Hewlett-Packard Development Company, L.p.

Sun Microsystems, Inc.

Lawrence Michael Ruane

Mark G Hayden

One embodiment of the present invention is directed to a multi-node data-storage system comprising a number C of component-data-storage-system nodes that store data pages of a data object in a mirror and in 0, 1, or more parity-encoded snapshots, the mirror comprising 2 or more replicates including a 0replicate and each parity-encoded snapshot containing 1 or more stripes, each stripe including snapshot data pages and snapshot parity pages. The multi-node data-storage system features a snapshot operation that generates a next parity-encoded snapshot and transfers data pages of the data object from the mirror to the next-generated parity-encoded snapshot, the snapshot operation distributing each snapshot data page to the node that contained the corresponding 0-replicate data page, distributing successive snapshot data pages to successive nodes in a cyclic order with respect to the C nodes, and distributing snapshot data and parity pages to the C nodes so that the data and parity pages of each stripe reside on different nodes and so that the data and parity pages are evenly balanced among the C nodes following distribution of a threshold number of snapshot pages.

Method and system for parity-page distribution among nodes of a multi-node data-storage system

A logical address and a pointer entry for a file in an indirect address file system are translated into a physical address. A decision module tests a pointer flag in a present pointer entry. The pointer entry has a pointer and a pointer flag to identify whether the pointer points to a data storage area or a metadata storage area. The decision module indicates whether the pointer is a data pointer or a metadata pointer. In response to the decision module indicating the pointer is a data pointer, a set module combines the data pointer with the logical address to generate a physical address. A split module, in response to the decision module indicating the pointer is a metadata pointer, divides the logical address into a first portion as an index value and a remaining portion as an offset value. An update module then sets the logical address to the offset value. A retrieve module combines the metadata pointer with the index value to get the next pointer entry. Each metadata pointer points to a next successive level of indirect addressing in the plurality of levels of indirect addressing. Also, each data pointer points directly to the data storage area. As a result all levels of indirect addressing, which are from a level having the pointer entry with a data pointer to a last indirect addressing level, are by-passed.

Growing community of inventors

Parker, CO, United States of America

Lawrence Michael Ruane