Capturing snapshots of offload applications on many-core coprocessors

1. A method, comprising:

capturing a snapshot of an offload process being executed by one or more many-core processors, the offload process being in signal communication with a host process being executed by a host processor, at least the offload being in signal communication with a monitoring process;

executing a communication channel emptying method, by a process on one end of a communication channel, to empty the communication channel, the emptying method comprising:

forming a consistent global state by collaboration between the host process, a Coprocessor Offload Infrastructure (COI) daemon, and the offload process; and

blocking one or more threads in the host process and the offload process to prevent use of existing Symmetric Communications InterFace (SCIF) channels between the host process and the offload process;

terminating the offload process on the one or more many-core processors, by the monitor process responsive to a communication between the monitor process and the offload processing being disrupted; and

simultaneously executing multiple processes whose aggregate memory footprint exceeds physical memory on the processors by swapping out the offload process and swapping in another offload process based on scheduling and resource management policies,

wherein the snapshot comprises a respective predetermined minimum set of information required to restore a same state of the process as when the snapshot was taken, and

wherein the swapping out comprises preemptively migrating the offload process from a failing processor to a healthy processor.

2. The method of claim 1 , further comprising executing a snapshot correctness ensuring method, by at least one of the host process, the offload process, and the monitor process, to ensure correct snapshots.

3. The method of claim 1 , wherein the communication channel emptying method comprises a two-ended message transmission confirmation method for ensuring that resulting snapshots of both sides of the communication channel are in agreement regarding whether a message was sent or not sent.

4. The method claim 2 , further comprising executing a communication channel disconnecting method, by a process on one end of a communication channel, to disconnect from the communication channel.

5. The method of claim 1 , further comprising executing a post-snapshot execution-resuming method, by at least one of the host process, the offload process, and the monitor process, to resume execution after the snapshot is taken.

6. The method of claim 1 , further comprising executing an offload process execution restarting method, by at least one of the host process, the offload processes, and the monitor process, to restart an execution of the offload process from the snapshot.

7. The method of claim 6 , wherein the offload process execution restarting method comprises sending information through the monitor process to the offload process to reestablish the communication channels with the host process.

8. The method of claim 1 , further comprising executing a host process execution restarting method, by at least one of the host process, the offload process, and the monitor process, to restart an execution of the host process from the snapshot.

9. A non-transitory article of manufacture tangibly embodying a computer readable program which when executed causes a computer to perform the steps of claim 1 .