Link to home
Start Free TrialLog in
Avatar of Muhammad Asif
Muhammad Asif

asked on

Raid 0 with One Disk Using RaidController

I am looking into Dell architecture for Exchange 2016 implementation using Dell servers. In storage section they are saying " Fourteen independent Raid 0 disks can be used to store copies of Exchange databases ( A Raid Controller Can be used to create independent Raid 0 volumes by using internal drives. ) " 


Are they saying to create 14 Raid and each Raid 0 with single disk?  Can any one elaborate this please?
Avatar of Ravi Kumar Atrey
Ravi Kumar Atrey
Flag of India image

It is self explanatory "Fourteen independent Raid 0 disks"

As minimum 2 disks are required for RAID 0 so minimum 28 disks would require but they would all be separately created in RAID 0 in pair of 2.
 
1. 2 disk in RAID 0
2 2 disk in RAID 0

and so on.

There would be 14 RAID 0 disk. Each RAID 0 disk will be a separate disk so.

This is what I understood.
Avatar of Muhammad Asif
Muhammad Asif

ASKER

Actually in Storage Architecture, Dell servers have only 14 drives for Exchange DB and Logs. So I am quite sure they are referring to create 14 Raids ( Raid0) each with only one Disk using Raid Controller.  Is there any advantage or benefits to create Raid0 with one disk using Raid Controller?
Raid0 is only for situations where data loss isn't critical but speed is...so for Exchange what you need is Raid 10 or Raid 6...everything else is slow or with limited redundancy..
Hi John,

From above mentioned dell statement what you understand? Does they mean 14 independent Raid 0 groups each with one group or they mean to one Raid0 group with 14 drives?
Normally when we talk about Raid...we also have a minimum of drives according to Raid Specs...
Raid 0 --> 2 drives
Raid 1 --2 drives
Raid 5  --> 3 Drives
Raid 10 -->4 Drives
For some bizzare reason which i have never looked up ...when you have a single HDD (1) and you want to connect it to a Raid controller its only accepted when you "construct" a Raid 0....its a bit obscure to say the least but as i said this is how it works....i always had storage with multiple drives and some cases with spare drives and this 1 HDD <--> Raid  0 i have seen it in some testing cases...
Probably its a good idea to share with us the paper from Dell to get a 2nd look...
Do remember that if in Raid0 1 drive goes bad you simply loose everything....so its only good for temporary files as i mentioned before and when speed matters the most...
But if your up to speed then its better to invest on SSD...
Do remember that if in Raid0 1 drive goes bad you simply loose everything.

Yes, you are right. But In High availability mode where you have at least 3 copies of one Mailbox database then Microsoft recommends to JBOD etc. to reduce the cost. As database would be available on other server in case of disk failure where database reside.  For your reference I have attached the Exchange 2016 Reference document by Dell.

You may go through the page 38 related to storage architecture.
Design-Guide-for-Microsoft-Exchange-.pdf
Well from the paper its crystal clear the it uses 14 HDDS in 14 Raid 0..
Quoting from the paper:
When a PowerEdge R730xd server is configured with 16 LFF drives, fourteen independent RAID0 disks can
be used to store copies of Exchange databases. These RAID0 disks can be created by using
12 disks in the
front bay and 2 disks in the internal drive tray
.
I guess the confusion came from the Figure 3 where it mentions:
2 x Exchange DB RAID0 LUN
They have mentioned that 12 drives would be used from front bay and those drives are also used which are located in server at internal tray bay. So In total 14 LFF drives Drives would be used for Exchange DB and Logs.  Two drives would be used for OS which would be configured with Raid 1.

The confusion is regarding the configuration of 14 drives with Raid for Exchange DB and Logs.  Is they trying to say that we would need to configure 14 drives with 14 Raid 0 group  and each raid 0 would have only one disk?
This what they say based on the word "independent " at least to my understanding....
On the practical side ...if you really going to deploy such a demanding infrastructure Dell as any other vendor would be more than pleased to help properly implement it...that's for sure...but (there is always a but....i don't know why)....try to get a "demo" on this...a similar infrastructure (client case) ...because to my bitter experience the engineer that you would get assigned maybe will operate more with a commercial view on his/her mind than on the technical....
Even if you have the slightest doubt especially on the performance don't hesitate to put everything on hold and reconsider....
So you are doing a good job on trying to clarify every little detail because its a wild world out there and after the buying procedure has closed there is no turning back...
@ john Thanks for your details.  I have posted the same question on  Dell forum and are awaiting for their response.  I hope they will post some satisfied answer. However, if anyone else have implemented the same kind of infrastructure then please share views.
To be blunt, all the previous answers are 100% WRONG!  The experts give good information but they are dead wrong on what the documentation is saying.  (I have a developer NDA with the vendor who makes the controller, so trust me here  ...)

What they are saying is no more or no less than saying you have the option to configure one or more drives so that it appears as a single disk EMULATED drive.  

This would make sense (14 individual disks) if you wanted the O/S to do  software RAID.  (Like Solaris + ZFS or LINUX + md device driver).  You might want to do hardware mirrored RAID1 for boot disk then do software raid for the 12 as another example

Do not read anything else in that statement about why you may want to do that.  Not all Dell controllers allow you to do this, which is why they mentioned it.  Some controllers do not present the disk as a RAID0, but a standard pass-through, unmodified drive.  

Note also that the "RAID0" disk is not the raw drive.  That controller EMULATES the drive.   All I/O goes through the RAID controller, and the controller puts metadata on the disk..  A few MB of metadata goes on the beginning of the disk, and the controller gives emulated drive that is a few MB less than the physical disk.   N blocks of metadata means block X of the logical disk is block N + X of physical disk.  IF capacity of disk is Y blocks then LOGICAL block 37 of the RAID0 = PHYSICAL block N + 37 of the disk.  That means addressable capacity of the RAID0 drive is Y - N blocks.    

If there are 100 blocks of metadata, then think of it as everything is shifted 100 blocks.   If that disk is formatted as a boot drive, and you take it away from controller then it will not boot w/o controller because everything is shifted.

Hope that explains it.   Again, this is not a RAID why you might want to do a certain raid level philosophical thing. the manual is just saying you are not dealing with a raw pass-through drive, you are dealing with an emulated drive with contents shifted over.  (All transparent if this disk will ALWAYS be used with this particular controller make/model)
Personally i think i am totally lost...good info about  inner working of controllers...but none here mentioned the "single disk EMULATED drive"
ASKER CERTIFIED SOLUTION
Avatar of David
David
Flag of United States of America image

Link to home
membership
This solution is only available to members.
To access this solution, you must be a member of Experts Exchange.
Start Free Trial
So if i got it right you mean that the controll takes the drives and presents them up to 14 raid0 fake drives like the Luns we get from SANs
Close, Tom ... but it is not fake RAID. The controller adds buffering/caching at expense of a few MB.   How about this definition:  This controller is just like any other decent RAID controller, but it simply sets the MINIMUM number of disks of a RAID0 to 1 rather than 2.

You can do software RAID1 or any other operating system-based virtualization with any RAID group on any controller, even fake-raid controllers,  so really the only config difference is a single-disk RAID0.
Thanks David. I have got the correct understanding now. You have provided awesome knowledge on this topic.
This looks like Dell using confusing terms again.

Try looking up RAID10 in the Dell documentation - you will find at least one reference to it being a mirror of two stripes, exactly the reason the industry tried to switch to a single definition after the RAID0+1, RAID1+0 contradiction! (that was Dell again)
@Gerald -- Storage metrics have become hopelessly munged ever since a marketeer at (now long gone) Micron decided they could win some bids on price per MB by redefining 1MB as 1,000,000 bytes instead of 1024x1024 bytes.      

Besides, we both know you can create a perfectly valid RAID1+0 with only a single disk drive under LINUX, or a RAID1, or a RAID10, or all of these combinations on the same drive at the same time ;)