Iterative computations with ordered read–write locks

Abstract

Abstract:We introduce the framework of ordered read-write locks, ORWL, that are characterized by two main features: a strict FIFO policy for access and the attribution of access to lock-handles instead of processes or threads. These two properties allow applications to have a controlled pro-active access to resources and thereby to achieve a high degree of asynchronicity between different tasks of the same application. For the case of iterative computations with many parallel tasks which access their resources in a cyclic pattern we provide a generic technique to implement them by means of ORWL. We show that the possible execution patterns for such a system correspond to a combinatorial lattice structure and that this lattice is finite iff the configuration contains a potential deadlock. In addition, we provide efficient algorithms: one that allows for a deadlock-free initialization of such a system and another one for the detection of deadlocks in an already initialized system.

Memory Access Classification for Vertical Task Parallelism

“…A proof of this theorem follows from the fact that the dependency relation ℘ is acyclic and from the theory developed for Ordered Read-Write Locks (ORWL), see [11]. This formalizes a model for autonomous execution of tasks or processes that manage their shared resources by means of cyclic FIFOs that are attached to data resources.…”

“…This is a pattern that is found widely in the field, e.g many iterative algorithms that traverse matrices or geometric objects fall into that category. Using the assumption of a constant data access pattern, we detect dependencies and derive an implicit scheduling at runtime using Ordered Read-Write Locks (ORWL) during some initial iterations, see [11]. After this instrumentation phase, we do not need to check any condition or recompute dependencies, thus eliminating the drawbacks of the Inspector-Executor model.…”

“…A fine grained (theoretical) model is presented that guarantees that all computational results are exactly as they would have been produced by the originating sequential program, Section III. Based on [11] we are able to prove that our parallel execution model is correct, fair and deadlock free.…”

Optimizing Locality by Topology-Aware Placement for a Task Based Programming Model

“…Among these programming models ORWL [4] "Ordered Read-Write Locks" provides a framework based on the management of shared resources in a parallel environment. These resources are abstracted in the ORWL model by the notion of location.…”

Automatic Code Generation for Iterative Multi-dimensional Stencil Computations

“…ORWL, see [6], is an inter-task synchronization paradigm for iterative dataoriented parallel and distributed algorithms. It favors algorithmic control and data consistency by introducing a FIFO-based lock mechanism that handles data-dependencies between tasks and data resources.…”