Raid 0 vs. Raid 3 Speed Test at DVinfo.net

Adam Gold · August 11th, 2009, 06:35 PM

Obviously this is far from definitive, but as I (and I gather many others) always wonder what the "best" setup is, I decided to run a couple of tests comparing RAID0 vs. RAID3 when reformatting my video drives today.

For the past few months I've been using 7 x 1TB drives in RAID0 because I've been told it was faster, but I was always concerned about lack of redundancy. So after backing up all my final renders, I blew all my raw footage and projects off the array and reformatted as RAID3 plus Hot Spare (total capacity = 5TB).

Before and after, I ran HDTach to get a sense of speed.

Anyway, here are the results:

RAID0
Quick Test
Burst Speed 624 MBps
Avg Read 356 MBps

Long Test
Burst Speed 675 MBps
Avg Read 487 MBps

RAID3
Quick Test
Burst Speed 660 MBps
Avg Read 502 MBps

Long Test
Burst Speed 649 MBps
Avg Read 489 MBps

So it appears that you certainly don't give up speed by opting for the safety of RAID3.

Note: Formatted new volume as GPT after Vista 64 recommended it for volumes greater than 2TB and I found a post from Harm who also pointed out this was a good idea.

Tripp Woelfel · August 11th, 2009, 06:55 PM

Your results are not surprising at all. Read performance for long block data such as video is generally equivalent between Levels 0 and 3. It can be with 5 as well, but there's a shed load of variables that will contribute to your performance. Where performance will differ between redundant and non redundant arrays is on writes. Depending upon your stripe size and other parameters the difference may be small or large. Regardless, the computation of the parity data and then writing it will take time. On reads parity data is not used unless you have a failed drive. On a fully working read performance of RAID 3 should be comparable to a non fault tolerant array. If you had an equal number of data drives, it should be nearly identical.

Adam Gold · August 11th, 2009, 06:59 PM

So do you have an opinion on what some of those parameter values should ideally be? I just sort of accepted Areca's defaults when I told it to create the new array....

Harm Millaard · August 12th, 2009, 02:32 AM

Quote:

Originally Posted by Adam Gold

So do you have an opinion on what some of those parameter values should ideally be? I just sort of accepted Areca's defaults when I told it to create the new array....

The only adjustment I made was setting the stripe size to 128K.

My HDTach long test gave a burst of 1079 MB/s and average read of 853 MB/s.

When using the Atto benchmark my average read was similar, but the average write results topped off around 550 MB/s. That is the penalty of using parity that Tripp mentioned. In SiSoft the drive index in MB/s was nearly 1500.

To freshen your memory, my raid is a 12 (x 1 TB) disk Raid30 for 10 TB effective space.

Adam Gold · August 12th, 2009, 01:16 PM

Good advice, Harm. I'll give that a shot when I next reformat the array, which will likely be when I add the five extra eSATA drives to max out the Areca's 12-disk capacity.

I assume your extra speed also comes from having more disks? As I recall your charts, while the speed goes up with more disks, there is a point of diminishing returns, no?

Harm Millaard · August 12th, 2009, 02:17 PM

Yes, Adam, there is a point of diminishing returns. It hardly made any difference in performance going from 10 to 12 disks. From a space point of view I can still add two more disks internally, bringing the total to 19, and my Areca still allows for 4 more disks, but performance wise it will give not me any benefits, due to those diminishing returns.

What may give me the extra performance is the use of 2 GB cache on the controller, instead of the standard 512 MB. Going to 4 GB would be a further performance boost, but the price of that memory is not worth it IMO.

Tripp Woelfel · August 12th, 2009, 07:54 PM

Harm's right... as he normally is. Data is always going to move through memory much faster than through rotating magnetic (disk). Adding memory cache might save you a few disk reads which will speed things up. Will you notice it in the real world? That depends.

One thing to remember is that data in memory is not protected from failures. Lose power or have a RAID controller failure and whatever's in cache is gone... see ya... bye. This may not matter since video editing is almost exclusively reading until you save a project or render video. It's probably mouse walnuts.

Harm Millaard · August 13th, 2009, 01:05 AM

Tripp,

You are right again and that is the reason I also have a battery backup module BBM installed on the Areca, so it can finish writing from cache to disk in the case of power failure.

Adam Gold · August 13th, 2009, 11:12 AM

...and thanks to Harm's advice a year and a half ago I have both 2GB RAM and the BBM on my Areca...

August 11th, 2009, 06:35 PM	#1
Adam Gold Inner Circle Join Date: Jan 2007 Location: Woodinville, WA USA Posts: 3,467	Raid 0 vs. Raid 3 Speed Test Obviously this is far from definitive, but as I (and I gather many others) always wonder what the "best" setup is, I decided to run a couple of tests comparing RAID0 vs. RAID3 when reformatting my video drives today. For the past few months I've been using 7 x 1TB drives in RAID0 because I've been told it was faster, but I was always concerned about lack of redundancy. So after backing up all my final renders, I blew all my raw footage and projects off the array and reformatted as RAID3 plus Hot Spare (total capacity = 5TB). Before and after, I ran HDTach to get a sense of speed. Anyway, here are the results: RAID0 Quick Test Burst Speed 624 MBps Avg Read 356 MBps Long Test Burst Speed 675 MBps Avg Read 487 MBps RAID3 Quick Test Burst Speed 660 MBps Avg Read 502 MBps Long Test Burst Speed 649 MBps Avg Read 489 MBps So it appears that you certainly don't give up speed by opting for the safety of RAID3. Note: Formatted new volume as GPT after Vista 64 recommended it for volumes greater than 2TB and I found a post from Harm who also pointed out this was a good idea. Last edited by Adam Gold; August 11th, 2009 at 09:09 PM.

August 11th, 2009, 06:55 PM	#2
Tripp Woelfel Trustee Join Date: Jan 2008 Location: North Conway, NH Posts: 1,745	Your results are not surprising at all. Read performance for long block data such as video is generally equivalent between Levels 0 and 3. It can be with 5 as well, but there's a shed load of variables that will contribute to your performance. Where performance will differ between redundant and non redundant arrays is on writes. Depending upon your stripe size and other parameters the difference may be small or large. Regardless, the computation of the parity data and then writing it will take time. On reads parity data is not used unless you have a failed drive. On a fully working read performance of RAID 3 should be comparable to a non fault tolerant array. If you had an equal number of data drives, it should be nearly identical.

August 11th, 2009, 06:59 PM	#3
Adam Gold Inner Circle Join Date: Jan 2007 Location: Woodinville, WA USA Posts: 3,467	So do you have an opinion on what some of those parameter values should ideally be? I just sort of accepted Areca's defaults when I told it to create the new array....

August 12th, 2009, 01:16 PM	#5
Adam Gold Inner Circle Join Date: Jan 2007 Location: Woodinville, WA USA Posts: 3,467	Good advice, Harm. I'll give that a shot when I next reformat the array, which will likely be when I add the five extra eSATA drives to max out the Areca's 12-disk capacity. I assume your extra speed also comes from having more disks? As I recall your charts, while the speed goes up with more disks, there is a point of diminishing returns, no?

August 12th, 2009, 02:17 PM	#6
Harm Millaard Trustee Join Date: Aug 2006 Location: Rotterdam, Netherlands Posts: 1,832	Yes, Adam, there is a point of diminishing returns. It hardly made any difference in performance going from 10 to 12 disks. From a space point of view I can still add two more disks internally, bringing the total to 19, and my Areca still allows for 4 more disks, but performance wise it will give not me any benefits, due to those diminishing returns. What may give me the extra performance is the use of 2 GB cache on the controller, instead of the standard 512 MB. Going to 4 GB would be a further performance boost, but the price of that memory is not worth it IMO.

August 12th, 2009, 07:54 PM	#7
Tripp Woelfel Trustee Join Date: Jan 2008 Location: North Conway, NH Posts: 1,745	Harm's right... as he normally is. Data is always going to move through memory much faster than through rotating magnetic (disk). Adding memory cache might save you a few disk reads which will speed things up. Will you notice it in the real world? That depends. One thing to remember is that data in memory is not protected from failures. Lose power or have a RAID controller failure and whatever's in cache is gone... see ya... bye. This may not matter since video editing is almost exclusively reading until you save a project or render video. It's probably mouse walnuts.

August 13th, 2009, 01:05 AM	#8
Harm Millaard Trustee Join Date: Aug 2006 Location: Rotterdam, Netherlands Posts: 1,832	Tripp, You are right again and that is the reason I also have a battery backup module BBM installed on the Areca, so it can finish writing from cache to disk in the case of power failure.

August 13th, 2009, 11:12 AM	#9
Adam Gold Inner Circle Join Date: Jan 2007 Location: Woodinville, WA USA Posts: 3,467	...and thanks to Harm's advice a year and a half ago I have both 2GB RAM and the BBM on my Areca...