| 摘要 |
Disclosed are methods, apparatuses, and systems for leveraging an offload programming model for local checkpoints. Computing entities in a computing environment are implemented as at least one source and a larger number of sinks. A task dispatcher dispatches tasks including executable codes to source(s), and execution of task codes is managed by the source(s). Code sections in the task code, which is designated to be offloaded, are offloaded to the sinks by generating offload context information. In conjunction with each offload, an offload object is generated and written to a storage. The offloaded code sections are executed by the sinks, which return result data to the source, e.g., through a direct writing to a memory buffer which is specified in the offload context information. The states of sinks are monitored to detect failures, and, when a failure occurs, the source retrieves the offload object which corresponds to the code section, which is offloaded on a failed sink, regenerates the offload context information for the code section, and transmits the offload context information to other sinks for execution. According to the present invention, a substantial storage consumption is prevented, and it is possible to prevent a big ratio of an overall process bandwidth from being practically wasted. |
