it on every OS X installer DVD and on every OS X installation.
/Applications/Utilities/Disk Utility > Restore
Select erase destination to get a bootable clone in block-copy mode (fast!).