Automatic Locality Exploitation in the Codelet Model

Abstract

State-of-the-art codelet scheduling focuses on dynamic workload balance of codelets (similar to tasks). While this approach may achieve reasonable performance since computation resources are fully utilized, it may not attain optimal energy savings. In this paper, targeting at IBM Cyclops64 -a manycore system, we propose a novel polynomial time algorithm that finds out the optimal codelet scheduling in terms of maximum locality and minimum global memory accesses. Our algorithm leverages static information regarding locality among codelets to achieve better performance and energy efficiency. By using local buffers to pass data produced in one codelet to another, global memory accesses can be greatly reduced. The experimental results on our developed IBM Cyclops-64 emulator show that the codelet scheduling of our algorithm removes up to 59.7% of global memory accesses, achieves up to 68.1% of performance improvement, and reduces up to 40.7% of energy consumption comparing to the state-of-the-art codelet scheduling.

On communication efficient dataflow computing in software defined networking enabled cloud

“…Because of its fine‐grain and event‐driven characteristics, much effort has been devoted to studying how to optimize codelet program execution model. For example, Chen et al 24 study the problem of scheduling codelets with the aim of maximum locality exploitation among the codelets for performance efficiency optimization. Zuckerman et al 25 propose a potential solution to solve the resource management issue in exascale position by using codelet execution model.…”

Genetic scheduling policy on codelet model

“…Due to the fine‐grained attribute of the Codelet Model, locality exploitation and energy efficiency are the main emphasis of recent research studies on this model. Chen et al 21 proposed a series of codelet scheduling algorithms (ie, min‐cost flow based, max first, and graph partitioning based algorithm) to find an optimal schedule for maximizing locality. These three algorithms use different methods to solve the problems of balancing locality and energy consumption by reducing the access number of global memory.…”

Position Paper: Locality-Driven Scheduling of Tasks for Data-Dependent Multithreading

“…al [7] in the codelet model [19] uses a codelet graph to find the best scheduling algorithm of an application targeting a many core architecture. In the graph, edges are annotated with the amount of data shared by a source and a sink codelet and codelets are grouped in order to minimize the sum of intergroup weights.…”

“…For applications requiring fine-grain parallelism, other works have used the dataflow and the codelet model for the implementation of such techniques [5], [6], [7]. Even though these models are preferred for applications that need to adapt dynamically during runtime, overheads could be produced in the scheduling of the many small tasks generated by the model.…”