Anticipated copying during idle time to increase i/o performance of snapshotted volumes
Original Publication Date: 2004-Aug-30
Included in the Prior Art Database: 2004-Aug-30
Disclosed is a method to increase the i/o performance of snapshotted volumes using anticipated copying. During idle time, blocks of a snapshotted volume are copied in anticipation to the corresponding snapshot(s). The blocks to be copied are chosen according to one of many possible schemes.
Anticipated copying during idle time to increase i Anticipated copying during idle time to increase iAnticipated copying during idle time to increase i Anticipated copying during idle time to increase i /
///o performance of snapshottedo performance of snapshottedo performance of snapshotted o performance of snapshotted
volumes volumesvolumes volumes
Summary of Invention Summary of InventionSummary of Invention Summary of Invention :
Anticipated copying during idle time can be used to increase the performance of snapshotted volumes by increasing the number of snapshotted blocks . The benefit of this method is that every write access to a block in the original volume that was copied in anticipation can change this block without causing additional i /o operations, which otherwise would decrease performance. The blocks to be copied are chosen according to one of many possible schemes .
Disclosed is a method to increase the i/o performance of snapshotted volumes using anticipated copying during idle time.
A volume is a storage space, handled by some kind of volume manager, e .g.
the i/o layer of a system. A snapshot is a logical point-in-time copy of a volume,
called original volume. When a block, i.e. a storage unit, is changed within the
original volume, the original block is copied to the associated snapshot (s) before
being changed, if it was not copied there before ("copy on write"; see figures 1 and
2). Thus a snapshot always represents the state of the original volume at the time the snapshot was created.
This method has a major problem: the performance of write operations to the original volume is degraded, as every write to a block that was not snapshotted before causes additional i/o operations to copy this block to the snapshot(s). If no special method is used, this problem even gets worse for a particular original volume the more snapshots are created of this volume . Summarizing:
- Writes to blocks that have not been snapshotted yet are slowed down . - Writes to blocks that have been snapshotted before are affected slightly (due to look-ups) or not at all.
The method disclosed herein circumvents this problem by increasing the number of snapshotted blocks. It anticipates copy-on-write operations and executes them during idle time, so i/o performance is not negatively impacted by doing so.
Anticipated copying is controlled by the volume manager. The vol...