I'm preparing a Dell R520 PowerEdge server with 8x 1TB disks as a data archive/warehouse (file, RDBMS, NoSQL) server. Out of the box, the PERC H710P RAID controller was configured with 3x 2TB RAID6 virtual disks. My understanding is that the 2TB limit is something to do with Windows "basic disks" (http://technet.microsoft.com/en-us/library/cc773268.aspx).
I'll be installing either Debian 7.4 or Ubuntu 12.04 LTS as the OS. For simplicity's sake I would like to create a ~5TB partition for data storage.
Is it advisable to create an LVM partition of that size that spans the three virtual disks? Does that lessen the effective 2-disk tolerance that I get from RAID6? Or are there performance issues when spanning LVM over multiple disks like that?
Or, would it be better to delete and recreate the virtual disks as one large 6TB RAID6 virtual disk? Is Debian or Ubuntu also limited to 2TB? I believe that used to be the case. Is it still?
The 2 TB limit is not so much a Windows or Linux limitation, it is mostly a limitation of the partition table counters. The old-style MBR partition table uses 32-bit counters to address sectors and specify the length of a partition. This would allow to specify partitions of
512 * (2^32)
Bytes in size, which is the 2 TB limit.Of course, you might hit other limits - old RAID controllers would not allow you to create volumes larger than 2 TB, virtualization solutions might be restricted to 2 TB per virtual disk, but if you do not have any of these, I would strongly recommend simply creating a larger logical disk and partitioning it using GPT, which does not have the aforementioned limit. Virtually all modern OSes do support GPT, although some tools might not (like Linux
fdisk
).Whether to use LVM is entirely up to your own taste and is probably largely independent of the logical drive size in your case. It also would not affect availability if all your LDs would span over the same set of disks. LVM has the specific disadvantage of adding to complexity should you ever decide to grow your array (e.g. by swapping disks for a higher-capacity ones and resizing using the Online Capacity Expansion feature of your controller). But as it also is an additional storage abstraction layer, it comes with a set of nice features you might need one day.