Live migration of hardware accelerated applications

Patent metadata

Field	Value
Publication number	US-10169065-B1
Application number	US-201615197260-A
Country	US
Kind code	B1
Filing date	Jun 29, 2016
Priority date	Jun 29, 2016
Publication date	Jan 1, 2019
Grant date	Jan 1, 2019

Abstract

Live migration of a hardware accelerated application may be orchestrated by cloud services to transfer the application from a source server to a destination server. The live migration may be triggered by probe circuitry that monitors quality of service metrics for migration conditions at the source server. When live migration is initiated by the cloud services, a snapshot of all state information relevant to the application at the source server may be saved to network attached storage accessible by the destination server. Changes to said state information at the source server may be mirrored onto the network attached storage. The destination server may copy the snapshot and subsequent changes and run the application in parallel before taking complete control of the application. After a handshake operation between the source and destination servers, the application may be shut down at the source server.

First claim

What is claimed is:

1. A method of migrating a hardware accelerated application from a source server to a destination server, wherein the source server comprises a processor connected to an external migration controller and reconfigurable circuitry, wherein the reconfigurable circuitry includes a plurality of accelerator resource slots and a memory management unit for the accelerator resource slots, the method comprising:
at the source server, receiving a migration notification from the migration controller, wherein the migration notification specifies a set of accelerator resource slots of the plurality of accelerator resource slots to be migrated and an identifier for the resources in the memory management unit to be migrated; wherein the migration controller is further configured for:
saving an image of state information associated with the hardware accelerated application from the source server to network attached storage in response to receiving the migration notification;
copying the image of state information associated with the hardware accelerated application from the network attached storage to the destination server; and
running the hardware accelerated application in parallel on the source server and the destination server.

2. The method defined in claim 1 , wherein saving the image of state information comprises:
configuring memory management circuitry on the source server in a snapshot mode; and
in response to configuring the memory management circuitry in the snapshot mode, copying entries in the memory management circuitry associated with the hardware accelerator resources used by the hardware accelerated application to the network attached storage.

3. The method defined in claim 2 , further comprising:
after saving the image of state information associated with the hardware accelerated application to the network attached storage, mirroring state changes at the source server to the network attached storage.

4. The method defined in claim 3 , wherein mirroring the state changes at the source server comprises:
configuring the memory management circuitry on the source server in a mirroring mode; and
in response to configuring the memory management circuitry on the source server in the mirroring mode, updating the image of state information saved at on the network attached storage based on the state changes at the source server.

5. The method defined in claim 1 ,
wherein the image of state information associated with the hardware accelerated application is saved to a location of the network attached storage specified in the migration notification.

6. The method defined in claim 1 , further comprising:
at the destination server, receiving a migration notification, wherein the image of state information associated with the hardware accelerated application is copied from a location of the network attached storage specified in the migration notification.

7. The method defined in claim 1 , further comprising:
performing a handshake operation with the source server and the destination server to transfer control of the hardware accelerated application to the destination server.

8. The method defined in claim 7 , wherein performing the handshake operation comprises:
using the source server to write a first token to the network attached storage indicating that the source server is ready for migration to continue;
using the destination server to detect the first token at the network attached storage; and
using the destination server to write a second token to the network attached storage indicating that the destination server is available to assume control of the application.

9. The method defined in claim 8 , wherein performing the handshake operation further comprises:
using the source server to detect the second token at the network attached storage; and
in response to detecting the second token, halting state mirroring operations at the source server.

10. The method defined in claim 1 , further comprising:
using probe circuitry at the source server to measure quality of service metrics associated with the hardware accelerated application;
monitoring the measured quality of service metrics for at least one migration condition; and
in response to detecting the at least one migration condition, notifying network control services of the at least one migration condition.

11. The method defined in claim 1 , wherein saving the image of state information associated with the hardware accelerated application comprises:
traversing a plurality of entries in the memory management unit for the accelerator resource slots and detecting which of the plurality of entries are associated with the identifier received in the migration notification; and
copying the entries associated with the identifier received in the migration notification to the network attached storage.

12. A method of migrating a hardware accelerated application from a source server, the source server comprising a processor coupled to memory-stored instructions, which when executed by the processor, implement virtualized instances of a daemon and migration process server code, the method comprising:
at a daemon on the source server, receiving a migration notification from a migration controller; wherein the migration process code is configured for:
saving a snapshot of state information associated with the hardware accelerated application from the source server to network attached storage in response to receiving the migration notification by configuring memory management circuitry associated with a plurality of hardware acceleration resources on the source server in a snapshot mode, wherein the snapshot of state information includes state values for selected hardware acceleration resources of the plurality of hardware acceleration resources specified in the migration notification;
mirroring state changes associated with the hardware accelerated application from the source server to the network attached storage by configuring the memory management circuitry in a mirroring mode, wherein the state values for the selected hardware acceleration resources specified in the migration notification are updated on the network attached storage in response to the memory management circuitry being configured in the mirroring mode; and
monitoring the network attached storage for a notification that another server is available to assume control of the hardware accelerated application.

13. The method defined in claim 12 , further comprising:
mounting the network attached storage; and
saving the snapshot of state information associated with the hardware accelerated application from the source server to a location on the network attached storage specified in the migration notification.

14. The method defined in claim 13 , wherein saving the snapshot of state information associated with the hardware accelerated application comprises:
using the daemon to configure memory management circuitry associated with programmable logic circuitry on the source server in a snapshot mode;
in the snapshot mode, searching a plurality of entries stored in the memory management circuitry for particular entries associated with an application identifier specified in the migration notification; and
writing the particular entries associated with the application identifier to the location on the network attached storage specified in the migration notification.

15. The method defined in claim 13 , wherein saving the snapshot of state information associated with the hardware accelerated application comprises:
copying state information associated with hardwar

Assignees

• Altera Corp

Inventors

• Nye Jeffrey L
• Rao Shiva

Classifications

• G06F9/5088

  involving task migration · CPC title

• G06F9/4856 Primary

  resumption being on a different machine, e.g. task migration, virtual machine migration (G06F9/5088 takes precedence) · CPC title

• G06F21/554

  involving event detection and direct action · CPC title

• G06F9/45558 Primary

  Hypervisor-specific management and integration aspects · CPC title

• H04L67/34

  involving the movement of software or configuration parameters  (network booting or remote initial program loading [RIPL] G06F9/4416) · CPC title