Kill-safe synchronization abstractions
Matthew Flatt, Robert Bruce Findler
Abstract
Matthew Flatt, Robert Bruce Findler
Abstract
When an individual task can be forcefully terminated at any time, cooperating tasks must communicate carefully. For example, if two tasks share an object, and if one task is terminated while it manipulates the object, the object may remain in an inconsistent or frozen state that incapacitates the other task. To support communication among terminable tasks, language run-time systems (and operating systems) provide kill-safe abstractions for inter-task communication. No kill-safe guarantee is available, however, for abstractions that are implemented outside the run-time system.In this paper, we show how a run-time system can support new kill-safe abstractions without requiring modification to the run-time system, and without requiring the run-time system to trust any new code. Our design frees the run-time implementor to provide only a modest set of synchronization primitives in the trusted computing base, while still allowing tasks to communicate using sophisticated abstractions.
OpenAlex reports 2 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
When an individual task can be forcefully terminated at any time, cooperating tasks must communicate carefully. For example, if two tasks share an object, and if one task is terminated while it manipulates the object, the object may remain in an inconsistent or frozen state that incapacitates the other task. To support communication among terminable tasks, language run-time systems (and operating systems) provide kill-safe abstractions for inter-task communication. No kill-safe guarantee is available, however, for abstractions that are implemented outside the run-time system.In this paper, we show how a run-time system can support new kill-safe abstractions without requiring modification to the run-time system, and without requiring the run-time system to trust any new code. Our design frees the run-time implementor to provide only a modest set of synchronization primitives in the trusted computing base, while still allowing tasks to communicate using sophisticated abstractions.
Key concepts: Computer science, Task (project management), Synchronization (alternating current), Set (abstract data type), Object (grammar), State (computer science), Code (set theory), Distributed computing