multiple hard drives, not for server but for RAID speed

Silencing hard drives, optical drives and other storage devices

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

multiple hard drives, not for server but for RAID speed

Post by photonblaster » Wed Jan 26, 2011 1:34 pm

The recent server silent build articles show that you can have multiple hard drives and still keep a system relatively quiet.

I want to go to multiple hard drives not to get more storage capacity but for Raid 5 to get more I/O speed. Use enough drives and you can get SSD speed with a lot more storage for the same price. Three top end 3.5" hard drives in RAID 5 is about the same as a single SSD as far as i can tell, maybe a little slower, unless you want to spend $500+ for a high end Crucial RealSSD C300 Review (256GB) (http://www.storagereview.com/best_drives.)

The problem of course is the sound level penalty, which I want to minimize.

2.5" drives are slower than 3.5", so for the same speed you would need more drives. But they are attractive because they can be suspended in 3.5" cage with the rubber string mounting technique. And probably lower total power and less heating issues per drive.

AND I suspect they be lower noise in total for the same speed (ie maybe 4-5 of 2.5" drives may be quieter than three 3.5" drives). The fact that hard drive noise does not appear to add in classical manor (6 HDD are not 3dBA louder than 3 in a system) also suggests going to RAID 5 may not be such a bad idea. (http://www.silentpcreview.com/Silent_Ho ... st_Results)

Mike said back in 2004 "my recent discovery of extremely quiet notebook hard drives that easily put the quietest of 3.5" desktop drives to shame" (http://www.silentpcreview.com/article145-page1.html), but the SPL numbers in the recommended hard drives list says otherwise ( http://www.silentpcreview.com/article29-page2.html)


So what should I beileve about sound levels 2.5" vs 3.5" drives? Anyone have some experience or insights that can help guide me in my decision?
Last edited by photonblaster on Wed Jan 26, 2011 5:26 pm, edited 1 time in total.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Wed Jan 26, 2011 3:44 pm

It's not that simple. We all wish it were but it isn't.
There are different performance characteristics you can look at.
In some respects, even a single hard drive is still better than most SSDs in practice. But in other respects no amount of money thrown at RAID arrays will come close to the performance of a cheap SSD.
The impact of going for 2.5'' drives is different depending on the performance characteristic you're looking at.
RAID5 has a performance impact. It's not all positive. RAID0 (or RAID 10) is better depending on what kind of performance you want.
If you want the best performance, you're probably looking at a hybrid solution mixing spinning and solid-state drives. Hardware is moving in that direction but you can do something similar without spending big bucks on RAID cards or waiting for cheaper hybrid hardware. ZFS for instance can use SSDs for cache without special hardware (or so I've been led to believe).
And last but not least, judicious use of RAM can do miracles.
So the question is, what kind of performance do you need? What problem are you trying to solve exactly?

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Wed Jan 26, 2011 5:24 pm

HFat wrote:It's not that simple. We all wish it were but it isn't.
There are different performance characteristics you can look at.
In some respects, even a single hard drive is still better than most SSDs in practice. But in other respects no amount of money thrown at RAID arrays will come close to the performance of a cheap SSD.
The impact of going for 2.5'' drives is different depending on the performance characteristic you're looking at.
RAID5 has a performance impact. It's not all positive. RAID0 (or RAID 10) is better depending on what kind of performance you want.
If you want the best performance, you're probably looking at a hybrid solution mixing spinning and solid-state drives. Hardware is moving in that direction but you can do something similar without spending big bucks on RAID cards or waiting for cheaper hybrid hardware. ZFS for instance can use SSDs for cache without special hardware (or so I've been led to believe).
And last but not least, judicious use of RAM can do miracles.
So the question is, what kind of performance do you need? What problem are you trying to solve exactly?
Thanks for the reply. I guess I should not have given some of the background that lead up to my question since it appears to have detracted from the question I need help with. I do not want to get into a discussion of merits of different RAID systems. Nor SSD vs HDD vs Hybrid, RAID or not, etc. I edited my first post to make this more clear.

I am looking for help about noise levels, 2.5" vs 3.5" HDD. I have found conflicting information as described in my post and would like to hear some real world experience, or better yet, links to side by side, scientifically sound, sound level measurements.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Thu Jan 27, 2011 11:42 am

I don't think the matter is up for discussion: your premise (getting "SSD performance" from a RAID array) is irredeemably flawed. You need to understand that.

Now, as to the noise of 2.5'' vs. 3.5'' drives...
I don't know a better source than SPCR for measurements. Sure, SPCR only measures one drive which could be a problem (sample variance) but SPCR has recordings and frequency measurments, not only a SPL value.

I'm surprised by the table as well. I don't have the quietest 3.5'' models to compare (like I've got single-platter F3EGs and a dual-platter F2EG but no single-platter F2EG) but, subjectively, my 2.5'' drives (again, not the same models: I have single-platter 5K.500Bs for isntance) are quieter than my 3.5'' drives when I hold both in my hand.
I know that humans make for really poor SPL meters so I don't doubt SPCR's SPL values. They don't matter very much anyway. When you get that low the noise is easily dampened under the noise-floor unless you're installing the PC in some kind of sound-proof studio.
What matters is the vibration and the seek noise which can be amplified by the case. The "whoosh" can be amplified a bit too but, with fairly quiet 3.5'' drives (but without the acoustic management thing), it's the seeks which are annoying in my opinion. And I feel there's a dramatic difference between 2.5'' and 3.5'' in that respect: hard-mounting 2.5'' drives is OK as far as I'm concerned because they have little vibrational and seek energy. It's obvious when you hold them in your hand. If I put my F2EG hard-mounted in a case on my desk, it makes quite a racket. The seeks in particular seem to make the case and the desk resonate. I can barely hear a quiet 2.5'' drive in the same situation. So I wonder if SPCR's tables use the same vibration scale for 2.5'' and 3.5'' drives. But maybe I misunderstand what "vibration" is supposed to refer to.

So if for some reason a RAID of quiet 2.5'' drives (an odd choice!) satisfied a need, I'd definitely go for 2.5''. But understand there's a tradeoff!
If you can suspend or otherwise soft-mount 3.5'' drives and you're careful with the model choice, I expect your array could be very quiet as well. But I expect it'd involve more work as compared to 2.5'' drives: your case would have to be heavier, to have more dampening material, less airflow and so on. I've never put much work into quieting RAID arrays myself (I put them in servers which aren't standing on anyone's desk!) so this last paragraph is purely speculative on my part.

Fire-Flare
Posts: 422
Joined: Sun Jun 13, 2010 12:44 pm
Location: Seattle

Re: multiple hard drives, not for server but for RAID speed

Post by Fire-Flare » Thu Jan 27, 2011 12:21 pm

HFat wrote:So if for some reason a RAID of quiet 2.5'' drives (an odd choice!) satisfied a need, I'd definitely go for 2.5''.
Luckily for Photon Blaster I'm an odd person!

I used to have a Samsung Spinpoint 3 but traded up to a pair of Western Digital AV-25 drives. (Which aren't reviewed here, sadly)

Unlike the Samsung, their 'whoosh' and seek noises are almost undetectable with your ear next to them and Icy Dock's dual 2.5" to 3.5" adapter (the passive black one) absorbs all the vibrations when mounted with rubber grommets.

The AV's spin at 1800 fewer RPMs, but with two in RAID 0 my Windows 7 Performance Index was identical to the single Samsung.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Thu Jan 27, 2011 12:39 pm

Your choice (RAID0 with 2 drives) makes a lot more sense than RAID5 if you ask me.
The Windows thing is probably understating the performance of your mini-array.

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Thu Jan 27, 2011 12:56 pm

HFat wrote:Your choice (RAID0 with 2 drives) makes a lot more sense than RAID5 if you ask me.
Curious as to why you would say that? 3 drives in RAID5 give almost the same performance level as 2 in RAID0, but with drive failure reduncancy. With the new high performance 2.5" dirves, $50-$70 each, you can get saturated SATAII performance, or close to it, at least in the outer edges of the platters, with four drives.

But, at the sacrifice of more noise than a single drive. I am evaluating that trade off.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Thu Jan 27, 2011 1:29 pm

You're focusing on a metric at which RAID5 is pretty good but which is probably not the most relevant. Spinning drives have terrible random access performance and so does RAID5 when you're writing. While two $40 drives in RAID0 are hard to beat for the money, for the cost of three $50-$70 drives you could get a decent SSD coupled with a quiet spinning drive. You could get good random access performance on that (depending on your application)... and of course very low noise.
Plus RAID5 is trouble. In a server, you'd want the redundancy and it might be worth the trouble. But in a PC? If you can mount three drives you can probably mount four and get two separate RAID0 arrays. You could then mirror one to the other and your data would be much safer than with RAID. It obviously depends on the rest of your environment: if you already have decent mirroring going on, RAID5 makes more sense if you want the extra safety but don't want to pay more to get RAID10.

I doubt you'd get much more noise than a single drive as long as there's no freaky interactions between the vibrations or seeks of the drives (if you soft-mount you should be safe). I'm no sound engineer but I thought noise sources had to be in phase or something to add up as if you doubled the power. 3db louder doesn't sound twice as loud in any case.

Fire-Flare
Posts: 422
Joined: Sun Jun 13, 2010 12:44 pm
Location: Seattle

Re: multiple hard drives, not for server but for RAID speed

Post by Fire-Flare » Thu Jan 27, 2011 1:33 pm

photonblaster wrote:
HFat wrote:Your choice (RAID0 with 2 drives) makes a lot more sense than RAID5 if you ask me.
Curious as to why you would say that? 3 drives in RAID5 give almost the same performance level as 2 in RAID0, but with drive failure reduncancy. With the new high performance 2.5" dirves, $50-$70 each, you can get saturated SATAII performance, or close to it, at least in the outer edges of the platters, with four drives.

But, at the sacrifice of more noise than a single drive. I am evaluating that trade off.
I'm not worried about drive failure, I make weekly backups.

2.5" drives give the additional bonus of allowing you to install more drives in a given case than 3.5". There are adapters that let you fit 4 in a 5.25" bay.

They can't match the capacity of a 3.5" but they can give better performance.

inti
Posts: 118
Joined: Thu Feb 10, 2005 7:09 am
Location: here

Re: multiple hard drives, not for server but for RAID speed

Post by inti » Sat Jul 09, 2011 11:16 am

RAID is very noisy because any disk action requires simultaneous seeks from all drives. So your +3dB power law will hold true for RAID.

To explain, in a normal multi-drive system, all drives are spinning but typically only one drive is seeking at a time. If you do a complex drive-to-drive operation involving seeking, like a video edit or something, maybe you will hear two drives seeking at the same time which gets fairly noisy, but even then the seeks are not necessarily synchronized.

In a RAID system, every drive access, even the smallest read (unless it is out of cache) will cause all drives to seek simultaneously. Because it is simultaneous, the sound reverberates through the case ('chunking'). The seek noise becomes the loudest sound in your system. At least, when I tried it I found it unbearable even with just two drives in a RAID0.

Sure, you can reduce the seek noise issue by using suspended 2.5" drives and selecting drives with the quietest seek. But, that seems counter-productive if you are going to build an array of three or four of them, you will be back up to a loud seek noise.

Alternative solutions:
* RAID0 of cheapest (GB/$) of SSDs
* NAS in a different room - no speed issues if it's on a wired network
* Hybrid 2.5" drive, like Seagate Momentus XT 500GB - this is good value and still gives you the speed of SSD for commonly used apps
* A lot of RAM and good caching software like SuperCache
* RAID with 3.5" drives in a fully silenced external case such as the one from A-tech Fabrication (very costly)
http://www.atechfabrication.com/product ... e_2500.htm

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Tue Jul 12, 2011 9:00 am

Since I did the first post on this thread, SSD performance has skyrocketed, and SATA 6Gb/s SSDs are so fast now that a SATA 3Gb/s interface slows them down. Except for cost, the newest SSDs are the way to go if you want speed. And PCI RAID SSD configurations are insanely fast it seems.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Tue Jul 12, 2011 9:16 am

At the low end, you still get better sequential writing performance for the money with spinning drives... and not by a small margin.

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Tue Jul 12, 2011 10:12 am

HFat wrote:At the low end, you still get better sequential writing performance for the money with spinning drives... and not by a small margin.
I am not sure what your comment is supposed to convey. Do you mean "low end", ie old cheapest SSDs are not as good as a spinning drive?

I just went to storage review, leaders board, looked at a few sequential read/write graphs side by side and it would appear the best SSDs are many times faster than the best spinning drive.

As I mentioned below, EXCEPT FOR COST, SSDs seem to be the way to go for speed, tough to justify a RAID configuration for spinners to get close. I think SPCR is tending towards this as they often look at systems with one SSD for boot speed and one huge spinner for storage capacity. And "inti" seemed to make a good point that seek noise may be an issue with RAID even with very quiet 2.5" spinners.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Tue Jul 12, 2011 10:27 am

photonblaster wrote:Do you mean "low end", ie old cheapest SSDs are not as good as a spinning drive?
The problem has not fundamentally changed. Some old SSDs are pretty good for the money actually.
The newest SSDs are still not as fast at writing sequentially as a pair of hard drives costing less than half as much. That's what I mean.

Storage Review's numbers are useless.

Abula
Posts: 3620
Joined: Tue Nov 02, 2004 12:22 pm
Location: Guatemala

Re: multiple hard drives, not for server but for RAID speed

Post by Abula » Tue Jul 12, 2011 10:50 am

Personally i wouldn't try a array of 2.5 drives, out of the following reasons
1) 2.5 Drives are not design for arrays
2) Mechanical drives will do more noise the more you have
3) Size and speeds lower than 3.5 drives.
4) $/gb is more exepensive on 2.5 than 3.5

Mechanical drives market is dying, slowly but will die eventually, they will have to keep up with size as its the only thing that will keep them afloat for some time. SSDs limitation atm is price and capacity, in time with better tech and better manufacturing they will lower it, but probably will be a lot of time, maybe 5 to 10 years.

Now it can be done though, i mean synology has a server that uses 2.5 drives in a raid array, Synology America Corp DiskStation 4-Bay USB 2.0 Network Attached Storage DS411slim Black. But to me in a server isnt that great, i can place the server anywhere around the network if drives are too loud, but green drives are fine to me, atm i have my server in the same room but the activity on it its low, spends like 90% of its time at idle.

To me the best solution currently is,
Desktop ---> Small SSD OS/Apps + 2.5 5400rpm laptop drive for storage + Home server for main storage and backups

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Tue Jul 12, 2011 11:20 am

Lots of people are using 2.5'' in arrays because you get higher capacity and performance in the same volume. 3.5'' are better at capacity per dollar, and often at performance per dollar too.

Hard drives will be used a servers and workstations long after they disappear from laptops and desktops.

awolfe63
Posts: 29
Joined: Sat Oct 27, 2007 3:25 pm

Re: multiple hard drives, not for server but for RAID speed

Post by awolfe63 » Tue Jul 12, 2011 12:35 pm

I use an 8-drive 2.5" RAID-6 NAS all the time. Quiet, low power, and works great. (QNAP SS-839)

I wouldn't hesitate to try building your own 2.5" RAID array.

jhhoffma
Posts: 2131
Joined: Mon Apr 25, 2005 10:00 am
Location: Grand Rapids, MI

Re: multiple hard drives, not for server but for RAID speed

Post by jhhoffma » Fri Jul 15, 2011 9:06 am

HFat wrote:
photonblaster wrote:Do you mean "low end", ie old cheapest SSDs are not as good as a spinning drive?
The problem has not fundamentally changed. Some old SSDs are pretty good for the money actually.
The newest SSDs are still not as fast at writing sequentially as a pair of hard drives costing less than half as much. That's what I mean.

Storage Review's numbers are useless.
HFat,

If you don't mind, it might be a good idea to post some reference or metrics to what you are saying so that it's a little clearer what the benefits/failings of SSD vs RAID might be.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Fri Jul 15, 2011 9:40 am

The benefits of both are well-known.
It's silly to try to determine which is better than the other. The optimal solution if you have the resources is of course to use both! Hybrid solutions are not better in all respects but they're better all-around if you don't want to take the trouble of setting up your software optimally.

SSD: much, much better random access performance and pretty good sequential read performance too

RAID: much lower prices, especially for capacity but also to get the same level of sequential write performance (probably not at all levels however) or reliability

And one shouldn't forget the option to use RAM (either dedicated cards loaded with RAM or simply system RAM) when comparing because performance is better than SSDs and so it sometimes makes sense to pair loads of RAM with RAID without bothering with SSDs.

Specific figures are all over the web but the details depend on the implementations are lots of numbers are useless. For instance Sandforce has compression so you shoudn't use benchmark with non-random data unless you want numbers about their compression.
If you have a specific concern, I'll see if I can provide something but I don't want to write an article on the general topic!

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Wed Jul 20, 2011 4:23 pm

HFat wrote: Storage Review's numbers are useless.
Why do you say such a extreme statement? They have probably a couple dozen metrics for each drive review, you can pick what metric is important to you.
What review site is better?

You also said that "The newest SSDs are still not as fast at writing sequentially as a pair of hard drives costing less than half as much." Certainly I agree that two of spinning drives usually cost much less than a single high end SSD, and $/gb is blown away by two HD vs any SDD, but could you please show me some test that, for instance, shows any two HD beat a single OCZ Vertex 3 in sequential writes?

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Thu Jul 21, 2011 12:17 am

I was being generous actually. These numbers are evidently worse than useless since they led you into believing that the Vertex3 has particularly good sequential write performance. Misleading data with no warnings is worse than no data at all. You can actually find single hard drives which are faster than the smaller Vertex3 in that respect. Common hard drives won't be but a pair can easily beat even the larger Vertex3. This shouldn't be surprising. You can learn about the performance of hard drives on lmgtfy.com. Then see Anandtech for background, tests and benchmarks about the actual (see above) performance of Sandforce drives.

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Thu Jul 21, 2011 4:15 pm

HFat wrote:I was being generous actually. These numbers are evidently worse than useless since they led you into believing that the Vertex3 has particularly good sequential write performance. Misleading data with no warnings is worse than no data at all. You can actually find single hard drives which are faster than the smaller Vertex3 in that respect. Common hard drives won't be but a pair can easily beat even the larger Vertex3. This shouldn't be surprising. You can learn about the performance of hard drives on lmgtfy.com. Then see Anandtech for background, tests and benchmarks about the actual (see above) performance of Sandforce drives.
OK, I guess you are saying that Storage Review uses compressible data while Anandtech does not: Quote AT:AS-SSD Incompressible Sequential Performance
The AS-SSD sequential benchmark uses incompressible data for all of its transfers. The result is a pretty big reduction in sequential write speed on SandForce based controllers.
http://www.anandtech.com/show/4341/ocz- ... reviewed/7

So. ya, if you have incompressible data you do not benefit (again a guess since unlike apparently the many to whom the "benefits are well known", I do not know for sure, just using logic) if the controller does not strip out "empty space" in the file it slows the write down because it has to write most of the original file. But on the other hand if you do have compressible files, the SR stats should be OK. The why is not really important though since to me the end result is what is what's critical.

But lets forget that for now, and assume that no one in their right mind would consider using anything but incompressible file writes to characterize the performance of SSDs. The best SSDs seem to be in the range of 300MB/sec according to AT if I read their graphs correctly, and the Vertex 3 is 284.

Now look at HDD, fastest I saw on AT, the velociraptor. It shows 144MB/sec sequential writes. Assume best case scenario you can double this in RAID, and you do not beat by any margin the best SSD on uncompressible files sequential writes and barely beat the Vertex 3 unlike your statement to the contrary. You can add more drives to the RAID to get it faster at the expense of added noise and higher data risk. It also means that you need 3 velociraptors to get better than the best SSD, and then the $/GB AND total GBs are similar, as I just priced on Newegg.

3, possibly 4 of 7Krpm HDD in RAID possibly could keep up with the best single SSD, at a lower price and with more storage. But again at risk of storage failure much more frequent than a SSD. But this is more than a pair which you claim is faster.

Add to this that most files are probably at least a little compressible, and the fact that you do not double write rates with two drives, then two HDD NEVER beat a single large Vertex 3 SSD.

It would seem clear to me that HDD RAID is now really only of major benifit if you want tons of capacity. When I started this thread, this was not true, but since then SSD throughput has doubled.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Fri Jul 22, 2011 11:03 am

Your dishonesty is appalling. Why do you bother?
The Velociraptor you're talking scored 144MB/s in IOmeter. The Vertex 3 you're talking about (the one that costs about $500, the affordable doesn't perform like that) scored 225MB/s in IOmeter. You had to compare different benchmarks to get the result you want.
Not only that. Velociraptors have good access times compared to other consumer drives but not particularly good sequential transfer speeds. The one you're talking about is old and has a lower throughput than cheaper drives with 660G platters which were introduced long before the Vertex 3. Platters with even higher density (and therefore throughput) are available now. But you have the chutzpah to claim that no drive can possibly beat the one you selected!
Since you're talking about such an expensive SSD, let me amend by statement above: you can beat its sequential write performance with two hard drives that are older than that Vertex and costing together less than a fifth of its price, not merely less than half.

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Fri Jul 22, 2011 6:01 pm

HFat wrote:Your dishonesty is appalling. Why do you bother?
The Velociraptor you're talking scored 144MB/s in IOmeter. The Vertex 3 you're talking about (the one that costs about $500, the affordable doesn't perform like that) scored 225MB/s in IOmeter. You had to compare different benchmarks to get the result you want.
Not only that. Velociraptors have good access times compared to other consumer drives but not particularly good sequential transfer speeds. The one you're talking about is old and has a lower throughput than cheaper drives with 660G platters which were introduced long before the Vertex 3. Platters with even higher density (and therefore throughput) are available now. But you have the chutzpah to claim that no drive can possibly beat the one you selected!
Since you're talking about such an expensive SSD, let me amend by statement above: you can beat its sequential write performance with two hard drives that are older than that Vertex and costing together less than a fifth of its price, not merely less than half.
If I was being dishonest, I would have used a chart on AT that had the lowest possible write rate for the velociraptor. I used the only available sequential write test I found from the most recent HDD drive review at AT...http://www.anandtech.com/show/3981/west ... reviewed/4. Note that is does not discuss whether or not it is un-compressable or not. Sorry if that does not meet your criteria for comparison, but I am doing the best I can to give hard data under the rules you give.

I would think the comparison I used is the exact opposite to being dishonest, since it places the SSD in poor light by choosing a slow, if not its slowest write metric. Oh well, let's use IOmeter only then, which you say shows 144 and 225 respectivley. Go to http://www.anandtech.com/bench/Product/182?vs=352 and AT says, yes, the velociraptor is 144, but says the Vertex 3 newest drive is not what you claim, but is 366, which is 2.5 times faster on this metric.

Honestly, I really have tried to find something to support your claim that two regular HDD in RAID are faster than the newest SSDs on AT, little alone the fastest avaialbe HDD. Being an honest person, I would not have made my post to you if I had not tried.


Will you kindly, to quote jhhoffma
jhhoffma wrote:"If you don't mind, it might be a good idea to post some reference or metrics to what you are saying so that it's a little clearer what the benefits/failings of SSD vs RAID might be."
To paraphrase, please give us data rather than rhetoric and link to that data.

HFat
Posts: 1753
Joined: Thu Jul 03, 2008 4:27 am
Location: Switzerland

Re: multiple hard drives, not for server but for RAID speed

Post by HFat » Fri Jul 22, 2011 9:57 pm

I won't bee goaded into replacing lmgtfy.com.

But since I've made a mistake, let me correct it: it's 272 and not 225, an unaccountably low value people would be right to suspect. I should have been more careful. But that doesn't change the conclusion.
366 is the approximate speed at which zeroes can be compressed by the controller. Slower drives with the same controller get the same speed if you feed them zeroes.

multiplexer
Posts: 41
Joined: Wed Jul 13, 2011 4:11 am

Re: multiple hard drives, not for server but for RAID speed

Post by multiplexer » Sat Jul 23, 2011 1:17 am

Now you are just trying to prove each other wrong instead of making a constructive argument.

SSDs are vastly superior to HDDs in every practical scenario. There should be no question about this. There is only one scenario where HDDs perform better, and that is sequential write. That's it. The amount of sequential writes in practical situations is minimal and, especially, the absolute time savings you get with higher sequential writes are well compensated by for instance the sequential read speeds on even the slowest SSDs. I have Intel X25-Vs in all my computers, and guess what, even those $50 second hand drives do 190MB/s sequential read and 20+ MB/s 4-512k read. They do this all with higher recording density and much lower power requirements than regular hard drives.

The big issue with SSDs is price per GB, and that is more or less the only reason why HDDs are still interesting to have. No amount of HDDs, even tens of HDDs in RAID, will get close to the performance of just one modern day SSD. If you are planning to get a RAID array of HDDs just for performance, well, don't. Get an SSD. However, if you need the storage space there is no alternative. Well, you can shell out $10.000 or so for a pair of 2TB SSDs. However, RAIDing multiple hard drives just for performance won't cut it even then. In order to get a tangible, lasting performance advantage over JBOD you need to for instance consider a ZFS array with L2ARC SSD (or lots of RAM as L2ARC).

However, both RAIDing and ZFS are kind of a hassle (at least as of now). I consider plain RAID for performance too much of a hassle for too little of an advantage. ZFS is worth it, but you'll need to invest some time into it. In any other case, just go JBOD + SSD and put whatever needs quick access on the SSD, reserve the HDDs for stuff that you will most likely sequentially access (video media, for instance).

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Sat Jul 23, 2011 9:25 am

multiplexer wrote:.....SSDs are vastly superior to HDDs in every practical scenario. ..... There is only one scenario where HDDs perform better, and that is sequential write. ........
But multiplexer, havn't the newest SSD drives surpassed HDD even in this one scenario?

photonblaster
Posts: 59
Joined: Sun May 17, 2009 9:26 am
Location: North America

Re: multiple hard drives, not for server but for RAID speed

Post by photonblaster » Sat Jul 23, 2011 9:37 am

multiplexer wrote:Now you are just trying to prove each other wrong instead of making a constructive argument.
Actually, multiplexer, Hfat never tried to prove anything, just gave (arrogant) opinion with posts with innapropriate hyperbole and numbers not backed up. If it happens on my thread, I like to take such posters to task a little, certainly a "tilting at windmills" activity but has some amusement value.

Thanks for you post with some good overall insights into the SSD/HDD situation.

multiplexer
Posts: 41
Joined: Wed Jul 13, 2011 4:11 am

Re: multiple hard drives, not for server but for RAID speed

Post by multiplexer » Sat Jul 23, 2011 11:57 am

Let's look at it from another perspective with respect to your sequential write comment. Even the best SSDs are still in the same order of magnitude where HDDs live. Certainly if we're talking cost/performance. As a counterexample take random I/O: even the best HDDs don't manage better than 300kB/s random I/O performance (block-size transfers) while the cheapest SSDs get almost two orders of magnitude better performance. That's a race HDDs can never win.

In sequential reads and writes it's a different story. Yes, you can buy a Vertex 3 drive and get roughly the write and read performance of three hard drives, but for that you need a $300 SSD compared to three $65 HDDs. HDDs still win in some regard, and even if you're not willing to buy into that performance metric, there's still the unproven reliability of consumer SSDs. To alleviate that, you need to buy into more expensive enterprise hardware which skews the advantage towards HDDs still more.

Post Reply