Scheduling Light-Weight Parallelism in ArTCoP

Abstract

Abstract. We present the design and prototype implementation of the scheduling component in ArTCoP (architecture transparent control of parallelism), a novel run-time environment (RTE) for parallel execution of high-level languages. A key feature of ArTCoP is its support for deep process and memory hierarchies, shown in the scheduler by supporting light-weight threads. To realise a system with easily exchangeable components, the system defines a micro-kernel , providing basic infrastructure, such as garbage collection. All complex RTE operations, including the handling of parallelism, are implemented at a separate system level. By choosing Concurrent Haskell as high-level system language, we obtain a prototype in the form of an executable specification that is easier to maintain and more flexible than conventional RTEs. We demonstrate the flexibility of this approach by presenting implementations of a scheduler for light-weight threads in ArTCoP, based on GHC Version 6.6.

PAEAN: Portable and scalable runtime support for parallel Haskell dialects

“…System modularity is achieved similarly to PAEAN components: driven by a workload distribution logic, tasks are distributed to processors and controlled by a scheduling component; scheduling relies on a communication subsystem; and a monitoring component informs dynamic and adaptive scheduling decisions. In (Berthold, 2008;Berthold et al, 2008), we presented a Haskell scheduling framework in the spirit of the ARTCOP design. Complex scheduling policies and heuristics can be defined in Haskell, based on a system of type classes for parallel jobs and scheduler states.…”

“…ArTCoP 11 is a highly modular micro-kernel design for parallel Haskell (Berthold et al, 2008) that is closely related to PAEAN. The kernel level implements simple and generic routines for asynchronous communication, basic execution management and system information.…”

“…Coordination constructs can be used at both a Library and an Application level either to implement algorithms directly or to implement parallel algorithmic patterns and skeletons (Cole, 1989). System modularity is achieved similarly to PAEAN components: driven by In Berthold (2008), Berthold et al (2008), we presented a Haskell scheduling framework in the spirit of the ArTCoP design. Complex scheduling policies and heuristics can be defined in Haskell, based on a system of type classes for parallel jobs and scheduler states.…”

Eden – Parallel Functional Programming with Haskell

“…Details on the parallel runtime system and Eden's concept of implementation can best be found in [8,10,4]. The technique of layered parallel runtime environments has been further developed and generalised by Berthold, Loidl and Al Zain [3,12]. The Eden trace viewer tool EdenTV is available on Eden's web site.…”