System and method for I/O optimized data migration between high performance computing entities and a data storage supported by a de-clustered raid (DCR)architecture with vertical execution of I/O commands

摘要

A data storage system employs a de-clustered RAID (DCR) organization with a Vertical Rotation approach to prevent overflow of cache buffers. The Vertical Rotation Controller limits the size of the I/O operation relative to a respective Physical Disk Extent (PDE) to the cache buffer, and writes/reads a computed parity stripe matching the size of the cache, to form chunk bundles in the PDEs. When a bundle is filled in a first PDE residing at a first row, a PDE residing at a second row vertically displaced from the first row is engaged for forming another bundle. The process continues until all PDEs in a Vertical Rotation Group (VRG) are filled, or continues to a next VRG vertically displaced from a previous one to end I/O requesting.

主权项

1. A method for data migration in a data storage system, comprising the steps of:
(a) operatively coupling at least one data generating entity and a plurality of arrayed physical data storage devices, wherein each of said physical data storage devices includes an on-board cache buffer of a first predetermined size and a predetermined number of Physical Disk Extends (PDEs), each PDE constituting a data storage unit of a second predetermined size and having a physical address in relation to said each physical data storage device, wherein said second predetermined size exceeds said first predetermined size; (b) forming a de-clustered Redundant Array of Independent Disks (RAID) structure containing a plurality of PDEs arranged in a first number of horizontally arranged rows and a second number of vertically arranged columns; (c) issuing, by said at least one data generating entity, at least one respective I/O request for data migration between said at least one data generating entity and respective data storage devices, wherein said at least one respective I/O request includes data units and Logical Block Addresses (LBAs) of PDEs of interest; (d) computing at least one parity unit associated with said data units, and forming a parity stripe composed of said data units and said at least one parity unit; (e) in said de-clustered RAID (DCR) structure, formulating a collection of PDEs of interest in accordance with said LBAs identified in said I/O request; (f) striping said parity stripe throughout at least a portion of said PDEs in said collection thereof through a Vertical Rotation routine performed through the steps of: (g) computing, in said de-clustered RAID (DCR) structure, at least one Vertical Rotation Group (VRG) arranged with a predetermined number of rows from said first number of rows included in said DCR structure, (h) limiting an I/O operation relative each PDE in said collection thereof to said first predetermined size of said on-board cache buffer, (i) writing a respective stripe portion of said parity stripe into a respective PDE residing at one of said rows of said at least one Vertical Rotation Group, thus forming a first data chunk bundle of a size not exceeding the first predetermined size of said on-board cache buffer, and (j) switching to a next PDE residing at a next row vertically displaced from said row one along a respective column from said second number thereof in said at least one Vertical Rotation Group, and filling a second data chunk bundle in said next PDE residing at said next row with another portion of said parity stripe, wherein the size of said second data chunk bundle does not exceed the first predetermined size of said on-board cache buffer.