Understanding Hard Drive Performance

[pschmid]

We've added 15 more hard drives to our hard drive lists, and analyzed a full model family to try to understand why some models perform better than others. Which capacity point is the fastest? Does 16 MB cache offer advantages over 8 MB? We answer these questions and many more.

 

[dark_knight33]

Your first set of benchmark results with the 4 HDDs do not indicate which drives out of the four tested have NCQ or not. Since there is a grouping in many of the results which could be accounted for by this difference, I think you need to remake those charts indicating which of the drives had NCQ. Additionally you don't discuss at all the effect, if any, NCQ has over the benchmark results you published at any point in the article. :?

How can you claim to make a fair "analysis" of a drive family after neglecting this?

Edit: Specifically, when I said "the 4 HDDs" I'm talking about the results on page 6 & 7:

http://www.tomshardware.com/2007/03/05/understanding-hard-drive-performance/page6.html

 

[00101010]

It will be interesting to see what the future of hard drives will be with the Heat-Assisted Magnetic Recording (HAMR) technology. I wonder though if the added heat generation will cause higher drive failure risks. It's a concern with existing drives, and they're now planning to add more heat to future drives with this.

When these drives are commerially available, computer manufacturers will likely design their new PCs with proper airflow, but older PCs that people upgrade may not do as well without added cooling. Granted there are add-on coolers and fans people can buy, but in some situations where the PC is a small form factor, or a HTPC where people want low noise (meaning less fans) or cooling options are not available, then what? Next thing you know, hard drive manufacturers will either reduce the warranty time, add disclaimers in the packaging about airflow requirements, or change warranty rules because of many heat related failures.

Hopefully I'm wrong, but since the average person doesn't care about the inside of a PC, I hope the "heat assisted" drives will have the means to stay cool in both new and older PCs without a lot of added effort.

 

[Pihrana]

I think this is a very interesting article, and one which has got me thinking........

A great many PC enthusiasts like myself buy certain components because we know with some confidence that we'll be able to squeeze every last ounce of performance from them in the form of overclocking, or unlocking as in the case of certain ATI 9500's and Geforce 6800's for instance....

With the above in mind, is it possible to unclock this WASTED storage from hardrive platters?

In your article it becomes apparent that a great number of hardrives have wasted space (or locked space) which with the correct know how may be un-locked for usage....... I already own a 250Gb and a 320Gb 7200.10 drive and it would be great if someone knew how to unlock that wasted space.....

I certainly do not have the tallent or know how but if anyone does then please get in touch. - hehehehe

 

[hergieburbur]

I am pretty sure that couldn't be done without replacing ROM chips on the drive. In my opinion, not worth the hassle and potential bad sectors when you consider the price of storage today.

I could be wrong though...

 

[hergieburbur]

I finally actually finished the article, and I have two comments to make:

You could do a better job of organizing the contents of your graphs for readability. It doesn't make any sense to have a 320GB UATA drive as the first entry in one graph, and then have a 750 Gb as the fist entry in the SATA graph. That makes in unnecessarily hard to read and make comparisons.

You don't mention whether the tests for the SATA drives were done with NCQ enabled or not, and you don't mention the difference experienced when doing so. if NCQ was disabled for the purposes of comparing interfaces, fine, but it should be mentioned in the article (maybe I missed it).

Overall an interesting article though.

 

[4745454b]

Interesting article, to bad the "benchmarks" are a waste of time. It drives me nuts when I read an article about harddrives, only to find out that NO real world programs are used to test time. How about a time to load windows test? How about a zip/unzip test? File copy? How about differences in level load or FPS in games? Why not test with programs I use? I don't know about the rest of you guys, but I don't run IO meter that much...

Hergie, the drives are layed out differently because some are faster then others. When I read the article, I noticed that the drives were organised by avg transfer rate. This means the 320GB drive had the fastest avg transfer rate in the ATA group, while the 750GB had the fastest in the SATA group.

 

[hergieburbur]

I caught that, its just that to me it doesn't make sense as a comparison, especially when the difference in average rates was so small as to be nearly negligible.

 

[exisnet]

Nice article, thanks
Confirms I got a great deal on my 8M Raptors...

 

[vangvace]

Dark_Knight
The SATA drives have NCQ, the PATA drives do not. It's mentioned earlier in the article. As for NCQ being on or off though perhaps they shame on Tom's.

Here's a link to the 750G SATA review for comparison/divination of what should have been mentioned here. Sharky Extreme

 

[hergieburbur]

Another comment, I think THG as a whole needs to re-examine the way they name their articles. By the title, you would think this article would be an in-depth review of the different features of HD's and their effect on performance. This article is partly that, but it only explores the effect of two interfaces and capacity differences, and in a small part, access time. It could be a lot more in-depth based on the title.

I am not criticizing the article at all, but I have seen a lot of articles lately that the title would lead you to believe are about one thing, then you read the article, and they are barely related to the topic.

 

[joefriday]

Interesting article, to bad the "benchmarks" are a waste of time. It drives me nuts when I read an article about harddrives, only to find out that NO real world programs are used to test time. How about a time to load windows test? How about a zip/unzip test? File copy? How about differences in level load or FPS in games? Why not test with programs I use? I don't know about the rest of you guys, but I don't run IO meter that much...

Hergie, the drives are layed out differently because some are faster then others. When I read the article, I noticed that the drives were organised by avg transfer rate. This means the 320GB drive had the fastest avg transfer rate in the ATA group, while the 750GB had the fastest in the SATA group.

If you look closer, you'll see that they did in fact test Windows start up performance. As for the rest of your programs you want tested, there's a reason why IO meter was used as the benchmark: the difference in performance from one model to another is so small, only a Hard Drive-specific benchmark is going to tell you the difference.

 

[Red_Frog]

Sure would be nice to see some RAID tests once in a while.

I haven't used a single drive in a PC for the better part of the last decade. I'd love to see four of those 16MB cache Seagates tested in a RAID 0.

 

[hergieburbur]

Ahh! That just screams danger to me!

Of course, I come from a business consulting background where RAID 0 is almost taboo.

 

[4745454b]

If you look closer, you'll see that they did in fact test Windows start up performance.

No they tested bandwith/MBs. They didn't list how many seconds each drive took to boot.

the difference in performance from one model to another is so small, only a Hard Drive-specific benchmark is going to tell you the difference.

This is one of those exactly things. I believe for the most part this applies to Raptor/AID0 drives. Raptor/AID0 will allow faster boot and copy times, but you'd need a stopwatch to see the difference for gaming/general use.

 

[joefriday]

If you look closer, you'll see that they did in fact test Windows start up performance.

No they tested bandwith/MBs. They didn't list how many seconds each drive took to boot.



Actually, it's throughput, not bandwidth, and it has inverse correlation to load times.

 

[Luscious]

In terms of price, one high capacity drive is still less expensive than multiple smaller ones.

Uh, I'm not so sure:

Seagate 750GB $289.99 Link
Seagate 320GB $89.99 Link
3x320GB > 750GB and 3x$89.99 < $289.99

Put those three drives in RAID and you will get better performance too.

 

[ziggie]

Excellent article.........period!!!

BUT having said that, I'd much -- much Reuther have long jeopardy than gigs, controller and cache being my choice for purchase.

I'd like to see an article on which drives last longer.

 

[dark41]

I thought it was a good article, and long overdue to have the latest Seagate drives included. I just wish there was a link provided to the complete hard drive comparisions with the Seagates now included.

I've been using Seagates with NCQ for a couple years now. This is the first I've heard of enabling or disabling NCQ. Anyone care to explain how this is done?

And for the record, any business that doesn't use RAID 0 is missing out. Taboo? We've implemented RAID 0 on every business computer that we've built for the past 4 years. The only problems we've had were with Nvidia chipsets corrupting RAID arrays. The solution was to avoid Nvidia chipsets, not RAID 0. But then we don't need consultants either. ;-)

 

[hergieburbur]

I thought it was a good article, and long overdue to have the latest Seagate drives included. I just wish there was a link provided to the complete hard drive comparisions with the Seagates now included.

I've been using Seagates with NCQ for a couple years now. This is the first I've heard of enabling or disabling NCQ. Anyone care to explain how this is done?

And for the record, any business that doesn't use RAID 0 is missing out. Taboo? We've implemented RAID 0 on every business computer that we've built for the past 4 years. The only problems we've had were with Nvidia chipsets corrupting RAID arrays. The solution was to avoid Nvidia chipsets, not RAID 0. But then we don't need consultants either. ;-)

I'm certainly not getting into a debate about the relative merits of RAID 0 in a business environment, other than to say I ALWAYS recommend 1+0 or 5 over 0.

Your not hurting my feelings, I'm not that kind of consultant anymore.

I think some controllers provide the option to disable NCQ. I am pretty sure I've seen that on one of the Promise controllers I used to support.

 

[dark41]

It wasn't my intent to hurt anyone's feelings. I believe most people would agree with me that RAID 5, 0+1, and 10 are all versions of RAID 0. They just include backup (RAID 1) as well, in some form. We also recommend RAID 5 or 10 for what its worth.

I don't use Promise controllers much. I assumed you knew of a trick in Windows or via the BIOS to disable NCQ that I wasn't aware of. I don't see any options to disable NCQ on Intel nor Gigabyte boards (our weapons of choice).

Anyone wanting to see actual startup times for Windows or other real time comparisons can look at the TG hard drive charts. A link should have been provided with the article.

http://www23.tomshardware.com/storage.html

Bang for the buck, the 750GB Seagate 7200.10 is pretty impressive.

 

[hergieburbur]

The 320GB model is pretty nice as well.

You could technically say that ALL RAID forms are just different derivations of 1 and/or 0.

 

[CLLO]

I found the article very informative.

It confirms my thought that sATA and pATA doesn't make much difference given that the either interface is a good bit faster then what the platters can deliver.

It surprise me that at high queue depth, invariably, the pATA out performs the eATA. A longer queue should give NCQ more opportunity to make a positive difference, but instead, there is a clear cross over point and then difference drops to negative.

Entirely logical, but counter intuitive. At certain queue size, the amount of time it takes to play with the queue exceeded the time of merely waiting for the darn thing to come under the read/write head.

I am happier with my pATA now, at least until their on-disk logic get faster.

 

[rvjc]

This was a very useful article but it did not elaborate on a number of critical performance-related issues.

With LBA, bad sector reallocation and other HDD controller smarts in play, is it possible anymore to make assumptions about contiguous storage areas and physical CHS head position.

For example, I would like my swap file/partition to be PHYSICALLY contiguous. How can I guarantee this?

Traditionally, swap files were located as close as posible to track 0 in order to minimise access times. This was based on proximity to the head's rest position. Does this rule still apply? How can I be sure that I am close to physical (as distinct to logical) track 0? Does the head return to a rest position during quiescent periods or does it wait at an optimal position based on historical accesses?