Transactional commits with hardware assists in remote memory

• US 10,140,149 B1
• Filed: 05/19/2015
• Issued: 11/27/2018
• Est. Priority Date: 05/19/2015
• Status: Active Grant

• Current Assignee: Pure Storage, Inc.
• Original Assignee: Pure Storage, Inc.
• Inventors: Hayes, John, Gold, Brian, Gupta, Shantanu, Lee, Robert, Kannan, Hari
• Primary Examiner(s): Daye, Chelcie
• Application Number:
  US14/716,543
• Time in Patent Office: 1,288 Days
• Field of Search: 707615
• CPC Class Codes: G06F 12/0802 Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
• 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 13/16 for access to memory bus G06F13/28 takes precedence
• CPC Class Codes: G06F 16/10 File systems; File servers
• CPC Class Codes: G06F 2212/1032 Reliability improvement, data loss prevention, degraded operation etc
• CPC Class Codes: G06F 2212/152 Virtualized environment, e.g. logically partitioned system
• CPC Class Codes: G06F 2212/214 Solid state disk
• CPC Class Codes: G06F 2212/222 Non-volatile memory
• CPC Class Codes: G06F 2212/261 Storage comprising a plurality of storage devices
• CPC Class Codes: G06F 3/0619 in relation to data integrity, e.g. data losses, bit errors
• CPC Class Codes: G06F 3/0656 Data buffering arrangements
• CPC Class Codes: G06F 3/0688 Non-volatile semiconductor memory arrays
• CPC Class Codes: G06F 9/466 Transaction processing
• CPC Class Codes: G06F 9/467 Transactional memory G06F9/528 takes precedence

Abstract

A method for a transactional commit in a storage unit is provided. The method includes receiving a logical record from a storage node into a transaction engine of a storage unit of the storage node and writing the logical record into a data structure of the transaction engine. The method includes writing, to a command queue of the transaction engine, an indication to perform an atomic update using the logical record and transferring each portion of the logical record from the data structure of the transaction engine to non-persistent memory of the storage unit as a committed transaction. A storage unit for a storage system is also provided.

Claims

1. A storage unit for a storage system, comprising:

non-persistent memory;

persistent memory; and

a transaction engine, configured to receive logical records and configured to transfer the logical records to the non-persistent memory as atomic updates, wherein each of the atomic updates constitutes a transfer of a logical record to the non-persistent memory, the transaction engine configurable to complete an atomic update that is in progress responsive to an event, wherein a transaction interlock assures coherency of the atomic update that is in progress, and wherein the contents of the non-persistent memory, having a coherent image of a logical record associated with the atomic update, are available in the persistent memory, after the event, for use during a recovery of the storage system.

2. The storage unit for a storage system of claim 1, further comprising:

the logical record having a length greater than a bus width of a processor associated with the transfer of the logical record; and

the processor configured to transfer contents of the non-persistent memory to the persistent memory, triggered by the event.

3. The storage unit for a storage system of claim 1, wherein the transaction engine comprises:

a transaction record buffer having a byte length sufficient to hold at least one logical record that is destined for writing into the non-persistent memory; and

a command queue configured to hold information about a logical record, wherein the transaction engine is coupled to the transaction record buffer and the command queue, the transaction engine configured to perform a transfer of the logical record from the transaction record buffer to the non-persistent memory as a committed transaction.

4. The storage unit for a storage system of claim 1, further comprising:

an energy reserve coupled to the non-persistent memory, the persistent memory, and the transaction engine, the energy reserve having sufficient electrical capacity to power a completion of the transfer of the logical record to the non-persistent memory and power a further transfer of contents of the non-persistent memory to the persistent memory;

the non-persistent memory including dynamic random-access memory (DRAM); and

the persistent memory including flash memory.

5. The storage unit for a storage system of claim 1, further comprising:

a transaction interlock that tracks information about the logical record;

the information about the logical record includes a destination address of the logical record and information about a size of the logical record;

the transaction interlock includes a transaction record buffer; and

a byte length of the transaction record buffer is greater than one logical record.

6. The storage unit for a storage system of claim 1, further comprising:

a transaction interlock having a transaction record buffer configured to receive at least one logical record;

a first bus coupling the transaction record buffer and the non-persistent memory; and

a second bus coupling the processor and the transaction record buffer, wherein the first bus is wider than the second bus.

7. The storage unit for a storage system of claim 1, further comprising a transaction interlock, configured to direct each of the atomic updates, wherein completing the transfer of the logical record to the non-persistent memory constitutes a transactional commit which is recorded in information about the logical record as written to the transaction interlock.

8. A storage system, comprising:

a plurality of storage nodes coupled as a storage cluster;

each of the plurality of storage nodes having one or more storage units;

each storage unit having non-persistent memory, persistent memory, and a transaction engine with a transaction interlock configured to assure coherency of an atomic update; and

the transaction engine and the transaction interlock of each storage unit configured to transfer logical records from the transaction engine with the transaction interlock to the non-persistent memory with each logical record transferred as the atomic update interlocked by the transaction interlock, wherein at least one logical record has a length greater than a bus width of a processor associated with the transfer of the logical record.

9. The storage system of claim 8, further comprising:

each storage unit configured to complete an atomic update in progress during an event;

each storage unit configured to dump at least a portion of contents of the non-persistent memory to the persistent memory, triggered by the event and completion of the atomic update, wherein the at least a portion of the contents of the non-persistent memory includes a coherent image of a logical record transferred as the atomic update interlocked by the transaction interlock; and

each storage unit having an energy reserve, coupled to at least the non-persistent memory, the persistent memory, the transaction engine and the transaction interlock, the energy reserve having sufficient electrical energy to support dumping the at least a portion of the contents of the non-persistent memory to the persistent memory.

10. The storage system of claim 8, wherein the transaction engine and the transaction interlock comprise:

a transaction record buffer, configured to receive and hold a logical record from a processor of one of the plurality of storage nodes;

a command queue, configured to hold a description of a transfer of the logical record;

a transfer engine, configured to move the logical record from the transaction record buffer to the non-persistent memory in accordance with the description of the transfer of the logical record to the command queue; and

a transaction indicator, configured to show a start of a next logical record for transfer in the transaction record buffer, and an end of logical records in the transaction record buffer, wherein the start of the next logical record is updated in the transaction indicator following an update to the command queue, and wherein the end of logical records is updated in the transaction indicator using flow control for receiving logical records into the transaction record buffer.

11. The storage system of claim 8, further comprising:

each storage unit having a direct memory access (DMA) unit, configured to copy contents of the non-persistent memory to the persistent memory; and

an energy reserve coupled to the non-persistent memory, the persistent memory, the DMA unit, the transaction engine and the transaction interlock, the energy reserve configured to store and release sufficient electrical energy so that the transaction engine and the transaction interlock engine can complete the atomic update and the DMA unit can complete a copy of the contents of the non-persistent memory to the persistent memory, wherein the energy reserve includes a super capacitor, a battery backup, or a power hold up device.

12. The storage system of claim 8, further comprising:

the transaction interlock having a command queue; and

the transaction engine configured to receive a description of a transfer of a logical record into the command queue to be used for the transfer of the logical record from the transaction engine with the transaction interlock to the non-persistent memory, wherein the description of the transfer of the logical record includes a sequence number, a destination starting address of the logical record, and one of a destination ending address of the logical record or a size of the logical record.

13. The storage system of claim 8, wherein:

the transaction engine and the transaction interlock are configured to complete moving a logical record from the transaction engine with the transaction interlock to the non-persistent memory, as the atomic update, responsive to an event;

the transaction interlock includes a command queue, configured to hold a description of the atomic update; and

the command queue and the description of the atomic update provide an interlock mechanism of the atomic update.

14. The storage system of claim 8, wherein the non-persistent memory, the persistent memory, and the transaction engine with the transaction interlock are in a power loss hold up domain and wherein a logical record has a length greater than a bus width in the storage unit.