Switched direct attached shared storage architecture

• US 9,756,128 B2
• Filed: 04/16/2014
• Issued: 09/05/2017
• Est. Priority Date: 04/17/2013
• Status: Active Grant

• Current Assignee: Angel Acosta
• Original Assignee: Apeiron Data Systems
• Inventors: Lomelino, Larry, Christ, Christopher, Marripudi, Gunneswara Rao, Laswell, Harry
• Primary Examiner(s): Truong, Lan-Dai T
• Application Number:
  US14/254,868
• Publication Number:
  US 20140317206A1
• Time in Patent Office: 1,238 Days
• Field of Search: 709206, 709219, 709223, 709220, 709222, 709238
• CPC Class Codes: H04L 12/4633 Interconnection of networks using encapsulation techniques, e.g. tunneling
• CPC Class Codes: H04L 51/04 Real-time or near real-time messaging, e.g. instant messaging [IM]
• CPC Class Codes: H04L 67/1097 for distributed storage of data in networks, e.g. transport arrangements for network file system [NFS], storage area networks [SAN] or network attached storage [NAS]

Abstract

Various embodiments for implementing a switched direct attached shared storage architecture as disclosed herein include: providing a plurality of compute nodes, each compute node having integrated compute capabilities, data storage, and a network interface controller (Host NIC), the plurality of compute nodes being in data communication with each other via a local area network, the plurality of compute nodes each including distributed storage processing software resident thereon; providing a plurality of physical data storage devices in data communication with a storage controller; and enabling data communications in a data store switch fabric between the plurality of compute nodes and the plurality of physical data storage devices via the Host NIC and the storage controller, the data store switch fabric encapsulating data requests from the plurality of compute nodes into data frames for transport to corresponding physical data storage devices.

Claims

1. A switched direct attached shared data storage system comprising:

a plurality of compute nodes, each compute node having integrated compute capabilities, data storage, and a network interface controller (Host NIC), the plurality of compute nodes being in data communication with each other via a local area network, the plurality of compute nodes each including distributed storage processing software resident thereon;

a plurality of physical data storage devices in data communication with a storage controller; and

a data store switch fabric enabling data communications between the plurality of compute nodes and the plurality of physical data storage devices via the Host NIC and the storage controller, the data store switch fabric encapsulating data requests from the plurality of compute nodes into Ethernet frames for transport to corresponding physical data storage devices, the encapsulated data requests including information formatted in a native NVMe (Non-Volatile Memory Express) storage protocol compatible with the corresponding physical data storage devices, wherein the data store switch fabric presents an Ethernet MAC (Media Access Control) table based virtual to physical remapping layer enabling the plurality of compute nodes to access the plurality of physical data storage devices in a virtual manner, the Host NIC presenting a Peripheral Component Interconnect Express (PCIe) endpoint to a corresponding compute node and the storage controller, which provides a root complex to a group of NVMe (Non-Volatile Memory Express) solid-state drives.

2. The system as claimed in claim 1 wherein the plurality of physical data storage devices is a plurality of solid-state NVMe (Non-Volatile Memory Express) drives (SSDs).

3. The system as claimed in claim 1 wherein the data store switch fabric is implemented using a protocol from the group consisting of:

Ethernet and Peripheral Component Interconnect Express (PCIe).

4. The system as claimed in claim 2 wherein the data store switch fabric enables data communications between the plurality of compute nodes and the plurality of NVMe (Non-Volatile Memory Express) drives in a manner to emulate a direct data connection.

5. The system as claimed in claim 1 wherein the data store switch fabric enables a distributed object store using an object tag format.

6. The system as claimed in claim 1 wherein the data store switch fabric overlaps processing operations to accelerate input/output operations between the plurality of compute nodes and the plurality of physical data storage devices using an Ethernet connection.

7. The system as claimed in claim 1 wherein the storage controller offloads a processing burden of the plurality of compute nodes, thereby allowing the plurality of compute nodes to write to a single data storage replica and have the corresponding data storage device automatically move the updated data to a designated replica within the data store switch fabric using native NVMe (Non-Volatile Memory Express) storage commands.

8. The system as claimed in claim 1 wherein the data store switch fabric implements a peer to peer (P2P) messaging protocol using NVMe (Non-Volatile Memory Express) storage command formats.

9. The system as claimed in claim 1 including an analytics framework to build an analytical model based on storage fabric performance, which can be utilized to estimate compute application runtimes, adjust resource assignments, and optimize the storage fabric performance.

10. A method comprising:

providing a plurality of compute nodes, each compute node having integrated compute capabilities, data storage, and a network interface controller (Host NIC), the plurality of compute nodes being in data communication with each other via a local area network, the plurality of compute nodes each including distributed storage processing software resident thereon;

providing a plurality of physical data storage devices in data communication with a storage controller; and

enabling data communications in a data store switch fabric between the plurality of compute nodes and the plurality of physical data storage devices via the Host NIC and the storage controller, the data store switch fabric encapsulating data requests from the plurality of compute nodes into Ethernet frames for transport to corresponding physical data storage devices, the encapsulated data requests including information formatted in a native NVMe (Non-Volatile Memory Express) storage protocol compatible with the corresponding physical data storage devices, wherein the data store switch fabric presents an Ethernet MAC (Media Access Control) table based virtual to physical remapping layer enabling the plurality of compute nodes to access the plurality of physical data storage devices in a virtual manner, the Host NIC presenting a Peripheral Component Interconnect Express (PCIe) endpoint to a corresponding compute node and the storage controller, which provides a root complex to a group of NVMe (Non-Volatile Memory Express) solid-state drives.

11. The method as claimed in claim 10 wherein the plurality of physical data storage devices is a plurality of solid-state NVMe (Non-Volatile Memory Express) drives (SSDs).

12. The method as claimed in claim 11 wherein the data store switch fabric enables data communications between the plurality of compute nodes and the plurality of NVMe (Non-Volatile Memory Express) drives in a manner to emulate a direct data connection.

13. The method as claimed in claim 10 wherein the data store switch fabric is implemented using a protocol from the group consisting of:

Ethernet and Peripheral Component Interconnect Express (PCIe).

14. The method as claimed in claim 10 wherein the data store switch fabric enables a distributed object store using an object tag format.

15. The method as claimed in claim 10 wherein the data store switch fabric overlaps processing operations to accelerate input/output operations between the plurality of compute nodes and the plurality of physical data storage devices using an Ethernet connection.

16. The method as claimed in claim 10 wherein the storage controller offloads a processing burden of the plurality of compute nodes, thereby allowing the plurality of compute nodes to write to a single data storage replica and have the corresponding data storage device automatically move the updated data to a designated replica within the data store switch fabric using native NVMe (Non-Volatile Memory Express) storage commands.

17. The method as claimed in claim 10 wherein the data store switch fabric implements a peer to peer (P2P) messaging protocol using NVMe (Non-Volatile Memory Express) storage command formats.

18. The method as claimed in claim 10 including providing an analytics framework to build an analytical model based on storage fabric performance, which can be utilized to estimate compute application runtimes, adjust resource assignments, and optimize the storage fabric performance.