The number of columns specifies how many physical disk data is striped across and has an impact on performance regardless of stripe and block size. Storage Spaces intelligently scales the column count up to eight by default, but you can adjust this parameter by using Windows PowerShell.

One example of how the number of columns affects performance is as follows. Assume that you create a simple (no resiliency) storage space in a storage pool with 12 disks, and that each disk is capable of 150 MB/s sequential throughput. Creating a simple space with one column yields a maximum throughput of 150 MB/s (the concept of spanning creates exceptions – see next section). Using two columns increases this to almost 300 MB/s, with further increases scaling linearly (for example, four columns yields close to 600 MB/s throughput). The more columns a storage space is assigned, the more disks can be striped across and the higher performance.

If you choose a high column count storage space and you intend to expand the capacity of the storage space in the future, you should add at least as many disks as the storage space has columns (in case of a simple or parity space), or columns times number of data copies (in the case of a mirrored space). For instance, when deploying a two-way mirrored space with four columns, you should have eight disks with available capacity (two columns times two data copies), and add disks in groups of eight when expanding the storage pool.

The following table provides real-world performance numbers comparing two simple spaces with different numbers of columns. In both conditions, the storage pool contained four SSDs, and each storage space received a sequential read request of 1 MB I/Os, akin to reading a large number of 1 MB files.

Disks

Total Capacity (TB)

Columns

Throughput (MB/s)

Avg. Latency (ms)

4

1.45

2

902.45

8

4

1.45

4

1622.50

4


The data shows that increasing the number of columns can significantly improve performance for sequential workloads, though at the expense of some flexibility when expanding capacity (you should add disks in groupings that match the number of columns x data copies). Random workloads do not experience as significant a performance increase, exhibiting more uniform performance across different column counts.

Another factor that ties in directly with the column count of the space is the amount of outstanding I/Os or data that is to be read from or written to the storage space. A large number of columns benefits applications that generate enough load to saturated multiple disks, but introduce unnecessary limitation on capacity expansion for less demanding applications.

The following example describes this trade-off: A typical hard disk is saturated writing a large number of 1MB files or block data in a sequential fashion. To saturate a simple space with four hard disks and the same workload, it is necessary to provide roughly 32 outstanding I/Os of this type. If less data is outstanding, not all disks are fully utilized. If the application using this storage space will never have more than 16 I/Os outstanding at a time, creating a storage space with two manually assigned disks could service this workload and preserve the other two disks for capacity expansion or for use with another  application (using its own manually allocated storage space). This also has the advantage that the administrator can add two disks to extend the storage space as opposed to four.

To adjust the number of columns used by storage spaces, you must use Windows PowerShell. For more information, see:http://go.microsoft.com/fwlink/p/?LinkId=260763 

Ref: http://social.technet.microsoft.com/wiki/contents/articles/15200.storage-spaces-designing-for-performance.aspx