Technologies for accelerator fabric protocol multipathing

Patent metadata

Field	Value
Publication number	US-11301407-B2
Application number	US-201916242928-A
Country	US
Kind code	B2
Filing date	Jan 8, 2019
Priority date	Jan 8, 2019
Publication date	Apr 12, 2022
Grant date	Apr 12, 2022

Abstract

Technologies for accessing pooled accelerator resources over a network fabric are disclosed. In disclosed embodiments, an application hosted by a computing platform accesses remote accelerator resources over a network fabric using protocol multipathing mechanisms. A communication session is established with the remote accelerator resources. The communication session comprises at least two connections. The at least two connections at least include a first connection having or utilizing a first transport layer and a second connection having or utilizing a second transport layer that is different than the first transport layer. Other embodiments may be disclosed and/or claimed.

First claim

The invention claimed is:

1. At least one non-transitory computer readable storage medium (CRSM) comprising instructions stored thereon, wherein execution of the instructions is to cause a computing platform to:
operate an initiator application executed by the computing platform to establish, for another application executed by the computing platform, a communication session with a hardware accelerator remote from the computing platform and communicatively coupled over a network fabric, wherein the communication session comprises at least two concurrent connections to the hardware accelerator, the at least two concurrent connections at least including a first connection having a first transport layer and a second connection having a second transport layer that is different than the first transport layer; and
communicate data, for the another application, with the hardware accelerator over the first connection or the second connection.

2. The at least one CRSM of claim 1 , wherein, to establish the communication session with the hardware accelerator, execution of the instructions is to cause the computing platform to operate the initiator application to:
perform an establishment procedure with the hardware accelerator for the first connection; and
receive a session identifier (ID) from the hardware accelerator during the establishment procedure.

3. The at least one CRSM of claim 2 , wherein the establishment procedure is a first establishment procedure and, to establish the communication session with the hardware accelerator, execution of the instructions is to cause the computing platform to operate the initiator application to:
perform a second establishment procedure with the hardware accelerator for the second connection; and
send a session ID to the hardware accelerator during the second establishment procedure.

4. The at least one CRSM of claim 1 , wherein the computing platform is communicatively coupled with network interface circuitry (NIC) and, to establish the communication session with the hardware accelerator, execution of the instructions is to cause the computing platform to operate the initiator application to:
instruct the NIC to establish the first connection with the hardware accelerator;
obtain a session ID from the NIC upon establishment of the first connection with the hardware accelerator; and
send the session ID to the hardware accelerator during an establishment procedure for the second connection.

5. The at least one CRSM of claim 4 , wherein the first transport layer is implemented by the NIC, and the second transport layer is implemented by the computing platform.

6. The at least one CRSM of claim 1 , wherein execution of the instructions is to cause the computing platform to operate the initiator application to:
detect a link failure of the first connection;
activate the second connection;
send a session ID to the hardware accelerator over the second connection; and
attempt reactivation of the first connection.

7. The at least one CRSM of claim 1 , wherein execution of the instructions is to cause the computing platform to:
operate a virtual machine (VM) or an application container, wherein the initiator application and the another application are to operate within the VM or the application container.

8. The at least one CRSM of claim 7 , wherein execution of the instructions is to cause the computing platform to operate the initiator application to:
detect a trigger for migration of the VM or the application container to another computing system;

before the migration:
deactivate the first connection,
activate the second connection, and
send a session ID to the hardware accelerator over the second connection; and after the migration:
deactivate the second connection, and
re-activate the first connection, and
send the session ID to the remote hardware accelerator over the first connection.

9. The at least one CRSM of claim 1 , wherein the first transport layer is a Remote Direct Memory Access (RDMA) based transport layer, and the second transport layer is a Transmission Control Protocol (TCP) layer.

10. The at least one CRSM of claim 1 , wherein the computing platform is in a server computer system; the hardware accelerator comprises a field programmable gate array (FPGA), (ASIC), or programmable System-on-Chip (SoC); and the network fabric is an Ethernet fabric or a Fibre channel (FC) fabric.

11. Hardware accelerator circuitry comprising:
one or more configurable logic blocks (CLBs) configured with logic arranged to:
establish a communication session over a network fabric with a computing platform remote from the hardware accelerator circuitry, wherein the communication session comprises at least two concurrent connections to the computing platform, the at least two concurrent connections at least including a first connection having a first transport layer and a second connection having a second transport layer that is different than the first transport layer; and
obtain data for execution by the hardware accelerator circuitry over the first connection or the second connection.

12. The hardware accelerator circuitry of claim 11 , wherein, to establish the communication session with the computing platform, the logic is arranged to:
receive, over the network fabric from the computing platform, a request to establish the first connection;
generate a session identifier (ID) based on the request; and
send, over the first connection to the computing platform, an acknowledgement (ACK) including the session ID.

13. The hardware accelerator circuitry of claim 12 , wherein the request is a first request, the ACK is a first ACK, and, to establish the communication session with the computing platform, the logic is arranged to:
receive, over the network fabric from the computing platform, a second request to establish the second connection, the second request including the session ID;
store the session ID in association with a first connection ID of the first connection and a second connection ID of the second connection; and
send, over the second connection to the computing platform, a second ACK to the computing platform to indicate successful establishment of the communication session.

14. The hardware accelerator circuitry of claim 13 , wherein the logic is arranged to:
activate the first connection in response to receipt of a control or data packet from the computing platform over the first connection.

15. The hardware accelerator circuitry of claim 14 , wherein the logic is arranged to:
activate the second connection in response to detection of a disconnection of the first connection, wherein detection of the disconnection of the first connection is based on expiration of a timer or based on receipt of a disconnection data packet.

16. The hardware accelerator circuitry of claim 15 , wherein the computing platform is a first computing platform, the disconnection data packet is a first message, and the logic is arranged to:
receive, over the network fabric from a second computing platform, a second message to re-activate the first connection based on migration of a virtual machine (VM) or an application container from the computing platform, the second message including the session ID, and the second computing platform being different than the first computing platform;
activate the first connection in response to receipt of the second message.

17. The hardware accelerator circuitry of claim 11 , wherein the first transport layer is a Remote Direct Memory Access (RDMA) based transport layer, and the

Assignees

• Intel Corp

Inventors

• Sen Sujoy
• Ranganathan Narayan

Classifications

• H04L67/141

  Setup of application sessions (admission control or resource allocation in data switching networks H04L47/70) · CPC title

• G06F15/17331

  Distributed shared memory [DSM], e.g. remote direct memory access [RDMA] · CPC title

• G06F3/0604

  Improving or facilitating administration, e.g. storage management · CPC title

• G06F3/061

  Improving I/O performance · CPC title

• G06F13/4027 Primary

  using bus bridges (G06F13/4022 takes precedence) · CPC title