>Expand my RAID-6 array on 12 additional drives Or
>create another RAID-6 array from 12 new drives, and then use LVM to expand the original volume.
In the first case, I will face a reconstruction time of about 4-6 weeks , And then the extra capacity can be used. Also, if I need to add another shelf to this array later and use the same method, the reconstruction time may be longer.
In the second case, I am losing some capacity and potential performance, but it seems easier to handle and the available time is easier to predict.
I also plan to increase this array to 7-10 shelves in the next 2-3 years, The entire array will contain a huge volume formatted with XFS. So, anyone can tell from their experience and elaborate on the pros and cons of each method and the potential problems I will face over time. /p>
Thank you
As a general rule, you should try to keep the RAID6 LUN not too large in terms of the number of disks-when you are facing terribly long rebuild times, it does not make any sense to use a RAID LUN with >> 20 disks. Large storage vendors (EMC, NetApp, etc.) limit their RAID group size to about 20 spindles (depending on the disk model). But these people know their RAID controllers, characteristics and rebuild times very well, and their RAID implementation is not your mill md( 1) RAID or cheap PCI-E HBA. However, when you expand and add storage as needed, it makes sense to use LVM to strip/connect RAID6 LUNs. This is exactly the use case for volume managers.
Since you did not explain anything about your IO model and requirements, I cannot comment on performance.
I am building a high-level video Capacity storage. It is currently at the 20TB mark and continues to increase Long. Currently, I have a 12-bay chassis with 2TB hard drives and configured as RAID-6. I am considering buying another 12-bay chassis and filling it with 2TB drives, I am considering the following two options:
>Expand my RAID-6 array on 12 additional drives or
>create another RAID-6 array from 12 new drives, then use LVM to expand the original volume.
p>
In the first case, I will face a rebuild time of about 4-6 weeks before I can use the extra capacity. Also, if I need to add another shelf to this array later and use the same method , The reconstruction time may be longer.
In the second case, I am losing some capacity and potential performance, but it seems easier to handle and the available time is easier to predict.
I also plan to increase this array to 7-10 shelves in the next 2-3 years. The entire array will contain a huge volume formatted with XFS. So, anyone can tell from their experience, And elaborate on the pros and cons of each method, as well as the potential problems that I will face over time.
Thank you
Need to consider Many things to solve the problem. Since you did not mention any specific requirements (throughput, IOPS/latency, RAID reconstruction time-just to name a few important parameters), I just answer about the size of the shelf/LVM/RAID group .
As a general rule, you should try to keep the RAID6 LUN not too large in terms of the number of disks-when you are facing a terribly long reconstruction time, use a RAID LUN with >> 20 disks It doesn’t make any sense. Large storage vendors (EMC, NetApp, etc.) limit their RAID group size to about 20 spindles (depending on the disk model). But these people know their RAID controllers, characteristics, and rebuild times very well, and they The RAID implementation is not your mill md(1) RAID or cheap PCI-E HBA. However, when you expand and add storage as needed, it makes sense to use LVM to strip/connect RAID6 LUNs. This is exactly the use case for Volume Manager.
Since you did not explain anything about your IO patterns and requirements, I cannot comment on the performance aspect.