Systems and methods for scalable object storage

• US 9,507,812 B2
• Filed: 04/22/2014
• Issued: 11/29/2016
• Est. Priority Date: 08/12/2011
• Status: Active Grant

• Current Assignee: Nexenta by DDN, Inc.
• Original Assignee: Nexenta Systems Incorporated (DataDirect Networks Incorporated)
• Inventors: Bestler, Caitlin, Aizman, Alexander
• Primary Examiner(s): Chbouki, Tarek
• Application Number:
  US14/258,791
• Publication Number:
  US 20140229455A1
• Time in Patent Office: 952 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/1752 based on file chunks
• CPC Class Codes: G06F 16/22 Indexing; Data structures therefor; Storage structures

Abstract

The present disclosure provides an advantageous systems and related methods for scalable object storage. One embodiment relates to a system that includes at least a manifest tracking subsystem and a chunk payload subsystem. The manifest tracking subsystem stores version manifests and encodes forward-references from the version manifests to chunks. The chunk payload subsystem stores the chunks and encodes back-references to version manifests that reference the chunk. Other embodiments, aspects and features are also disclosed.

Claims

1. A system for providing distributed storage of payload and metadata for data objects, the system comprising:

a manifest tracking subsystem comprising a plurality of manifest server computers communicatively interconnected by a network, the manifest tracking subsystem storing version manifests for the objects, wherein an object stored in the system has version manifests stored at multiple manifest server computers, wherein a version manifest for the object encodes the metadata for the object including forward-references to chunks of the object, wherein the chunks comprise sequences of payload bytes for the object; and

a chunk payload subsystem comprising a plurality of chunk server computers communicatively interconnected by the network, the chunk payload subsystem storing the chunks and encoding back-references to version manifests that reference the chunks, wherein each chunk is stored at multiple chunk server computers,

wherein the back-references maintained by the plurality of chunk server computers comprise speculative back-references and verified back-references,

wherein a speculative back-reference is created during a chunk put operation to put a chunk into the system, and

wherein a verified back-reference corresponding to the speculative back-reference is created after the manifest tracking subsystem has completed a transaction that included the chunk put operation.

2. The system of claim 1, wherein each forward-reference comprises a reference to a chunk identifier without identifying any server on which the chunk is stored, wherein a first consistent hash algorithm is used on the forward reference to obtain location information for the chunk identifier.

3. The system of claim 1, wherein each back-reference comprises a reference to an object identifier without identifying any server on which the object is stored, wherein a second consistent hash algorithm is used on the back-reference to obtain location information for the object identifier.

4. The system of claim 1, wherein the speculative back-reference encodes a transaction identifier supplied with the chunk put operation.

5. The system of claim 4, wherein a put cancelling operation that identifies the chunk and the transaction identifier causes deletion of the speculative back-reference.

6. The system of claim 4, wherein the speculative back-reference further encodes a timestamp of the object put transaction.

7. The system of claim 1, wherein the speculative back-reference is created even when the chunk previously existed in the chunk payload subsystem.

8. The system of claim 1, wherein the speculative back-reference is deleted if the speculative back-reference is of a lower class-of-storage than an existing verified back-reference from the chunk to the object.

9. The system of claim 1, wherein, in order to verify the speculative back-reference, the chunk payload subsystem queries the manifest tracking subsystem to confirm that there is a corresponding forward-reference to the chunk from the object.

10. The system of claim 9, wherein upon confirmation of the corresponding forward-reference, the verified back-reference from the chunk to the object is created if none already exists, and a class-of-storage attribute for the verified back-reference is updated if the verified back-reference already exists,

further wherein the speculative back-reference is then deleted.

11. The system of claim 10, wherein the class-of-storage attribute is updated to be a greater of the class-of-storage attributes in the speculative and the verified back-references.

12. The system of claim 9, wherein upon confirmation of the corresponding forward-reference, a reference count for the chunk for the specific class-of-storage is incremented,

further wherein the speculative back-reference is then deleted.

13. The system of claim 12, wherein reference counts are adjusted in response to notifications from the manifest tracking subsystem that a class-of-storage for the chunk from a specific object has changed.

14. The system of claim 1 further comprising:

a client which initiates an object put transaction,

wherein the client obtains a transaction identifier for the object put transaction from the manifest tracking subsystem by sending a request for transaction identifier that at least includes an identity of the object to be put.

15. The system of claim 14, wherein the request for transaction identifier includes a class-of-storage identifier.

16. The system of claim 14, wherein the manifest tracking subsystem cryptographically signs the transaction identifier, and wherein the transaction identifier encodes at least the identity of the object to be put and a unique transaction identifier.

17. The system of claim 16, wherein the transaction identifier is unique even when there are multiple concurrent transactions updating the object to be put.

18. The system of claim 14, wherein the chunk payload subsystem rejects a request for a chunk put operation if the request does not supply a transaction identifier which was cryptographically signed by the manifest subsystem.

19. The system of claim 1, wherein the manifest tracking subsystem and the chunk payload subsystem collaborate to maintain a class-of-storage attribute for each chunk stored in the system.

20. The system of claim 19, wherein the manifest tracking subsystem issues a notification of a class-of-storage update when a class-of-storage required for a chunk by an object has been updated.

21. The system of claim 20, wherein the chunk payload subsystem applies the class-of-storage update to the verified back-reference.

22. The system of claim 20, wherein the chunk payload subsystem creates the verified back-reference, if necessary, by confirming the speculative back-reference between the chunk and the object.

23. The system of claim 1, wherein a chunk get operation is validated using a transaction identifier obtained from the manifest tracking subsystem and supplied to the chunk payload subsystem with the chunk get operation.

24. The system of claim 23, wherein the chunk payload subsystem rejects the chunk get operation if the chunk does not have a back-reference to an object identified in the transaction identifier.

25. A method for providing storage of data objects using a distributed storage system, the method comprising:

storing chunks of a version of an object in a chunk payload subsystem comprising a plurality of chunk server computers, wherein each chunk is stored at multiple chunk server computers;

encoding forward-references to the chunks of the version of the object in a version manifest that is stored in a manifest tracking subsystem comprising a plurality of manifest server computers, wherein an object stored in the system has version manifests stored at multiple manifest server computers; and

encoding back-references from the chunks of the version of the object to the version manifest that contains the forward-references to the chunks,

wherein a consistent hash algorithm is used on the forward references to obtain location information for the chunks,

wherein the back-references comprise speculative back-references and verified back-references,

wherein a speculative back-reference is created during a chunk put operation to put a chunk into the system, and

wherein a verified back-reference corresponding to the speculative back-reference is created after the manifest tracking subsystem has completed a transaction that included the chunk put operation.

26. A method for putting an object into a distributed storage system, the method comprising:

receiving from a client of the system a request to put an object into the system;

obtaining a transaction identifier for the request from a manifest tracking subsystem;

dividing a payload of the object into a plurality of chunks;

putting the plurality of chunks into a chunk payload subsystem, wherein each chunk is replicated so as to be stored at multiple chunk server computers of the chunk payload subsystem;

putting a manifest to a manifest tracking subsystem, wherein the manifest includes forward references from the object to the plurality of chunks, wherein the manifest is written to multiple manifest server computers of the manifest tracking subsystem, and wherein a first consistent hashing algorithm is used on the forward references to obtain location information for the plurality of chunks,

wherein the manifest is put into the manifest tracking subsystem after the plurality of chunks are put into the chunk payload subsystem; and

writing back-references from the plurality of chunks to the object, wherein the back-references are written in the chunk payload subsystem, wherein a second consistent hashing algorithm is used on the back-references to obtain location information for the object,

wherein the back-references comprise speculative back-references and verified back-references,

wherein a speculative back-reference is created during a chunk put operation to put a chunk into the system, and

wherein a verified back-reference corresponding to the speculative back-reference is created after the manifest tracking subsystem has completed a transaction that included the chunk put operation.

27. The method of claim 26 further comprising:

incrementing a reference count for a chunk when the speculative back-reference is confirmed.

28. The method of claim 26 further comprising:

creating the verified back-reference when the speculative back-reference is confirmed.

29. The method of claim 28, wherein the transaction identifier is cryptographically signed by the manifest tracking subsystem.

30. The method of claim 26, wherein the transaction identifier encodes an identifier for the object.

31. The method of claim 26, wherein the transaction identifier is unique even when there are multiple concurrent transactions updating the object to be put.