Growing and splitting a disk array by moving RAID group members
In a storage system that implements RAID (D+P) protection groups a drive subset initially has (D+P) drives with (D+P) partitions. The drive subset is made to be symmetrical such that protection group members are symmetrically distributed in a matrix of drive rows and partition columns that represents the drive subset. A single new drive is added by partitioning the new drive with (D+P) partitions, moving existing protection group members from a selected partition of the (D+P) drives to partitions of the single new drive by rotating the selected partition by 90 degrees, and adding protection group members of a new protection group to the vacated selected partition of the (D+P) drives. The process is repeated until (D+P) new drives have been added in single drive increments. The resulting drive subset is then split by forming a non-symmetric drive subset from the (D+P) drives and forming a symmetric drive subset from the (D+P) new drives.
1. A method, comprising:
for a plurality of protection groups using data (D) members and parity (P) members, creating a drive subset using (D+P) drives with (D+P) partitions such that protection group members are distributed in a manner represented by a symmetric matrix of drive rows and partition columns that represents the drive subset and wherein the matrix is symmetric such that protection group distribution in a transpose of the symmetric matrix created by interchanging row and column indices matches protection group distribution in the symmetric matrix;
adding, in increments of a single new drive until there are at least (D+P) new drives, new drives having (D+P) partitions to the drive subset by:
moving existing protection group members from a selected partition of the (D+P) drives to partitions of the single new drive by relocating each existing protection group member at drive Dn of the selected partition of the (D+P) drives to partition Pn of the new drive; and
adding protection group members of a new protection group to the selected partition of the (D+P) drives; and
splitting the drive subset into two drive subsets when the drive subset includes at least (D+P) new drives, wherein at least one of the two drive subsets is characterized by protection group member distribution in a manner represented by the symmetric matrix.
2. The method of claim 1 wherein moving existing protection group members from the selected partition of the (D+P) drives to partitions of the single new drive comprises starting with a first or last partition and continuing to an adjacent partition that has not been moved with addition of each successive new drive.
3. The method of claim 1 wherein moving existing protection group members from the selected partition of the (D+P) drives to partitions of the single new drive comprises rotating the selected partition by 90 degrees.
4. The method of claim 1 wherein splitting the drive subset into two drive subsets comprises forming a non-symmetric drive subset from the (D+P) drives.
5. The method of claim 1 wherein splitting the drive subset into two drive subsets comprises forming a symmetric drive subset from the (D+P) new drives.
6. An apparatus, comprising:
a plurality of interconnected compute nodes that manage access to a plurality of drives; and
a drive manager that:
for a plurality of protection groups using data (D) members and parity (P) members, creates a drive subset using (D+P) drives with (D+P) partitions such that protection group members are distributed in a manner represented by a symmetric matrix of drive rows and partition columns that represents the drive subset and wherein the matrix is symmetric such that protection group distribution in a transpose of the matrix created by interchanging row and column indices matches protection group distribution in the matrix, adds, in increments of a single new drive until there are at least (D+P) new drives, new drives having (D+P) partitions to the drive subset;
moves existing protection group members from a selected partition of the (D+P) drives to partitions of the single new drive by relocating each existing protection group member at drive Dn of the selected partition of the (D+P) drives to partition Pn of the new drive;
adds protection group members of a new protection group to the selected partition of the (D+P) drives; and
splits the drive subset into two drive subsets when the drive subset includes at least (D+P) new drives, wherein at least one of the two drive subsets is characterized by protection group member distribution in a manner represented by the symmetric matrix.
7. The apparatus of claim 6 wherein the drive manager moves existing protection group members from the selected partition of the (D+P) drives to partitions of the single new drive starting with a first or last partition and continuing to an adjacent partition that has not been moved with addition of each successive new drive.
8. The apparatus of claim 6 wherein the drive manager moves existing protection group members from the selected partition of the (D+P) drives to partitions of the single new drive by rotating the selected partition by 90 degrees.
9. The apparatus of claim 6 wherein the drive manager splits the drive subset into two drive subsets by forming a non-symmetric drive subset from the (D+P) drives.
10. The apparatus of claim 6 wherein the drive manager splits the drive subset into two drive subsets by forming a symmetric drive subset from the (D+P) new drives.
11. A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for using a computer system to scale a storage system in a single drive increment, the method comprising:
for a plurality of protection groups using data (D) members and parity (P) members, creating a drive subset using (D+P) drives with (D+P) partitions such that protection group members are distributed in a manner represented by a symmetric matrix of drive rows and partition columns that represents the drive subset and wherein the matrix is symmetric such that protection group distribution in a transpose of the matrix created by interchanging row and column indices matches protection group distribution in the matrix; and
adding, in increments of a single new drive until there are at least (D+P) new drives, new drives having (D+P) partitions to the drive subset by:
moving existing protection group members from a selected partition of the (D+P) drives to partitions of the single new drive by relocating each existing protection group member at drive Dn of the selected partition of the (D+P) drives to partition Pn of the new drive; and
adding protection group members of a new protection group to the selected partition of the (D+P) drives; and
splitting the drive subset into two drive subsets when the drive subset includes at least (D+P) new drives, wherein at least one of the two drive subsets is characterized by protection group member distribution in a manner represented by the symmetric matrix.
12. The non-transitory computer-readable storage medium of claim 11 wherein the method further comprises moving existing protection group members from the selected partition of the (D+P) drives to partitions of the single new drive comprises starting with a first or last partition and continuing to an adjacent partition that has not been moved with addition of each successive new drive.
13. The non-transitory computer-readable storage medium of claim 11 wherein the method further comprises moving existing protection group members from the selected partition of the (D+P) drives to partitions of the single new drive comprises rotating the selected partition by 90 degrees.
14. The non-transitory computer-readable storage medium of claim 11 wherein the method further comprises splitting the drive subset into two drive subsets comprises forming a symmetric drive subset from the (D+P) new drives.