RAID-0 proved ineffective at boosting desktop application/game performance

[Lazn_Work]

short and sweet:

If you are doing several things at once that access the drives, separate spindles with the separate threads accessing the separate drives will see the most benefit from non RAID 0 drives.

If you are doing large reads/writes from a single thread, or threads that must access the same drive, then RAID 0 shows the most benefit.

And for general everyday use, do whatever want for yourself: Some like it hot, some like it cold, some like it RAIDed

For myself, RAID 10 is my fave

 

[drizzt81]

I think we should all just skip the whole e-penis comparison and skip to the information? Most of us do not know each other, hence statements aimed at discrediting other people's "analysis" through attacks on their person are a shot in the dark.

Rather, I'll just say that every single one of my customers that I've moved from single disk to raid-0 for their OS and application needs has expressed how their machine is more responsive after. I could care less what other people think about this subject, so I won't argue the point further.

Have you considered a single/double blind study to try and account for any placebo effect? Given that people are unwilling to perform quantitative tests, I think at least we should try and maximize the accuracy of the qualitative information.

I'll agree that in certain situations a raid-0 would be slower than a single disk... but not many.

Is it really the number of situations that is the driver for performance? Or rather how frequently they occur, the relative time spent on these operations as part of total I/O time or some combination thereof?

 

[kumquat]

Every person who buys $1500 speaker cables expresses how much better their theater system sounds, too.

 

[mikeblas]

If you are doing several things at once that access the drives, separate spindles with the separate threads accessing the separate drives will see the most benefit from non RAID 0 drives.

What is "non RAID 0 drives"? RAID1? JBOD? RAID10? RAID5?

 

[Brahmzy]

Every person who buys $1500 speaker cables expresses how much better their theater system sounds, too.

Pretty lame comparison dude.

 

[Blazestorm]

I plan to run Raid0 again mainly for transferring stuff to and from my gigabit wired server... It has an 8 x 750 Raid5 on it with 300MB/s sequential reads (I think the controller is capped about that). but I can only get 50-60MB/s because my HDD is too slow D=

I think I'll be getting a couple 640 AAKS...

 

[unhappy_mage]

Pretty lame comparison dude.

In some ways, yes. Audio is a fairly subjective topic, and it's hard to make quantitative measurements of whether a particular set of speakers or headphones are "good" (or even "accurate", for that matter). But a lot of the things that audio enthusiasts say also apply to RAID enthusiasts---you have to experience it for yourself, once you switch there's no going back, it completely blew me away---without doing any objective blind testing to see which is which.

With hard drives I think there should be more room for science: how long does it take for this piece of machinery to perform that set of tasks? A lot of the tests that are run in typical test beds lack either precision or relevance to the real consumer's systems so that they are repeatable. Run benchmarks, or do double-blind testing (you'll need another person to help you with this, methinks) trying to decide which of two otherwise identical machines has a raid 0 array and which has a single disk.

On an aside, my roommate swore he could hear a difference in quality between two digital media players playing the same song. One statistically valid double-blind test later, he changed his mind. He hasn't brought up raid 0 yet

 

[kumquat]

Pretty lame comparison dude.

No, it's not.

The fact that customers "express" that they're impressed with the performance of an "upgrade" they paid for means absolutely nothing without quantification.

 

[MentallyNormal]

RAID-0 is like that hot chick back in highschool
You wanted her,but she was just too popular.
You where too afraid to ask her out
Now she is a porn star
DAMN you RAID-0!
Well at least you got ole reliable miss rosy palms.

 

[Lazn_Work]

What is "non RAID 0 drives"? RAID1? JBOD? RAID10? RAID5?

Oh quick nit picking.. each thread accessing separate spindles is better than both threads accessing the same one.. you could even have two RAID 0 arrays if you wanted to accomplish this.

 

[xxEIEIOxx]

You are welcome to say what you want, but the simple fact remains that no RAID setup will deliver improved desktop performance. Power users looking to increase storage performance should look to faster single drives or adding indendent spindle(s) to service additional loads.

RAID-0's shortcomings are well documented:

A) Reduced reliability
B) Increased heat/power draw/noise
C) Increased system complexity
D) Greatly complicated backup and disaster recovery
E) Substantially increased cost

However, many have tried to justify/overlook those shortcomings by simply saying "It's faster." Anyone who does this is wrong, wasting their money, and buying into hype. Nothing more.

A)Reduced reliability is untrue, 7 years and no issues. A drive does not become less reliable if you have 2. Nobody makes this argument for SLI.
B)The heat/power draw/noise from RAID 0 is irrelevant, this is an enthusiast setup, a lot of these guys will be running SLI, which contributes a LOT more of those things.
C) See B
D)Doesn't complicate backup and recovery for me in the least.
E)Oooh nooo, an extra $80, and I get the benefit of using to full capacity of both drives.

The people that quote articles like these have never tried it. It is "faster", it doesn't cost any more than the same amount of storage non-RAID, and doesn't have to be justified to anyone else. No, it's not faster for situations where the hard drive is not accessed anyway. If you want to run it, go ahead, and don't let anyone else talk you out of it. If you don't, then don't worry about those of us that do. We'll be fine. Thanks for your concern.

 

[kumquat]

A)Reduced reliability is untrue, 7 years and no issues. A drive does not become less reliable if you have 2. Nobody makes this argument for SLI.

Sorry. It is true. It's a simple fact.

Your one anecdote doesn't negate the mathematical truth.

 

[unhappy_mage]

A)Reduced reliability is untrue, 7 years and no issues. A drive does not become less reliable if you have 2. Nobody makes this argument for SLI.

The fact that you've had no issues doesn't make it as reliable as a single drive. SLI doesn't involve things that have moving parts as hard drives do. The number of hard drives sold in a year completely dwarfs the number any individual (or in most cases, company) buys, so your results are simply not statistically relevant. You need only read a few threads about raid 0 to see counterexamples.

B)The heat/power draw/noise from RAID 0 is irrelevant, this is an enthusiast setup, a lot of these guys will be running SLI, which contributes a LOT more of those things.

Irrelevant or not, it's still there. You can ignore it if you like, but you can't argue it doesn't happen, and other people care about those things, enthusiast or not.

C) See B

See B

D)Doesn't complicate backup and recovery for me in the least.

Many programs (especially older ones) have problems with backing up RAID arrays.

E)Oooh nooo, an extra $80, and I get the benefit of using to full capacity of both drives.

Or an extra $300 for another Raptor. And another $300 for an offboard RAID controller if you want one of those.

The people that quote articles like these have never tried it. It is "faster", it doesn't cost any more than the same amount of storage non-RAID, and doesn't have to be justified to anyone else. No, it's not faster for situations where the hard drive is not accessed anyway. If you want to run it, go ahead, and don't let anyone else talk you out of it. If you don't, then don't worry about those of us that do. We'll be fine. Thanks for your concern.

I have tried it. It was faster in synthetic benchmarks, but I didn't see any advantage out of it other than that; real-world usage didn't speed up any that I noticed. I don't try to talk people out of it, but I do try to point out the advantages and disadvantages when people state without discussion that they're planning to use raid 0. There are some things to think about when considering any new piece of hardware, and RAID 0 has issues that are worth considering before charging headlong into it.

 

[xxEIEIOxx]

1) The fact that you've had no issues doesn't make it as reliable as a single drive. SLI doesn't involve things that have moving parts as hard drives do. The number of hard drives sold in a year completely dwarfs the number any individual (or in most cases, company) buys, so your results are simply not statistically relevant. You need only read a few threads about raid 0 to see counterexamples.

2) Or an extra $300 for another Raptor. And another $300 for an offboard RAID controller if you want one of those.

3) There are some things to think about when considering any new piece of hardware, and RAID 0 has issues that are worth considering before charging headlong into it.

My opinion. on these 3.
1) Everyone predicts absolute doom if you run RAID 0. If you have 1 drive, and it would have been the one that failed in the array, you are no better off. You don't know before hand what drive it will be.

2) The expense is your choice. If you want to buy 2 $300 raptors, your option. You get the full capacity of both, so nothing is wasted. As for a RAID card, you'd have to really be into the idea to spend that money, as most mobos have integrated RAID at this point. If you are willing to go that route, nobody is going to talk you out of the cost.

3) I agree with this point, there are considerations to be made, and it shouldn't be approached with complete ignorance.

Again, I don't see the same negativity on SLI, which costs much more, and may also give minimal returns. I can find benchmarks that say SLI is no faster and even sometimes slower than a single card. At the end of the day, you have to decide if it is worthwhile for you. I find that it is.

 

[mikeblas]

A)Reduced reliability is untrue, 7 years and no issues. A drive does not become less reliable if you have 2. Nobody makes this argument for SLI.

Certainly, you don't believe that because you have had no issues, nobody else has ever had issues, do you? Or that nobody else ever will, right?

Oh quick nit picking.. each thread accessing separate spindles is better than both threads accessing the same one.. you could even have two RAID 0 arrays if you wanted to accomplish this.

Sorry; I'm just trying to figure out what you mean, as what you wrote isn't clear.

Thing is that most programs aren't as configurable as you're hoping. Sure, I can set up a database server to try and have different objects on different drives, spreading out load. But most other applications just read from disk, and don't provide any ability to decide what's read from where, or which thread will do the reading. Hard partitioning of the data, then, is impossible; but RAID does help with that as multiple requests to the same logical spindle can be concurrently handled across multiple physical spindles, increasing the IOPS for the system without fooling around with any settings--inaccessible or not.

 

[nitrobass24]

Clicky, check out the bar graph towards the bottom of Eugene's post. SR will be posting a pretty comprehensive article on RAID, hopefully soon. The graph is a selection from the results obtained for that forthcoming article.

Even running _four_ WD740GDs in RAID-0 is not enough to match the single user performance of a single WD1500ADFD. The 4x RAID-0 array manages a tie in the OfficeDriveMark, but falls short in the remaining four of SR's Desktop DriveMarks. Also observe the 4x RAID-0 array delivering _slower_ performance than 2x RAID-0 in FarCry and WoW, and only managing a gain of a few percent in the Sims 2.

You are welcome to say what you want, but the simple fact remains that no RAID setup will deliver improved desktop performance. Power users looking to increase storage performance should look to faster single drives or adding indendent spindle(s) to service additional loads.

Well this isnt exactly an apples to apples comparision here is it?
two older 74gb drives vs. a newer model 150gb...if your not using the same hardware how can this be considered a fair test?

RAID-0's shortcomings are well documented:

A) Reduced reliability
B) Increased heat/power draw/noise
C) Increased system complexity
D) Greatly complicated backup and disaster recovery
E) Substantially increased cost

A) Reduced reliability? Yes with any raid array there are more points of failure, and you chances for realizing a drive failure is increased. But that does not mean that a single drive is more reliable than any given RAID array. You never know if a disk will fail or which one. Also lets say your single drive fails...your screwed just like in a raid0 situation.

B) Increased heat/power/noise? WTF ok so lets say you have 4x500gb drives in your system. heat/power/noise levels don't change if you run them in a raid or as single disk.

C)Increased system complexity? Last time I checked computers are pretty damn complex machines. Also someone that is doing raid is probably not your general computer user that goes to best buy buys a computer and has geek squad set it up for them because they cant do simple task like turn the computer on or connect a printer. People that are using raid, know they are using raid because they consider themselves enthusiast/power users/ etc and generally know what they are getting themselves into.

D)Greatly complicated backup and disaster recovery? How the hell is backup any more complicated? I have a RAID0 and I backup the exact same way that i do on my laptop...nothing special there.

E) Substantially increased cost? Again WTF. How is more expensive to to take your drives in your system and run them in a RAID0? Ok so it takes about 15 minutes of your time to go into the bios and set up the onboard raid. You just dont need expensive raid controllers to do a RAID0...there is almost no cpu overhead involved.


Not all applications/situations will benefit from RAID but some will. you just cannot make a generalization about a technology because there is not a one size fits all solution. People ask in the forum all the time if they should do RAID, and I dont have the same answer every time...that because you have to take what the person is doing/trying to accomplish and evaluate the situation. If he is trying to get his computer to boot faster yea RAID0 isnt really going to help, but if he is doing large video encoding that involves large contiguous writes he will see a significant improvement.

So quit forcing your uneducated opinions upon others
If you dont like RAID then dont do it.

 

[xxEIEIOxx]

Certainly, you don't believe that because you have had no issues, nobody else has ever had issues, do you? Or that nobody else ever will, right?

I'm not claiming it isn't possible. My complaint is that people throw it out there like it is absolutely going to happen. As if hard drives had a 50% failure rate, and RAID 0 had a 100% failure rate. Failures are most likely in the single digit percentages, and most likely, if you buy 2 drives you will be fine. If you stripe 100 drives together, you will probably get a bad one. If you buy 1 drive, it could just as well be bad. If you don't backup your system you are taking that risk regardless of the drive setup, and if you do it doesn't matter.

 

[kumquat]

I'm not claiming it isn't possible. My complaint is that people throw it out there like it is absolutely going to happen. As if hard drives had a 50% failure rate, and RAID 0 had a 100% failure rate.

Uh, no. No one said that. This is called a "straw man argument." And it's a logical fallacy.

The cold hard mathematical truth is that a RAID 0 array is significantly less reliable than a single drive. By going to a RAID 0 array instead of a single drive you more than double your chances of a catastrophic failure that will result in complete data loss.

End of story.

 

[Trepidati0n]

Uh, no. No one said that. This is called a "straw man argument." And it's a logical fallacy.

The cold hard mathematical truth is that a RAID 0 array is significantly less reliable than a single drive. By going to a RAID 0 array instead of a single drive you more than double your chances of a catastrophic failure that will result in complete data loss.

End of story.

Yes and no. The single drive system needs to be 2x as large as a typical 2 drive RAID 0 setup to get equal level of data. Therefore soon as you double the size of the drive to compete with RAID0, the test is not longer that of equivalence. A computer with X drives in will have the same drive failure rate regardless of operating in a single drive capacity or a RAID capacity. RAID does not increase failure...addition of drives does.

 

[unhappy_mage]

*Sigh* Not this thread again...

Well this isnt exactly an apples to apples comparision here is it?
two older 74gb drives vs. a newer model 150gb...if your not using the same hardware how can this be considered a fair test?

They're comparing what were (at the time, in early 2006!) two similarly-priced solutions.

But that does not mean that a single drive is more reliable than any given RAID array. You never know if a disk will fail or which one. Also lets say your single drive fails...your screwed just like in a raid0 situation.

If the RAID array and the single disk are made of similar drives... yes, it *does* mean more potential for data loss. Even comparing a 2-disk raid 0 to two independent disks there is an increased potential to lose data---if one disk of the raid fails, you lose everything; if one independent disk fails whatever was on the other disk is still okay.

B) Increased heat/power/noise? WTF ok so lets say you have 4x500gb drives in your system. heat/power/noise levels don't change if you run them in a raid or as single disk.

That's usually true (depending on whether you have an offboard RAID controller). But I don't think anyone is arguing that point; more disks mean more power draw. This thread is more about using RAID-0 not necessarily helping performance.

People that are using raid, know they are using raid because they consider themselves enthusiast/power users/ etc and generally know what they are getting themselves into.

Except when, for example, Dell sells RAID 0 as an option without explaining it at all.

D)Greatly complicated backup and disaster recovery? How the hell is backup any more complicated? I have a RAID0 and I backup the exact same way that i do on my laptop...nothing special there.

This was a better argument in 2006. Old versions of Ghost that many home users used at the time didn't support RAID arrays, since they were DOS-based and no drivers for the controllers were available.

E) Substantially increased cost? Again WTF. How is more expensive to to take your drives in your system and run them in a RAID0? Ok so it takes about 15 minutes of your time to go into the bios and set up the onboard raid. You just dont need expensive raid controllers to do a RAID0...there is almost no cpu overhead involved.

But you have to buy two drives to begin with.

If he is trying to get his computer to boot faster yea RAID0 isnt really going to help, but if he is doing large video encoding that involves large contiguous writes he will see a significant improvement.

On this point we agree. Raid of any sort (well, maybe except RAID 2) has its place. But for many home and enthusiast rigs, it's a waste of money and time to use it.

So quit forcing your uneducated opinions upon others
If you dont like RAID then dont do it.

Uneducated is apparently in the eye of the beholder.

 

[Lazn_Work]

Sorry; I'm just trying to figure out what you mean, as what you wrote isn't clear.

Thing is that most programs aren't as configurable as you're hoping. Sure, I can set up a database server to try and have different objects on different drives, spreading out load. But most other applications just read from disk, and don't provide any ability to decide what's read from where, or which thread will do the reading. Hard partitioning of the data, then, is impossible; but RAID does help with that as multiple requests to the same logical spindle can be concurrently handled across multiple physical spindles, increasing the IOPS for the system without fooling around with any settings--inaccessible or not.

Sure you can, install the program to a drive and most of it's writes will be do that drive, assign a drive or folder for swap space in Photoshop, set your temp folder locations, set download folders, virtual memory, etc. etc. There is lots of advantages to having separate spindles if you multi task.

 

[mikeblas]

Sure you can, install the program to a drive and most of it's writes will be do that drive, assign a drive or folder for swap space in Photoshop, set your temp folder locations, set download folders, virtual memory, etc. etc. There is lots of advantages to having separate spindles if you multi task.

That might give you process-level control, but doesn't give you thread-level control.

 

[Lazn_Work]

That might give you process-level control, but doesn't give you thread-level control.

Again you are nit picking. Technically you are right. I probably shouldn't have said "thread" because threads are written by the programmers.

My point still stands that: not having multiple separate simultaneous reads and rights to the same spindle/array is best and can be alleviated by properly setting up separate "drives."

And that when you have multiple accesses occurring regularly in your computer usage you are better served by separate "drives" serving those separate accesses than by one single large RAID0 array trying to serve it all.

 

[goldznnz]

just setup a raid0 w x2 velociraptors, shatt bee tite!

either i smokin crack or this setup is very quick!

and i was just coming off the old single raptors ADFD

 

[mikeblas]

And that when you have multiple accesses occurring regularly in your computer usage you are better served by separate "drives" serving those separate accesses than by one single large RAID0 array trying to serve it all.

I'm sure you'll dismiss my corrections as "nitpicking" again, but I think you're actually better off with RAID0 in the long run.

Under load, using multiple individual drives is about as fast as the same number of drives setup in RAID0 as one volume. It depends on the parameters of the scenario, as always, but the controller is going to distribute the work among the four spindles.

When not fully loaded, an application going after the RAID0 volume is still going to have its accesses partitioned among all the physical disks in the array. Since the RAID0 array has more IOPS capacity than the individual drive, this scenario is a win for RAID0. Realistically, I think the system will spend more time in this mode than it will in the mode where all drives are busy with individual applications.

Plus, you have the increased flexibility. If one application is configured to use one drive, performance will degrade substantially if another application using that same drive is run at the same time. RAID0 will better absorb the demand, since it still has all the physical disks at its disposal.

 

[Lazn_Work]

I'm sure you'll dismiss my corrections as "nitpicking" again, but I think you're actually better off with RAID0 in the long run.

Under load, using multiple individual drives is about as fast as the same number of drives setup in RAID0 as one volume. It depends on the parameters of the scenario, as always, but the controller is going to distribute the work among the four spindles.

When not fully loaded, an application going after the RAID0 volume is still going to have its accesses partitioned among all the physical disks in the array. Since the RAID0 array has more IOPS capacity than the individual drive, this scenario is a win for RAID0. Realistically, I think the system will spend more time in this mode than it will in the mode where all drives are busy with individual applications.

Plus, you have the increased flexibility. If one application is configured to use one drive, performance will degrade substantially if another application using that same drive is run at the same time. RAID0 will better absorb the demand, since it still has all the physical disks at its disposal.

Here is where I disagree with you. (and this time you are not nit picking, you are addressing the issue at hand)

I know you will just dismiss it as specific to databases.. But on the SQL server at work I had a choice to configure the 6 drives any way I wanted. I decided on three volumes of RAID 1.

Two drives are SATA 80gb 7200 RPM drives for the OS and SQL install in RAID 1.

The other 4 are 15000 RPM SAS drives where I have them as two RAID 1 arrays when I could have had one RAID 10 array.

Why did I choose two RAID 1 arrays when one RAID 10 array is "faster?"

Because I put the SQL databases on one, and the SQL transaction logs on the other.

The transaction logs are almost always writing, and almost never reading, and they write every time anything happens to the databases.. Forcing one volume to write twice for every write is slower than having each volume writing independent of each other.

This can also apply in other environments, but not as clearly or as often.

 

[mikeblas]

(and this time you are not nit picking, you are addressing the issue at hand)

I'm sorry to see you continue to focus on this. I ask clarifying questions when I'm not sure I completely understand someone's post. It saves lots of time... except when I talk to you, because you change the subject to your perception of nitpicking. One day, maybe you'll realize there's a possibility that what you write isn't as clear as you think it is.

This can also apply in other environments, but not as clearly or as often.

That's precisely the problem: it doesn't happen as often. For database applications (and some other apps) you know that the log is always writing (and writing sequentially, too), and that can easily be segregated from other disk activity.

Let's say you had eight drives available for your data. How would you distribute them? Would you continue to make small RAID1 arrays, and manually assign individual indexes and tables to them? Or would you pile up the spindles into a larger RAID1 or RAID10 array, and let the controller sort it out?

 

[Lazn_Work]

I'm sorry to see you continue to focus on this. I ask clarifying questions when I'm not sure I completely understand someone's post. It saves lots of time... except when I talk to you, because you change the subject to your perception of nitpicking. One day, maybe you'll realize there's a possibility that what you write isn't as clear as you think it is.

That's precisely the problem: it doesn't happen as often. For database applications (and some other apps) you know that the log is always writing (and writing sequentially, too), and that can easily be segregated from other disk activity.

Let's say you had eight drives available for your data. How would you distribute them? Would you continue to make small RAID1 arrays, and manually assign individual indexes and tables to them? Or would you pile up the spindles into a larger RAID1 or RAID10 array, and let the controller sort it out?

I don't think I was changing the subject or perspective at all.. You were focused on if I called it a program or a thread or all the things not pertaining to the array and accessing it.

And if you look at my posts all I was saying is that there are times when separate spindles are superior to aggregate arrays, not that it is always the case. Photoshop is one example where I'd want a RAID 0 array for swap and a separate drive (raid or not) for running the program and storing the data on. SQL is just the clearest example of my point, but not the only one.

As for your 8 drive example, I'd just RAID 10 them and be done as it is simple and provides quite a bit of speed.

When you get into a SAN environment you just throw more spindles at the controller and let it sort them out, sometimes not even setting the RAID level, just letting it determine what speed the disks are and what level of data protection you need and let it handle the details. Look in to Compellent's SANs to see some really cool stuff.