SYSTEM AND METHOD TO PRESERVE AND RECOVER UNWRITTEN DATA PRESENT IN DATA CACHE OF A DISK SUBSYSTEM ACROSS POWER OUTAGES

• US 20100174870A1
• Filed: 01/02/2009
• Published: 07/08/2010
• Est. Priority Date: 01/02/2009
• Status: Active Grant

• Current Assignee: Avago Technologies International Sales Pte Limited (Broadcom, Inc.)
• Original Assignee: LSI Corporation (Broadcom, Inc.)
• Inventors: BANERJEE, ARINDAM
• Granted Patent: US 8,347,041 B2
• Time in Patent Office: Days
• US Class Current: 711/143
• CPC Class Codes: G06F 1/30 Means for acting in the event of power-supply failure or interruption, e.g. power-supply fluctuations for resetting only G06F1/24
• CPC Class Codes: G06F 12/0866 for peripheral storage systems, e.g. disk cache
• CPC Class Codes: G06F 2212/1028 Power efficiency
• CPC Class Codes: G06F 2212/214 Solid state disk
• CPC Class Codes: Y02D 10/00 Energy efficient computing, e.g. low power processors, power management or thermal management

Abstract

Disclosed are a system and method to preserve and recover unwritten data present in data cache of a disk subsystem across power outages. In one embodiment, a method of a controller is described. The method includes applying a write-back technique between a host server and a data store, accessing a dirty data in a cache memory during a power outage. The method may apply an algorithm for efficiently offloading the dirty data to a non-volatile storage device during the power outage. In addition the method may apply the algorithm to efficiently transfer the dirty data from the non-volatile storage device to the data store when power is restored.

Claims

1. A method of a controller comprising:

applying a write-back technique between a host server and a data store;

accessing a dirty data in a cache memory during a power outage;

applying an algorithm for efficiently offloading the dirty data to a non-volatile storage device during the power outage; and

applying the algorithm to efficiently transfer the dirty data from the non-volatile storage device to the data store when power is restored.

2. The method of claim 1:

wherein the write-back technique is a transfer method in which an acknowledgement is sent to the host server from the controller when a data arrives from the host server to the cache memory.

3. The method of claim 2:

wherein the dirty data is an information arriving from the host server to the cache memory prior to transport to the data store;

wherein a parallel access of the non-volatile storage devices is ensured to improve overall backup performance when a backup time is limited;

wherein a maximum sequential access within the non-volatile storage device is ensured to increase Input Output (IO) throughput when the backup time is limited;

wherein a restore time is minimized to activate the write-back technique within a threshold amount of time after power is restored; and

wherein the data is packed and laid out in an order of a volume number and a logical block address.

4. The method of claim 1:

wherein the algorithm to access multiple non-volatile storage devices in parallel by creating a volume group or a disk group, and

wherein a data striping technique is used to achieve parallelism.

5. The method of claim 4 further comprising:

creating logical volumes that has a portion of a backup data on each of the storage devices constituting the volume group or the disk group during a backup operation.

6. The method of claim 1 further comprising:

packing cache blocks from different physical locations in the cache memory in a particular order based on both the volume number to which a cache block corresponds to and the logical block address within a volume; and

wherein the packing enables only the cache blocks which comprise the dirty data to be stored together in the non-volatile storage device without wasting storage space.

7. The method of claim 6 further comprising:

wherein the cache block is located in any physical location of the cache memory;

wherein the algorithm avoids storing a packing related information within the storage device so as to ensure maximum sequential access;

wherein by reading an entire metadata information from the non-volatile storage device the algorithm intelligently recreates the packing information every time there is an interruption during a backup and/or restore; and

wherein a packing information is critical to locate the respective cache block'"'"'s physical location within the non-volatile storage device.

8. The method of claim 7 further comprising:

wherein storing the packing information in the non-volatile storage device results in small random access to the non-volatile storage device.

9. The method of claim 7 further comprising:

wherein the algorithm to access the non-volatile storage device in a sequential manner to provide maximum throughput during the backup and restore operation.

10. The method of claim 7 further comprising:

wherein a metadata information is also stored in logical volumes created during the backup operation on the non-volatile storage device thereby ensuring minimum read times; and

wherein the metadata information maintains a state of the dirty cache blocks within the cache memory when the power outage first occurred.

11. The method of claim 1:

wherein the algorithm to restore the dirty data in a context of a cache sync operation to ensure that the power on to a host IO access time is similar to a normal boot time of the controller;

wherein the cache sync operation is performed after the controller initialization; and

wherein the data restoration post system initialization avoids effect to the power on to a host IO response time.

12. The method of claim 11 further comprising:

handling efficiently an interrupted write data operation during a system initialization phase without affecting the power on to host IO response time by ensuring the cache blocks corresponding to interrupted writes are laid out contiguously within the non-volatile storage device;

wherein the cache blocks corresponding to interrupted writes can be located in any physical location of the cache memory;

wherein storing the cache blocks corresponding to interrupted writes contiguously within the non-volatile storage devices enable all such cache blocks to be restored through one or more sequential read operations;

wherein an interrupted write is an unfinished in-flight write operation that was in progress between the cache memory and the data store when the power outage occurred; and

wherein interrupted writes need to be completed at system initialization to maintain Redundant Array of Independent Disk (RAID) data parity consistency.

13. The method of claim 1:

wherein the algorithm allows restoration of the dirty data in the context of a host server initiated IO;

wherein the host server initiated IO is another data transport which arrives from the host server and is targeted to the same logical unit within the data store as the previous dirty data which was stored in the non-volatile storage device during backup; and

wherein the host server initiated IO is allowed only post system initialization.

14. The method of claim 1:

wherein the algorithm to provide a minimal restore time by using an efficient data layout within the non-volatile storage devices;

wherein the efficient data layout is achieved by ordering the cache blocks in the non-volatile storage device sorted on their volume number and the logical block address within the volume thereby synchronizing the data layout with the cache sync operation algorithm;

wherein synchronizing the data layout with a cache sync operation enables sequential access to the non-volatile storage device during data restore; and

wherein the algorithm also allows sequential access to the data store during a restore operation thereby minimizing expensive seek operations on the data store.

15. The method of claim 1:

wherein the algorithm to invalidate the dirty data which is read from the non-volatile storage device when the power is restored; and

wherein an invalidation is achieved through marking a bit in a bitmap table and mapping the bit to the corresponding cache block related metadata information.

16. The method of claim 15:

wherein the invalidation requires updating specific fields within the metadata information stored within the non-volatile storage devices;

wherein updating specific fields of the metadata information within the non-volatile storage devices results in an expensive read-write-modify operations on the storage device; and

wherein a bitmap technique enables avoiding the expensive read-modify-write operation.

17. A system comprising:

a host server;

a data store;

a cache memory which comprise a dirty data;

a non-volatile storage device to store the dirty data during a power outage; and

a controller coupled between the host server and the data store to apply optimal algorithm to access the dirty data in the cache memory during the power outage and to offload the dirty data to the non-volatile storage device and to apply an algorithm to optimally transfer the dirty data from the non-volatile storage device to the data store when power is restored.

18. The system of claim 17:

wherein a write-back technique is a transfer method in which an acknowledgement is sent to the host server from the controller when the dirty data arrives from the host server to the cache memory.

19. A method comprising:

accessing a dirty data block optimally in a cache memory during a power outage;

storing the dirty data in a non-volatile storage device during the power outage; and

transferring the dirty data from the non-volatile storage device to a data store when power is restored.

20. The method of claim 19 further comprising:

applying a write-back technique between a host server and the data store.