Disaster Recovery

Non-standard RAID levels primer: RAID 1E

Scott Lowe introduces a RAID level that's a little off the beaten path. Learn more about RAID 1E and how it differs from the more common RAID levels.

Previously, I've presented you with a look at many RAID levels, including some hybrid RAID levels, such as RAID-10 and RAID-50. In a series of articles on data protection, I will provide you with a look at some non-standard RAID levels. Some of them are available only from specific manufacturers, but some are actually very slowly working their way into more general use.

RAID 1E (striped mirroring, enhanced mirroring, hybrid mirroring)

RAID 1E -- which, depending on the vendor, is also called striped mirroring, enhanced mirroring, and hybrid mirroring -- is a RAID level that combines RAID 0's striping capabilities with RAID 1’s mirroring protection. If you’re thinking that this RAID method sounds a lot like RAID 10, you should understand one critical difference between RAID 1E and RAID 10. RAID 1E uses an odd number of disks to achieve your data protection goals while RAID 10 requires the use of an even number of disks.

Whereas RAID levels 0 and 1 each require a minimum of two disks, RAID 1E requires a minimum of three disks. Keep in mind that RAID 10 requires at least four disks. As is the case under RAID 1, RAID 1E has a 50 percent disk capacity overhead. In other words, only half of the total capacity of the array is available for use.

RAID 1E works by striping data across all of the disks in the array a la RAID 0. As you know, however, RAID 0 is less than ideal for enterprise environments since the loss of any disk in the array results in the loss of all data stored on the array. Where RAID 1E becomes viable is in the next step, in which a copy of the data is then striped across all of the disks as well. Simply striping the data in the exact same way as the initial stripe would not be any good since copies of the data would still reside on the same disk as the original copy. RAID 1E therefore, shifts the second copy of the data over on each physical disk. Take a look at Figure A. Each number refers to a block of data. A number with an M refers to a mirrored block of that data.

Figure A

A simplistic look at how RAID 1E works

Now, imagine that disk five suffers a failure. Blocks three and eight are stored on that disk, along with the mirrored copies of blocks five and ten. However, the array can suffer this failure since mirrored copies of blocks three and eight are stored on disk one. In theory, you can lose multiple disks in a RAID 1E array as long as the disks are not adjacent. In Figure A, for example, you can lose disks one and three, with disk one's blocks one and six mirrored to disk two, and disk three's blocks two and seven mirrored to disk four."

RAID 1E can potentially provide more performance than a traditional RAID 1 array. With an odd number of disks, RAID 1E provides more spindles (in many RAID 1E cases, three disks/spindles instead of two). Like RAID 1, RAID 1E’s primary disadvantage is its 50 percent disk overhead. Another significant RAID 1E disadvantage is its relatively low support from controller manufacturers.

Summary

RAID 1E looks to be an interesting alternative to RAID 1 when somewhat better performance is necessary, and you don’t want to go the RAID 10 route. Are any of you running systems with RAID 1E? If so, leave a comment and let us know your reasoning and experience.

TechRepublic's free Storage NetNote newsletter is designed to help you manage the critical data in your enterprise. Automatically sign up today!

23 comments
cerulean_dragon
cerulean_dragon

I was troubleshooting an issue with a server today and discovered that we actually have several servers running RAID1E with 3 active and 1 hot spare drive. Frankly I had no idea what 1E was so I came here to find out. Thanks for the article, it was pretty good. Still trying to wrap my head around it. I thought it was mirroring across multiple drives. Personally I would think with 4 drives RAID 5 would be the optimal setup, but eh, I didn't build them. I haven't checked to see if the system's onboard SAS controller will even support 5. I know it'll do 0, 1, and (apparently) 1E. If not, that could be the reason my predecessor went with 1E. In my case, one of the drives was dead, so I moved in the hot spare and sync'd it. The LSI controller menu was pretty simple. Thanks again for the handy article.

brian
brian

I have been crawling the web looking for performance reviews between raid 10 and 1e (called IME on my LSI card, which does not support raid 10.) Are there any performance differences between 1e and 10, when the number of disks is even? I am designing a system for HD vfx compositing. It will be loading multiple layers of uncompressed 4k frames while writing the composited result. I have three options without adding additional system cost: 1: IME volume with even number of SATA drives on LSI SAS controller. (Asus PIKE 1098.) 2: Raid 10 on the same motherboard's built-in SATA controller, but limits number of drives. 3: Create two Raid-1 volumes on SAS controller, up to three Raid-1 volumes on SATA controller, and create a Windows Dynamic Volume (Software Raid-0) of those.

ehsan_nasco
ehsan_nasco

i am running my systems are RAID 1E what do I need to do if in case of failure please respond to my mail id : ehsan@nasco.com.sa

jaymdale
jaymdale

Is it possible that someone could post a diagram of what a 4-disk RAID1E data set would look like?

Tachyon
Tachyon

This is the RAID level used by ADP, at least on their 9400 series servers (which are IBM servers). It does seem to be a good compromise between performance and data security. I would imagine this scenario is typical. Where data integrity is the number one priority and cannot be compromised, but if some amount of performance can be gained without sacrificing data integrity, why not?

cdc
cdc

DISK 1 FAILED ALL ARRAY LOST IN MY IBM X3200 AND I DONT UNDERSTAND THE REASON . IT SHOULD WORK OK WITH THE NEXT 2?

michaellashinsky
michaellashinsky

Is this somehow better than a Raid 5? If it is, I''m not getting it. A raid 5 is spread out over 5 disks, for reading and writing, and any one drive can drop out without losing data. The only advantage I can see to a 1E is that two disks could drop out, as long as they are not adjacent. But you take a 50% overhead hit on the whole setup. I don't see a real world advantage. Am I missing something? Michael

laman
laman

The author seems to base the difference mostly on the number of disk requried for 1E and 10. However with the cost of hard disk getting lower and lower these days, I don't see much problem in getting 4 disks instead of 3.

Paul
Paul

The example should of said 2 and 7 for disk 3. The overhead is 50% like the author said but it is spread over multiple disks. Will it support reads from diffent disks? i.e. if you need data block 2 and 7, can it grab 2 and 7M instead? If so then the more disks you have the closer you will get to the read performance of a raid 5 in a scenario where you are doing mostly reads from disks for your application.

rjj
rjj

The overhead for RAID 1E is 100%.

joepizzi
joepizzi

The text following Figure A does not match the figure.

andrew_miga
andrew_miga

Hello, You are missing two things. First of all RAID 5 is just a name for the RAID level and not the number of disks. There is a minimum of 3 disks in a RAID 5 array and no maximum. It is common to have four or eight port controller cards using RAID 5. Secondly you are missing the tradeoff between overhead and write speed between RAID 10 or 1E arrays and RAID 5 arrays which require parity calculations for writing. With 100% hardware RAID 10/1E controllers disks are synchronized, write speed is very fast as there are no parity calculations. The real world advantage is faster write speed than RAID 5 at a trade-off of higher storage overhead. For many applications RAID 5 is impractical and RAID 10 is used. Andrew

andrew_miga
andrew_miga

Hello, The problem would be if your organization had a very limited budget for hardware. There are many people who would appreciate only having to buy one high performance SCSI disk instead of two. There may be also space or power/heat considerations for a particular server.

?/\/\?|???\/???
?/\/\?|???\/???

:) Had to chuckle at that - hope it was intentional!

jgaskell
jgaskell

So, none of the disk space is useable for data? Doesn't sound very useful ;-)

DWalker88001
DWalker88001

Yes, the text says "Now, imagine that disk three suffers a failure. Blocks three and eight are stored on that disk, along with the mirrored copies of blocks two and seven." It looks to me like blocks 2 and 7, along with mirrors of blocks 4 and 9, are stored on disk three. The author needs to correct the article.

michaellashinsky
michaellashinsky

Andrew, The second paragraph is what I needed to know. The real advantage would be write speed at a cost of more overhead. Thanks for saying it in plain English. Michael PS When I said 5 disks, I was referring to the example. When I re-read my post, I see I phrased it unclearly. But I still thank you for clear and simple explanations. You have a gift for teaching. Please keep using it.

andrew_miga
andrew_miga

Hello, It doesn't surprise me that you are a manager. The overhead refers to the extra storage needed for mirroring and not the perecent of usable storage. To mirror a 300 GB disk you would need another 300 GB disk for RAID 1 leading to 100% overhead for security purposes and 50% usability. In summary 100% overhead is 50% of total disk space usable for data.

Selena Frye
Selena Frye

You are correct that there is a problem between the text and figure. I apologize for the lag in getting the information verified. The text has now been amended to match up with the table in the figure. Thank you for pointing out the error to the author.

jpb
jpb

When you are dealing with professionals, you expect everyone to know the correct terms and definitions. This is a professional forum, not a consumer, prosumer, amateur, or otherwise non-professional forum. Overhead is what you consume over and above the usable amount, whether in retail, IT, or otherwise. It's not a matter of perspective or point of view. If he had said 50% of his total disk capacity was overhead, that would have been accurate, but that is a completely different statement from "his overhead is 50%", which would mean adding half the original amount of available space. I can see where you view overhead differently, being an Engineer, as overhead has a different meaning in Engineering where it is included as a part of the whole. This is not Engineering, it is IT.

sor
sor

Most people probably understand the basic math there, it's more about one's idea of 'overhead'. If I said, "you've got 300GB, but 150GB of that is overhead for RAID 1E", then you'd probably say that overhead is 50% as well. Generally I think that overhead is referred to as a part of the whole as my above example, but I could see someone thinking of it the other way too, i.e. "if I want 150GB of usable space, how much more will I need in overhead?".

Editor's Picks