Real-world demonstration of RAID level vs read speed
nadams
Posts: 5,877
The discussion of SSD drives in another thread led me to post this.
We all know that when you stripe (RAID 0) multiple drives, you get an increase in speed, until you hit the limit of the RAID controller. But of course, this doesn't provide any redundancy whatsoever. If you lose one drive, you lose the whole shebang.
This leads many people to opt for mirroring (RAID 1) for some sort of redundancy. The problem here is, you have exactly half the useable space of the array. The advantages are, you can lose either drive and stay up, and you only have to have 2 drives.
To get more space, some people will decide on RAID 5. Here, you have to have at least 3 drives, and you will lose the space of only one of the drives. Essentially, the third drive acts as a parity drive. However, this means that, while you will see a slight increase in speed, that increase is limited due to the fact that the controller always has to keep track of the parity information. This bogs the controller processor down.
So that takes us on to RAID 10, which is a stripe of two or more mirrors. What does this bring us? The speed of a striped array, with better redundancy than a RAID 5 array. You can lose any one disk from each mirror, and because there's no parity info to keep track of, speed is not comprimised. The obvious downside is the cost of having to have at least 4 drives, plus losing half the capacity of what you're buying. To get a large database array, you're going to need a lot of money.
To demonstrate this, I set up a server and configured it with all three RAID options, and ran read tests on each. Write tests were unable to be performed, because the 1u server didn't have enough open drive spots to have the array as a secondary drive.
The server is an HP DL360G6 with a single quad-core Xeon 2ghz processor, 4gb RAM, and HP 10k RPM dual-port SAS drives on the onboard controller with the optional 256mb cache upgrade.
First up we have the RAID 5. This is utilising three of the drives.
As we can see, performance is fairly predictable. We start out at just over 100MB/s, and finish with an average of 90MB/s. Not bad, but comparable to my single 750gb SATA drive that I have in my desktop.
Next up we'll do the RAID 10. This is utilising 4 drives. Two mirrors, striped together.
Oh yes... Now we see where the lack of parity and the addition of striping comes in handy! Peak read is over 250MB/s! And the average is almost twice the MAX of RAID 5. This is clearly the winner when redundancy is required!
Okay, but what about RAID 0. I mean, just for S&G's, right?
Obviously the true speed winner, but without the advantage of any type of redundancy (As such, it shouldn't really be considered a RAID level at all, but what else are you going to call it?)
Notice that with the RAID 0, there's no rotational loss from start to finish. My theory is, this controller isn't fast enough to process all that data from 4 drives, even at the slowest points of the test.
So there you have it, in pretty pictures. I'm going to do a base test of one of the drives standalone in a few minutes, as soon as the server finishes reloading the OS (I knew I should've made a basic image)
We all know that when you stripe (RAID 0) multiple drives, you get an increase in speed, until you hit the limit of the RAID controller. But of course, this doesn't provide any redundancy whatsoever. If you lose one drive, you lose the whole shebang.
This leads many people to opt for mirroring (RAID 1) for some sort of redundancy. The problem here is, you have exactly half the useable space of the array. The advantages are, you can lose either drive and stay up, and you only have to have 2 drives.
To get more space, some people will decide on RAID 5. Here, you have to have at least 3 drives, and you will lose the space of only one of the drives. Essentially, the third drive acts as a parity drive. However, this means that, while you will see a slight increase in speed, that increase is limited due to the fact that the controller always has to keep track of the parity information. This bogs the controller processor down.
So that takes us on to RAID 10, which is a stripe of two or more mirrors. What does this bring us? The speed of a striped array, with better redundancy than a RAID 5 array. You can lose any one disk from each mirror, and because there's no parity info to keep track of, speed is not comprimised. The obvious downside is the cost of having to have at least 4 drives, plus losing half the capacity of what you're buying. To get a large database array, you're going to need a lot of money.
To demonstrate this, I set up a server and configured it with all three RAID options, and ran read tests on each. Write tests were unable to be performed, because the 1u server didn't have enough open drive spots to have the array as a secondary drive.
The server is an HP DL360G6 with a single quad-core Xeon 2ghz processor, 4gb RAM, and HP 10k RPM dual-port SAS drives on the onboard controller with the optional 256mb cache upgrade.
First up we have the RAID 5. This is utilising three of the drives.
As we can see, performance is fairly predictable. We start out at just over 100MB/s, and finish with an average of 90MB/s. Not bad, but comparable to my single 750gb SATA drive that I have in my desktop.
Next up we'll do the RAID 10. This is utilising 4 drives. Two mirrors, striped together.
Oh yes... Now we see where the lack of parity and the addition of striping comes in handy! Peak read is over 250MB/s! And the average is almost twice the MAX of RAID 5. This is clearly the winner when redundancy is required!
Okay, but what about RAID 0. I mean, just for S&G's, right?
Obviously the true speed winner, but without the advantage of any type of redundancy (As such, it shouldn't really be considered a RAID level at all, but what else are you going to call it?)
Notice that with the RAID 0, there's no rotational loss from start to finish. My theory is, this controller isn't fast enough to process all that data from 4 drives, even at the slowest points of the test.
So there you have it, in pretty pictures. I'm going to do a base test of one of the drives standalone in a few minutes, as soon as the server finishes reloading the OS (I knew I should've made a basic image)
Ludicrous gibs!
Post edited by nadams on
Comments
-
I'm a RAID idiot, but would it be possible to do a RAID 0 for operating systems and programs and then RAID 1 for music, pictures, etc. that can't be replaced? Like I said, I am an idiot when it comes to this and didn't know if it would even be something that your RAID controller would handle.
Thank you for the info though. I never realized the speed increases.I know just enough to be dangerous, but don't tell my wife, she thinks I'm a genius.
Pioneer VSX-816
Monitor 40's - fronts, bi-amped
Monitor 30's - surrounds
CS1 - center
PSW10 - I'll let you guess
Blue Jeans Cable - speaker cable
Daewoo 27 incher - one step up from a console
Sony Progressive scan DVD
XBOX
SOPA since 2008
Here's my stuff. -
Sure, you can run multiple arrays. But they have to be separate drives. So you could have two drives in RAID 0 for the OS, and then two 1tb drives or whatever in RAID 1 for the other stuff. Then you'd see the benefits of a fast OS/program drive, and redundancy of the other bits.Ludicrous gibs!
-
Okay cool. I may have some shopping to do later this year!I know just enough to be dangerous, but don't tell my wife, she thinks I'm a genius.
Pioneer VSX-816
Monitor 40's - fronts, bi-amped
Monitor 30's - surrounds
CS1 - center
PSW10 - I'll let you guess
Blue Jeans Cable - speaker cable
Daewoo 27 incher - one step up from a console
Sony Progressive scan DVD
XBOX
SOPA since 2008
Here's my stuff. -
I thought this post was pretty cool, so I installed the trial version of the prgram you used. Here are the results from my 4x1Tb RAID 10 array. I'm guessing the access time difference is because my drives are 7500RPM vs. 10k??HT
RTi70 mains
CSi30 center
RTi28 Rears
Velodyne CHT-12
H/K AVR-247
ADCOM GFA-7000
Samsung PN58B860
Playstation 3
2-Channel
Polk Audio LSi15's
Rotel RCD-1072
Nakamichi CA-5 Pre
ADCOM GFA-555
Signal Cable Analog II IC's
Signal Ultra Bi-Wire Speaker Cables -
I thought this post was pretty cool, so I installed the trial version of the prgram you used. Here are the results from my 4x1Tb RAID 10 array. I'm guessing the access time difference is because my drives are 7500RPM vs. 10k??
Yep, the lower access times are due to the 10k drives. That's pretty much all they give you. Better random access for small files. Large file transfers are not effected.Ludicrous gibs! -
What would you suggest for a RAID for music....with little concern about speed or access? Or should I bring in my main drive and go RAID 5?
Nice stuff Noah, I enjoyed checking it all out.CTC BBQ Amplifier, Sonic Frontiers Line3 Pre-Amplifier and Wadia 581 SACD player. Speakers? Always changing but for now, Mission Argonauts I picked up for $50 bucks, mint. -
Imagine the speed you would get with SSDs configured in raid. It should be way faster correct ,since no moving parts?Shoot the jumper.....................BALLIN.............!!!!!
Home Theater Pics in the Showcase :cool:
http://www.polkaudio.com/forums/showcase/view.php?userid=73580 -
What would you suggest for a RAID for music....with little concern about speed or access? Or should I bring in my main drive and go RAID 5?
Nice stuff Noah, I enjoyed checking it all out.
If you're not concerned with speed or low access times, RAID1 is the cheapest redundant option for a home environment, provided you can get away with a maximum storage capacity of one drive. If you want two drive's worth of storage, with redundancy, then RAID5 would be the best option.
RAID5 should never be used in a corporate environment, as long as you have the money for RAID10. RAID 5 is far too slow.Ludicrous gibs! -
Imagine the speed you would get with SSDs configured in raid. It should be way faster correct ,since no moving parts?
Your random access times will be sub-1ms. With data transfers, you will eventually hit the limit of your controller, like I did with my 4 drive RAID 0 array above.
Also, write speeds with SSD's are still not as fast as mechanical hard drives. SSD's with TRIM support are getting there, but they still don't bench as fast when writing.Ludicrous gibs! -
RAID5 should never be used in a corporate environment, as long as you have the money for RAID10. RAID 5 is far too slow.
This is what it really boils down to. You should also have a hot spare in a corporate environment for either configuration which adds one more disk's worth of capacity lost.
Thanks for the testing data. Very interesting comparison.____________________________________________________________
polkaudio Fully Modded SDA SRS 1.2TLs + Dreadnaught, LSiM706c, 4 X Polk Surrounds + 4 X ATMOS, SVS PB13 Ultra X 2, Pass Labs X1, Marantz 7704, Bob Carver Crimson Beauty 350 Tube Mono Blocks, Carver Sunfire Signature Cinema Grande 400x5, ADCOM GFA 7807, Panasonic UB420, Moon 380D DAC, EPSON Pro Cinema 6050
