Let's end the RAID-0 debate

E

EnderW

[H]F Junkie
Joined
Sep 25, 2003
Messages
11,249
Anyone who browses this forum regularly know there is somewhat of a debate over the benefits of RAID-0 for a single user system.

I'd like to end that debate, so I'm going to pick up two WD360ADFD drives and set them up in RAID-0, run some benchmarks and then compare the results with those from a single drive.

This will probably be done over christmas break, which is about a month away, so I hope to post the results before the new year.

However, I need some input on what benchmarks to run and how best to record the results.
I'd like to focus on application loading times and things of a similar nature, but the bad thing is these can't really be measured by a benchmarking utility (that I know of), so I'll need to bust out the stop watch.
That introduces a certain amount of error, so I was thinking of doing each test 3 times and posting the average.

I'd also like to do some video, although not for every test as that would be too time consuming.
And since I only have one system, what I would do it edit the videos together later so you can see the difference "side by side"

I also need input on which stripe size to use. It's possible to use a couple different ones, but since I'll be performing the RAID-0 tests twice, once on the motherboard controller, and again on my Highpoint 2320, I can't do every possible choice for all the benchmarks.

The different disk configurations will probably look something like this:

  • Single WD360ADFD
  • 2 x WD360ADFD RAID-0 on motherboard controller Stripe Size A
  • 2 x WD360ADFD RAID-0 on motherboard controller Stripe Size B
  • 2 x WD360ADFD RAID-0 on Highpoint 2320 controller Stripe Size A
  • 2 x WD360ADFD RAID-0 on Highpoint 2320 controller Stripe Size B


So, I need input on the following:
  1. Testing methods
  2. List of benchmarks
  3. Stripe sizes
 
i'd love to see one last configuration:
  • two raptors, one for the OS, the other for the application under inspection

and I have done something similar here.

whatever you do, I think the most important part is to clearly describe what you are measuring. I remember there is a RAID-0 article on the web that doesn't clearly define what "windows load time" is. It makes little mention of whether the POST time was included or not and when the "load time" was over. Being precise in that area allows us to recreate the 'benchmark' at home and compare our systems.

Also, make sure that you consider rebooting between every run to eliminate caching.
 
RAID 0 is definitely faster than a single 7200 RPM drive or drives in non-RAID mode.

Very difficult to notice a difference if compared to a 10000 RPM drive.

10000 RPM drives in RAID 0 = WIN!


/End debate
 
Standpoint said:
RAID 0 is definitely faster than a single 7200 RPM drive or drives in non-RAID mode.

Very difficult to notice a difference if compared to a 10000 RPM drive.

10000 RPM drives in RAID 0 = WIN!


/End debate
Click to expand...

You ended the "end debate", so does that mean it's started again :D ;)
 
Why only the 36GB drives for testing?

I had 2x 36GB raptors in RAID0 that were fast -- when I got 2x 74GB raptors in RAID0, they sure seemed faster to me (loading windows, loading games mainly)
 
cleric_retribution said:
Why only the 36GB drives for testing?

I had 2x 36GB raptors in RAID0 that were fast -- when I got 2x 74GB raptors in RAID0, they sure seemed faster to me (loading windows, loading games mainly)
Click to expand...

That's interesting, becuase it shouldn't matter the saize of the drive, if the other factors, such as spindle speed and cache are the same...
 
cleric_retribution said:
Why only the 36GB drives for testing?

I had 2x 36GB raptors in RAID0 that were fast -- when I got 2x 74GB raptors in RAID0, they sure seemed faster to me (loading windows, loading games mainly)
Click to expand...
So you are saying that there is a nonlinearity in the speedup that R-0 can provide with regards to the size of the drives?
 
I have 4 10k scsi drives that I'll run any benchmarks on once I get 68->80s in for them. Then maybe we can settle the scsi/sata debate too. Maybe I should buy some Fujitsu MAXes too... :D

EnderW said:
However, I need some input on what benchmarks to run and how best to record the results.
Click to expand...
Here are my suggestions.
1) "Typical game" tests - pick BF2 or something, set up a local server, time connecting and loading map.
2) "Video editing" test - load vob file, time converting a short clip into low-quality divx. Take CPU out of the equation as much as possible - resize it down, that kind of thing.
3) Database thing, maybe? I'm not so sure we need this.
4) Boot time tests. How long does it take to boot a brand new Windows install? Power button to desktop, set to autologin.
5) Please *do* HDTach, but don't post it. It simply isn't a good benchmark, and it shouldn't be posted in the same thread with the rest of the tests.

 
drizzt81 said:
i'd love to see one last configuration:
  • two raptors, one for the OS, the other for the application under inspection

and I have done something similar here.

whatever you do, I think the most important part is to clearly describe what you are measuring. I remember there is a RAID-0 article on the web that doesn't clearly define what "windows load time" is. It makes little mention of whether the POST time was included or not and when the "load time" was over. Being precise in that area allows us to recreate the 'benchmark' at home and compare our systems.

Also, make sure that you consider rebooting between every run to eliminate caching.
Click to expand...
I will consider that, thanks.

I've seen your results, I just want to do something a little more comprehensive and nice write up with graphs and videos etc so I can "publish" it in every tech forum I can find.
 
Standpoint said:
RAID 0 is definitely faster than a single 7200 RPM drive or drives in non-RAID mode.

Very difficult to notice a difference if compared to a 10000 RPM drive.

10000 RPM drives in RAID 0 = WIN!


/End debate
Click to expand...
Get out of my thread.
 
cleric_retribution said:
Why only the 36GB drives for testing?

I had 2x 36GB raptors in RAID0 that were fast -- when I got 2x 74GB raptors in RAID0, they sure seemed faster to me (loading windows, loading games mainly)
Click to expand...
Because they are the cheapest.
And because all the ADFD Raptors are virtually the same.

Your 36GB Raptors, were they WD360GD or WD360ADFD? There is a significant difference.

I am considering using WD740ADFD drives though since I already have one, but I'm somewhat concerned about getting one with a different firmware than mine which could skew the results.

Anyone that just got a brand new WD740ADFD want to post their firmware revision?
 
unhappy_mage said:
I have 4 10k scsi drives that I'll run any benchmarks on once I get 68->80s in for them. Then maybe we can settle the scsi/sata debate too. Maybe I should buy some Fujitsu MAXes too... :D


Here are my suggestions.
1) "Typical game" tests - pick BF2 or something, set up a local server, time connecting and loading map.
2) "Video editing" test - load vob file, time converting a short clip into low-quality divx. Take CPU out of the equation as much as possible - resize it down, that kind of thing.
3) Database thing, maybe? I'm not so sure we need this.
4) Boot time tests. How long does it take to boot a brand new Windows install? Power button to desktop, set to autologin.
5) Please *do* HDTach, but don't post it. It simply isn't a good benchmark, and it shouldn't be posted in the same thread with the rest of the tests.

Click to expand...

Thanks for the suggestions. What I want to do is get a "rough draft" of the types of benchmarks to run and then settle on specific details for each one.
 
EnderW said:
Because they are the cheapest.
And because all the ADFD Raptors are virtually the same.

Your 36GB Raptors, were they WD360GD or WD360ADFD? There is a significant difference.

I am considering using WD740ADFD drives though since I already have one, but I'm somewhat concerned about getting one with a different firmware than mine which could skew the results.

Anyone that just got a brand new WD740ADFD want to post their firmware revision?
Click to expand...

FYI, all of the ADFD HDD's are NOT the same. At least 2 reviews showed the WD740ADFD's faster in almost every category over the 150GB'ers.

You do know that this testing has been done before, right? Every time it is done, people will fight it. Are you going to use a stopwatch to test boot load times? Don't do that because people ripped that methodology to shreads on this forum. Too inaccurate they said.

I have done these tests too many times for my own knowledge-to back up what I can see and "feel." If it makes you feel better, have fun - I did. But in the end, no matter what testing any one of us do - pro RAID0 or anti RAID0, this issue will still be here. People have good results with RAID0 while others have bad results. I have sat there with stopwatches timing level loads times and Windows boot times in single vs. RAID0 so many times it's rediculous - which is why I bleieve so strongly about this.

I just don't want you to be dissappointed when this thread DOESN'T END THE DEBATE. Because it won't - no matter the results. The same exact test has been run with different HDDs so many times with so many different results by so many different sites/people.

That being said, I will enjoy seeing the results and thanks for devoting the time n effort.

Some things to consider...what "mode" will the single HDDs be in? AHCI? IDE compatible? Timing Windows boot can be difficult between the 2 as different controllers will be loaded depending on configuration. You must pick a time before the splash screen, but after RAID controller or POST screens. And this varies with every motherboard. What you may gain in raw disk read time in RAID0 at boot may be negated by a 3 second RAID "Press CTRL-I" screen popping up. But you must try n time only the actual HDD activity time. I'm sure you already know this, just bringing it up. :)
 
Brahmzy said:
FYI, all of the ADFD HDD's are NOT the same. At least 2 reviews showed the WD740ADFD's faster in almost every category over the 150GB'ers.
Click to expand...
I only know of the tweakers.net review, what is the other one?
And what little difference exists is only due to the number of platters (or sides of a platter) being used, so it's not something that would be an advantage for either single drive or RAID-0 configuration.
Also, if the 74GB ones are faster than the 150GB version, wouldn't logic dictate that the 36GB ones would be even faster?


Brahmzy said:
You do know that this testing has been done before, right? Every time it is done, people will fight it. Are you going to use a stopwatch to test boot load times? Don't do that because people ripped that methodology to shreads on this forum. Too inaccurate they said.

I have done these tests too many times for my own knowledge-to back up what I can see and "feel." If it makes you feel better, have fun - I did. But in the end, no matter what testing any one of us do - pro RAID0 or anti RAID0, this issue will still be here. People have good results with RAID0 while others have bad results. I have sat there with stopwatches timing level loads times and Windows boot times in single vs. RAID0 so many times it's rediculous - which is why I bleieve so strongly about this.

I just don't want you to be dissappointed when this thread DOESN'T END THE DEBATE. Because it won't - no matter the results. The same exact test has been run with different HDDs so many times with so many different results by so many different sites/people.
Click to expand...
Yes, there have been scattered comparisons here and there, but nothing on the magnitude of what I'm trying to accomplish here.

Do you have a better recommendation on how to measure level load times?
I personally think a stopwatch is fine. Sure, there is a slight margin of error, but taking the average of 3 times will reduce that.
If the difference is so small that a stopwatch can't accurately measure it, than that kinda proves the point that there isn't a significant difference IMO.

I'm aware there are people who will always argue regardless of what the evidence shows. I don't care about them. I'm doing this for people who want to see a conclusive comparison and for myself.


Brahmzy said:
That being said, I will enjoy seeing the results and thanks for devoting the time n effort.
Click to expand...
You're welcome.
 
Hmmm. I don't know the nuances of hard drive technology, but I've been using 2 x WD360GD drives in RAID0 mode (32K stripe, I think) for about two years already. It definitely feels a lot faster than the IBM 75GXP IDE HD that I had before...
 
No Limit said:
Hmmm. I don't know the nuances of hard drive technology, but I've been using 2 x WD360GD drives in RAID0 mode (32K stripe, I think) for about two years already. It definitely feels a lot faster than the IBM 75GXP IDE HD that I had before...
Click to expand...
A single WD360GD is already faster than the IBM 75GXP.
And you can remember what a hard drive "feels" like from 2 years ago? :eek:
 
FYI, all of the ADFD HDD's are NOT the same. At least 2 reviews showed the WD740ADFD's faster in almost every category over the 150GB'ers.
Click to expand...
I'd say they pretty much are the same. You'll see a little better STR on the bigger drives since there are more heads and platters, and quicker seeks with the 36'er because of less hardware to move. Production variances alone will make them even them out.

EnderW,

Good luck with your test. I'm sure you have all the recommendations you need. I've been tempted to do something similar, but figured I just didn't need to prove to anyone else the benefits I've reaped from RAID0 as I experience them everyday. Whichever setup your test ends up in favor of, the argueing will continue as there will never be a "conclusive" test, but at least you'll have the best answer for yourself/current setup.
 
I doubt if the tests are going to resolve any issues.

I replaced my daughter's Athlon 1800+ with an Athlon64 X2 4400+, replaced the IDE (100) hard drive with an SATA II hard drive, increased ram form 1G to 2G, new OS install.

She cannot tell the difference.

---

I just replaced a 4 drive RAID 0 with individual hard drives - too much fragmentation on the RAID 0 from recording TV programs. Recording 4 channels on 4 independent drives produces almost no fragmentation. So you might want to include defrag time in your tests ...
 
Brahmzy said:
You must pick a time before the splash screen, but after RAID controller or POST screens. And this varies with every motherboard. What you may gain in raw disk read time in RAID0 at boot may be negated by a 3 second RAID "Press CTRL-I" screen popping up. But you must try n time only the actual HDD activity time. I'm sure you already know this, just bringing it up. :)
Click to expand...
Boot time ought to be defined as the time it takes from pushing the power button until the "hit CTRL+ALT+DEL" screen comes up. Yes, the POST ought to be included, since the POST time may be influenced by whether RAID is configured or not. Yes, there are tons of settings that ought to be disclosed:
Where the other IDE channels "disabled" or "auto" ...
 
tuskenraider said:
I'd say they pretty much are the same. You'll see a little better STR on the bigger drives since there are more heads and platters, and quicker seeks with the 36'er because of less hardware to move. Production variances alone will make them even them out.
Click to expand...
STR doesn't depend on the number of platters or heads. Drives only transfer from one side of a platter at once - or everyone would want five-platter drives, all the time.

Brahmzy: why wait until after the POST screens? Those have to happen*, and it would be unrealistic to test otherwise.

Could anyone load EnderW a 740ADFD or 2 to compare with the 360s? Maybe use a deposit or something to make sure you get it back. It'd be really nice to see the difference in practice.

*unless you're using LinuxBIOS. But that's not real likely.

 
drizzt81 said:
Boot time ought to be defined as the time it takes from pushing the power button until the "hit CTRL+ALT+DEL" screen comes up. Yes, the POST ought to be included, since the POST time may be influenced by whether RAID is configured or not. Yes, there are tons of settings that ought to be disclosed:
Where the other IDE channels "disabled" or "auto" ...
Click to expand...
I agree with you 100% about the boot time.
And I'm glad you mentioned the other settings.
 
unhappy_mage said:
Could anyone load EnderW a 740ADFD or 2 to compare with the 360s? Maybe use a deposit or something to make sure you get it back. It'd be really nice to see the difference in practice.
Click to expand...
Well I already have 1 WD740ADFD, so as long as I can get a second one with the same firmware, I think I'll go that route.

However, maybe as part of a separate thing, I would like to compare the WD360ADFD, WD740ADFD, and WD1500ADFD as single drives.

If anyone wants to help, I would suggest just making a small donation via paypal and I will put that towards the money that will be lost buying the drives and then reselling them.
 
unhappy_mage said:
3) Database thing, maybe? I'm not so sure we need this.
Click to expand...

I'd argue IO tests are necessary, lets keep this one!

Might I suggest a benchmarking suit/method that is easy to duplicate on Linux as well? If standardized we could create a "sticky" displaying an assortment of drives/controllers/configurations and their respective results.....
 
longblock454 said:
I'd argue IO tests are necessary, lets keep this one!

Might I suggest a benchmarking suit/method that is easy to duplicate on Linux as well? If standardized we could create a "sticky" displaying an assortment of drives/controllers/configurations and their respective results.....
Click to expand...
I have no experience with Linux and no interest in it either.
I'll provide all the details of my setup and methods, but any Linux testing will have to be done by someone else.
 
EnderW said:
I have no experience with Linux and no interest in it either.
I'll provide all the details of my setup and methods, but any Linux testing will have to be done by someone else.
Click to expand...

No interest in a larger pool of results?

Many benchmarking utilities are available for both Windoze and Linux, i'm simply suggesting to choose one of those......
 
longblock454 said:
No interest in a larger pool of results?

Many benchmarking utilities are available for both Windoze and Linux, i'm simply suggesting to choose one of those......
Click to expand...
If you have the names of those utilities, I'd be happy to use some of them in my testing.
 
unhappy_mage said:
I have 4 10k scsi drives that I'll run any benchmarks on once I get 68->80s in for them. Then maybe we can settle the scsi/sata debate too. Maybe I should buy some Fujitsu MAXes too... :D
Click to expand...

I have a pair of 73gb 15k fujitsu MAX's

these things are the definition of FAST FAST FAST
 
EnderW said:
If you have the names of those utilities, I'd be happy to use some of them in my testing.
Click to expand...

Thanks!

I'll post some suggestions once the testing method is closer to being finalized.
 
EnderW said:
I agree with you 100% about the boot time.
And I'm glad you mentioned the other settings.
Click to expand...

I think that is a silly comparison for a boot time test. This will vary a ton depending on the motherboard (SATA/RAID controllers) used. It tells us nothing of how fast the 2 scenerios can read the boot files/drivers etc. I can hear it now "Oh yeah? Well RAID0 is 7 seconds slower to boot to desktop!"

Silly way to test IMO as this will only be valid for that particular motherboard.
 
STR doesn't depend on the number of platters or heads. Drives only transfer from one side of a platter at once - or everyone would want five-platter drives, all the time.
Click to expand...
My mistake as I couldn't come up with the correct term. I meant to imply that if the data was on each side of each platter, and the heads were positioned over the file which was spread across those four platters, it would be read faster no? Considered quicker read/write? At least that's what I was told by someone I thought knew what they were talking about and sounds logical.
 
Brahmzy said:
I think that is a silly comparison for a boot time test. This will vary a ton depending on the motherboard (SATA/RAID controllers) used. It tells us nothing of how fast the 2 scenerios can read the boot files/drivers etc. I can hear it now "Oh yeah? Well RAID0 is 7 seconds slower to boot to desktop!"

Silly way to test IMO as this will only be valid for that particular motherboard.
Click to expand...
it's not the only way to test, although I believe it makes the most sense

I'll also compare how many bars XP takes too, just for fun
 
EnderW said:
...
Your 36GB Raptors, were they WD360GD or WD360ADFD? There is a significant difference.
...
Click to expand...
My drives were the "GD" models.
What is the difference? I've no idea
confused.gif
 
EnderW said:
The different disk configurations will probably look something like this:

  • Single WD360ADFD
  • 2 x WD360ADFD RAID-0 on motherboard controller Stripe Size A
  • 2 x WD360ADFD RAID-0 on motherboard controller Stripe Size B
  • 2 x WD360ADFD RAID-0 on Highpoint 2320 controller Stripe Size A
  • 2 x WD360ADFD RAID-0 on Highpoint 2320 controller Stripe Size B


So, I need input on the following:
  1. Testing methods
  2. List of benchmarks
  3. Stripe sizes
Click to expand...

I'd like to see at least 64k and 64k / number of drives for stripe size.

The 2320 doesn't allow changing the stripe size.

I suggested some simple benches in the other thread:

http://hardforum.com/showpost.php?p=1030088334&postcount=8
 
Silly way to test IMO as this will only be valid for that particular motherboard.
Click to expand...
Not really. If I have RAID and my computer takes longer to get to the desktop, the fact is I have to wait longer before I can start using it. So while I understand you're not isolating the drive(s) themselves for boot up testing this way, I think it's something many care about. You could test both ways. After BIOS initialization to desktop and pushbutton to desktop. Also, most SATA/RAID controllers take the same amount of time to initialize whether there is one drive on it or a RAID array.
 
Brahmzy said:
I think that is a silly comparison for a boot time test. This will vary a ton depending on the motherboard (SATA/RAID controllers) used. It tells us nothing of how fast the 2 scenerios can read the boot files/drivers etc. I can hear it now "Oh yeah? Well RAID0 is 7 seconds slower to boot to desktop!"

Silly way to test IMO as this will only be valid for that particular motherboard.
Click to expand...
Let me elaborate my point: I suggest that boot time should start "at the push of the power button" because that is what the end-user experience is. Unfortunately for you, me and most other people out there, my boot up time does not start after POST and RAID bios loading, but rather when I push the button. Since I do not know of a way to have a system boot without watching the POST screen etc, it would appear to me that any 'real-world' comparison must include this time, especially since it may not be constant across the configurations.

All-in-all the benchmarks will only serve as a real-world comparison, i.e. on this system, with this controller, these are my results. We all know that regardless of what controller EnderW will be using the 'losing' side of his comparison will either argue 'well if he had used a different controller, the results would be different'.

The best we could do would be to have a 'standardized' benchmark and other members could repeat the experiments at their leisure and we can have a 'benchmark db' not unlike the overclocking db in the intel forum.
 
drizzt81 said:
Let me elaborate my point: I suggest that boot time should start "at the push of the power button" because that is what the end-user experience is. Unfortunately for you, me and most other people out there, my boot up time does not start after POST and RAID bios loading, but rather when I push the button. Since I do not know of a way to have a system boot without watching the POST screen etc, it would appear to me that any 'real-world' comparison must include this time, especially since it may not be constant across the configurations.

All-in-all the benchmarks will only serve as a real-world comparison, i.e. on this system, with this controller, these are my results. We all know that regardless of what controller EnderW will be using the 'losing' side of his comparison will either argue 'well if he had used a different controller, the results would be different'.

The best we could do would be to have a 'standardized' benchmark and other members could repeat the experiments at their leisure and we can have a 'benchmark db' not unlike the overclocking db in the intel forum.
Click to expand...

True. I think both scenerios should be tested. Good grief I cannot type today - I don't even know if I spelled scenerios correctly.
 
Brahmzy said:
True. I think both scenerios should be tested. Good grief I cannot type today - I don't even know if I spelled scenerios correctly.
Click to expand...
scenarios, Firefox 2.0 has built in spell checker
 
To test the OS loading times you could get Windows to the point where you select a normal startup. Press F8 until it asks you if you would like to boot into safe mode or normal mode. Press enter on the normal mode at the same time you press start on your stopwatch.

Not that there is going to be that much of a difference in Windows loading times. It really depends on what you have in startup and what has been loaded. I am willing to bet that the average time for all your Windows loading tests will be the same.
 
You must log in or register to reply here.
Back
Top