Write-Back Storage Cache Based On Fast Persistent Memory

• US 20130086330A1
• Filed: 11/03/2011
• Published: 04/04/2013
• Est. Priority Date: 09/30/2011
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: HU, WEI-MING, BADDEPUDI, BHARAT CHANDRA, ERLIKHMAN, BORIS, LEE, J. WILLIAM, SHI, JIA, GOYAL, KIRAN B., LOAIZA, JUAN R., TSUKERMAN, ALEX, UMAMAGESWARAN, KOTHANDA, AYA, SELCUK
• Granted Patent: US 10,592,416 B2
• Time in Patent Office: Days
• US Class Current: 711/143
• CPC Class Codes: G06F 12/08 in hierarchically structured memory systems, e.g. virtual memory systems
• CPC Class Codes: G06F 12/0868 Data transfer between cache memory and other subsystems, e.g. storage devices or host systems
• CPC Class Codes: G06F 2212/502 using adaptive policy

Abstract

A storage device uses non-volatile memory devices for caching. The storage device operates in a mode referred to herein as write-back mode. In write-back mode, a storage device responds to a request to write data by persistently writing the data to a cache in a non-volatile memory device and acknowledges to the requestor that the data is written persistently in the storage device. The acknowledgement is sent without necessarily having written the data that was requested to be written to primary storage. Instead, the data is written to primary storage later.

Claims

1. A method, comprising steps of:

a storage device receiving from a storage device client a data block write request to write a data block to said storage device;

wherein said storage device comprises one or more primary storage devices and one or more persistent cache devices;

storing, to a persistent cache device of said one or more persistent cache devices, said data block as a cache copy of the data block version stored in said one or more primary storage devices;

in response to storing said data block to said persistent cache device, transmitting an acknowledgement to the storage device client to acknowledge commitment of the write of said data block; and

after transmitting said acknowledgment, said storage device storing said data block to said one or more primary storage devices.

2. The method of claim 1, the steps further including:

after said storing said data block as a cache copy and prior to transmitting said acknowledgement, updating persistent cache metadata stored in said one or more persistent cache devices to;

map a cache block address for said data block to a primary block address for said data block;

store in association with said cache copy a dirty flag to specify that said cache copy is dirty;

store in association with said cache copy a block check value computed from the contents of said cache copy.

3. The method of claim 2, the steps further including:

after transmitting said acknowledgment and before storing said data block to said one or more primary storage device, said storage device receiving from a storage device client a another data block write request to write said data block to said storage device, said request specifying data block content for said data block;

overwriting said cache copy with said data block content;

in response to overwriting said cache copy, transmitting another acknowledgement to the storage device client to acknowledge commitment of said data block; and

wherein transmitting said another acknowledgement is performed without updating said persistent cache metadata to reflect the overwriting of said cache copy.

4. The method of claim 1, wherein said one or more persistent cache devices are flash memory devices and said one or more primary storage devices are disk drive devices.

5. The method of claim 1, the steps further including:

while said storage device operates in write-back mode, monitoring the write operation load of said one or more primary storage devices; and

based on said write operation load, determining to transition from write-back mode to write through mode.

6. The method of claim 1, the steps further including:

said storage device receiving from storage device clients a plurality data block write requests to write a respective data block to said storage device, each particular data block write request of said plurality of block write requests requesting to write a different data block;

for each particular data block write request of said plurality of block write requests, storing, to said persistent cache device of said one or more persistent cache devices, the respective data block as a cache copy of the respective data block version stored in said one or more primary storage devices;

performing one write to a data block storing persistent cache metadata to update said persistent cache metadata to reflect the respective data block writes for all of the plurality of data block write requests;

only after performing said one write to a data block storing said persistent cache metadata, for each particular data block write request of said plurality of block write requests, transmitting an acknowledgement to the respective storage device client to acknowledge commitment of the respective data block.

7. The method of claim 1, further comprising steps of:

said storage device receiving from a storage device client a data block request to read another data block stored in said one or more primary storage devices;

storing, to a persistent cache device of said one or more persistent cache devices, said other data block as a cache copy of the other data block; and

updating persistent cache metadata stored in said one or more persistent cache devices to;

map a cache block address for said data block to a primary block address for said data block; and

store in association with said cache copy a block check value computed from the contents of said cache copy.

8. A method, comprising steps of:

a storage device receiving from a storage device client a data block write request to write a data block to said storage device;

wherein said storage device comprises one or more primary storage devices and one or more persistent cache devices;

selecting a victim cache copy stored in said one or more persistent cache devices to which to replace with said data block;

wherein said persistent cache metadata;

maps the cache block address of said victim to the primary block address of the data block of which said victim is a cache copy;

contains a dirty flag that specifies that said victim is clean;

associates with said victim a block check value;

overwriting said victim in said persistent cache device with the contents of said data block;

encountering a storage device failure before updating said persistent cache metadata to reflect said overwriting of said victim; and

said storage device performing recovery, wherein performing recovery includes;

generating a block check value based on said content of the data block stored in place of said victim in said persistent storage device,

making a comparison between (1) the block check value generated based on the content of the data block stored in place of said victim and (2) the block check value associated with said victim by said persistent cache metadata, and

based on the comparison, detecting that the persistent cache metadata contains a false dirty flag that species that the data block stored in place of said victim is clean.

9. A method, comprising steps of:

a storage device receiving from a storage device client a data block write request to write a data block to said storage device;

wherein said storage device comprises one or more primary storage devices and one or more persistent cache devices;

determining that a cache copy of said data block is stored in a persistent cache device of said one or more persistent cache devices;

wherein said persistent cache metadata;

maps the cache block address of said cache copy to the primary block address of a data block of which said cache copy is a copy,

contains a dirty flag that specifies that said cache copy is dirty, and

associates with said victim a block check value;

overwriting said cache copy in said persistent cache device with the contents of said data block specified in said data block request;

transmitting an acknowledgement to the storage device client to acknowledge commitment of the write of said data block; and

said storage device performing recovery, wherein performing recovery includes determining that the overwritten version of said cache copy is dirty based on said dirty flag.