Fault tolerance for persistent main memory

• US 10,452,498 B2
• Filed: 06/28/2013
• Issued: 10/22/2019
• Est. Priority Date: 06/28/2013
• Status: Active Grant

• Current Assignee: Snowflake, Inc.
• Original Assignee: Hewlett Packard Enterprise Development LP (Hewlett-Packard Enterprise Company)
• Inventors: Lesartre, Gregg B., Morris, Dale C., Gostin, Gary, Herrell, Russ W., Wheeler, Andrew R., Gaither, Blaine D.
• Primary Examiner(s): Butler, Sarai E
• Application Number:
  US14/901,559
• Publication Number:
  US 20160147620A1
• Time in Patent Office: 2,307 Days
• Field of Search: 714 63, 714 631
• CPC Class Codes: G06F 11/1004 to protect a block of data words, e.g. CRC or checksum G06F11/1076 takes precedence; security arrangements for protecting computers or computer systems against unauthorized activity G06F21/00
• CPC Class Codes: G06F 11/108 Parity data distribution in semiconductor storages, e.g. in SSD
• CPC Class Codes: G06F 11/1666 where the redundant component is memory or memory area
• CPC Class Codes: G06F 11/20 using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
• CPC Class Codes: G06F 11/2069 Management of state, configuration or failover
• CPC Class Codes: G06F 2201/805 Real-time

Abstract

A computing system can include a processor and a persistent main memory including a fault tolerance capability. The computing system can also include a memory controller to store data in the persistent main memory and create redundant data. The memory controller can also store the redundant data remotely with respect to the persistent main memory. The memory controller can further access the redundant data during failure of the persistent main memory.

Claims

1. A computing system, comprising:

a processor;

a persistent main memory comprising a fault tolerance capability; and

a memory controller to;

store data in the persistent main memory;

create redundant data;

store the redundant data remotely with respect to the persistent main memory; and

access the redundant data during failure of the persistent main memory.

2. The computing system of claim 1, wherein the fault tolerance capability comprises storing data to the persistent main memory and mirroring the data to a remote persistent memory.

3. The computing system of claim 1, wherein the fault tolerance capability comprises storing data to the persistent main memory and updating at least one parity remote memory based on a difference between old data and new data.

4. The computing system of claim 1, wherein the computing system comprises a plurality of computing nodes and a plurality of persistent main memories, each computing node comprising:

a processor; and

a memory controller to communicate with the persistent main memories in response to a request by the processor,

wherein the persistent main memories reside remotely to the computing nodes, the persistent main memories comprising a pool of shared persistent main memory, regions of which are allocated to each computing node.

5. The computing system of claim 1, wherein the persistent main memory is accessed as memory, is accessed through direct memory access, and is treated as storage.

6. A computing system comprising fault tolerance, the computing system comprising:

a persistent main memory for storing data;

a remote persistent main memory for storing redundant data; and

a memory controller,

wherein the memory controller accesses the redundant data when the data in the persistent main memory cannot be accessed.

7. The system of claim 6, wherein the redundant data comprises a mirrored copy of the data.

8. The system of claim 6, wherein the computing system comprises:

a plurality of nodes, each node comprising a persistent main memory for storing data; and

a remote persistent memory for storing redundant data, the remote persistent memory acting as a parity node, comprising a check sum of parity bits of data in the persistent main memories of the nodes of the computing system,

wherein the check sum is updated when data is written to the persistent main memory of a node; and

wherein upon failure to read the data, the check sum and all other data values contributing to the check sum are read and combined to reconstruct lost data.

9. The system of claim 6, wherein the computing system comprises a plurality of computing nodes and a plurality of persistent main memories, each computing node comprising:

a processor; and

a memory controller to communicate with the persistent main memories in response to a request by the processor;

the plurality of persistent main memories comprising a first persistent main memory selected by address for storing data, a second persistent main memory selected by address for storing a mirrored copy of the data,

wherein data written to the first persistent main memory is also written to the second persistent main memory; and

wherein upon failure to read the data, the mirrored copy of the data is read.

10. The system of claim 6, wherein the computing system comprises a plurality of computing nodes and a plurality of persistent main memories, each computing node comprising:

a processor; and

a memory controller to communicate with the persistent main memories in response to a request by the processor;

the plurality of persistent main memories comprising a first persistent main memory selected by address for storing data, a second persistent main memory selected by address for storing a check sum associated with the data;

wherein the check sum in the second persistent main memory is updated based on a difference between old data and new data stored in the first persistent main memory; and

wherein upon failure to read the data, the check sum and all other data values contributing to the check sum are read and combined to reconstruct lost data.

11. The system of claim 10, wherein the persistent main memories comprise a pool of shared persistent main memory, regions of which are allocated to each computing node.

12. The system of claim 11, wherein in event of a failure of an original computing node to which a region of the pool of shared persistent main memory is allocated, a functional computing node accesses the region of the pool of shared persistent main memory so that data in the region of the pool of shared persistent memory is always available.

13. A method, comprising:

receiving, in a memory controller, a request to write data to a persistent main memory of a computing system;

writing the data to the persistent main memory; and

writing redundant data to a persistent memory remote to the persistent main memory.

14. The method of claim 13, further comprising simultaneously writing the data to the persistent main memory and writing the redundant data to the persistent memory remote to the persistent main memory.

15. The method of claim 13, writing redundant data comprising updating a check sum of parity bits of data in the computing system, wherein the data in the computing system is stored on persistent main memory of at least two nodes.

16. The method of claim 13, further comprising:

receiving, in a memory controller, a request to access data in the persistent main memory;

attempting to access the persistent main memory;

receiving notice of a failure of the persistent main memory;

reconstructing the data from the redundant data;

saving the data; and

returning the data to complete the request.

17. The method of claim 13, wherein responsive to receiving the request, the memory controller writes the data to the persistent main memory, and wherein the memory controller writes the redundant data by:

generating a transaction to a plurality of remote persistent memories including the persistent memory remote to the persistent main memory, wherein generation of the transaction causes generation and storage of the redundant data within each remote persistent memory as the memory controller writes the data to the persistent main memory.

18. The method of claim 17, wherein in a mirroring mode, the redundant data is generated as an identical copy of the data written to the persistent main memory, the identical copy written to each remote persistent memory.

19. The method of claim 13, wherein responsive to receiving the request, the memory controller writes the data to the persistent main memory, and wherein the memory controller writes the redundant data to the persistent memory remote to the persistent main memory by:

sending a command to update the redundant data on the persistent memory remote to the persistent main memory.

20. The method of claim 19, wherein the memory controller, as a redundant array of independent disks (RAID) controller, performs a first exclusive-or (XOR) operation on the data of the request and old data in the persistent main memory, the memory controller sending the command by sending a write delta command including a result of the first XOR operation to the persistent memory remote to the persistent main memory,

and wherein the persistent memory remote to the persistent main memory responsively reads the redundant data, performs a second XOR operation on the read redundant data and the result of the first XOR operation, and stores a result of the second XOR operation as the redundant data.