VM_Yamato
Enthusiast
Enthusiast

Why we can’t use RAID 5 policy on 3 nodes vSAN?

Jump to solution

Normally, from hardware RAID view point, we can make RAID 5 virtual disk using three drives.

but on vSAN, at least we need 4 nodes on it.

could you advise me?

Yamato Sakai Technical Training Instructor | Dell Technologies Education Services
1 Solution

Accepted Solutions
jameseydoyle
VMware Employee
VMware Employee

As a matter of fact, the RAID5 configuration in vSAN does not use a Witness component as it is not required. But what you say about RAID5 being different when using object based storage over real storage is 100% correct.

In vSAN, we are implementing a form of RAID, but it cannot be called RAID really, as we are not working with disks (the D in RAID). We are working with objects and components of those objects. A more correct term would be RAIN (or Redundant Array of Independent Nodes). In the vSAN implementation of RAID5, we create 4 components of the parent vDisk object. The data is spread across these components in a manner similar to that of RAID5 in the disk world. To guarantee redundancy, each of the components must be placed on a different host, so a single host failure would take more than one component offline.

A witness is not required because RAID5 only allows for 1 failure. As there are 4 components, we would always need to have 3 out of the 4 components available. If we had a split-brain, then each partition would only have 2 out of 4 components, making the data useless in both partitions.

View solution in original post

4 Replies
daphnissov
Immortal
Immortal

4 nodes, but all-flash nodes as well. vSAN is not RAID in that it is object-based storage, so a witness is still necessary. In RAID erasure coding systems where you're dealing with simply blocks, that's not necessary. There's a really good article on some of this written by Christos Karamanolis last February available here for more information.

VM_Yamato
Enthusiast
Enthusiast

Hi daphnissov,

Thanks give me message. of course, I already know basic knowledge of vSAN Erasure coding. ( AF only, need advanced and above license, need at least 4 nodes, 10Gbps network must)

My question is, why we can't use Erasure coding with 3 nodes vSAN cluster. because for example Dell PowerEdge RAID controller H730, H710, PERC 6, 5 and more....not only Dell product, but also other hardware vendor Hardware RAID controller, we can use RAID 5 using 3 drives, it is very well know rule for RAID 5.

RAID - Wikipedia 

Yamato Sakai Technical Training Instructor | Dell Technologies Education Services
0 Kudos
daphnissov
Immortal
Immortal

You're trying to compare an object-based storage system with a block storage system. They aren't the same thing. vSAN erasure coding isn't really RAID, but it behaves in an equivalent manner. In a RAID subsystem implementing level 5 or 6, there is distributed parity just as there is in vSAN, but there is no witness component. The two implementations of technology providing somewhat equivalent outcomes is radically different.

jameseydoyle
VMware Employee
VMware Employee

As a matter of fact, the RAID5 configuration in vSAN does not use a Witness component as it is not required. But what you say about RAID5 being different when using object based storage over real storage is 100% correct.

In vSAN, we are implementing a form of RAID, but it cannot be called RAID really, as we are not working with disks (the D in RAID). We are working with objects and components of those objects. A more correct term would be RAIN (or Redundant Array of Independent Nodes). In the vSAN implementation of RAID5, we create 4 components of the parent vDisk object. The data is spread across these components in a manner similar to that of RAID5 in the disk world. To guarantee redundancy, each of the components must be placed on a different host, so a single host failure would take more than one component offline.

A witness is not required because RAID5 only allows for 1 failure. As there are 4 components, we would always need to have 3 out of the 4 components available. If we had a split-brain, then each partition would only have 2 out of 4 components, making the data useless in both partitions.

View solution in original post