Acyclic Partitioning of Large Directed Acyclic Graphs

Herrmann, Julien; Kho, Jonathan; Uçar, Bora; Kaya, Kamer; Çatalyürek, Ümit V.

doi:10.1109/ccgrid.2017.101

Cited by 30 publications

(20 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We argue that such decisions are local, and hence cannot take into account the global structure of the graph. Recently, we have developed one of the first multi-level acyclic DAG partitioners [11]. We hypothesize that clusters found using such a DAG partitioner are much more successful in putting together the tasks with complex dependencies, and hence in minimizing the overall interprocessor communication, and we confirm this hypothesis in our experiments.…”

Section: Introductionsupporting

confidence: 83%

“…We propose novel heuristic approaches to solve the MIN-MAKESPAN problem, using a recent directed graph partitioner [11]. We compare the results with classical list-based (Cluster-based only) task priority type (partition priority) placement clustering approach load balancing produced clusters task priority and clustering heuristics, that we first describe and adapt for the duplex single-port communication model (Section IV-A).…”

Section: Algorithmsmentioning

confidence: 99%

“…The partition-based heuristics start by computing an acyclic partition of the DAG, using a recent DAG partitioner [11]. This…”

Section: B Partition-based Heuristicsmentioning

confidence: 99%

“…The pseudo-code of *-PART can be found in the companion research report [19]. The time complexity of the partitioner is linear on the graph size [11], hence the complexity of BL-EST-PART (resp. ETF-PART) is the same as BL-EST (resp.…”

Section: B Partition-based Heuristicsmentioning

confidence: 99%

See 3 more Smart Citations

A Scalable Clustering-Based Task Scheduler for Homogeneous Processors Using DAG Partitioning

Benoît

Uçar

Herrmann

et al. 2019

2019 IEEE International Parallel and Distributed Processing Symposium (IPDPS)

Self Cite

View full text Add to dashboard Cite

When scheduling a directed acyclic graph (DAG) of tasks with communication costs on computational platforms, a good trade-off between load balance and data locality is necessary. List-based scheduling techniques are commonly-used greedy approaches for this problem. The downside of listscheduling heuristics is that they are incapable of making shortterm sacrifices for the global efficiency of the schedule. In this work, we describe new list-based scheduling heuristics based on clustering for homogeneous platforms, under the realistic duplex single-port communication model. Our approach uses an acyclic partitioner for DAGs for clustering. The clustering enhances the data locality of the scheduler with a global view of the graph. Furthermore, since the partition is acyclic, we can schedule each part completely once its input tasks are ready to be executed. We present an extensive experimental evaluation showing the tradeoffs between the granularity of clustering and the parallelism, and how this affects the scheduling. Furthermore, we compare our heuristics to the best state-of-the-art list-scheduling and clustering heuristics, and obtain more than three times better makespan in cases with many communications.

show abstract

Section: Introductionsupporting

confidence: 83%

Section: Algorithmsmentioning

confidence: 99%

“…The partition-based heuristics start by computing an acyclic partition of the DAG, using a recent DAG partitioner [11]. This…”

Section: B Partition-based Heuristicsmentioning

confidence: 99%

Section: B Partition-based Heuristicsmentioning

confidence: 99%

See 2 more Smart Citations

A Scalable Clustering-Based Task Scheduler for Homogeneous Processors Using DAG Partitioning

Benoît

Uçar

Herrmann

et al. 2019

2019 IEEE International Parallel and Distributed Processing Symposium (IPDPS)

Self Cite

View full text Add to dashboard Cite

show abstract

“…However, none of these tools can partition directed graphs under the constraint that the quotient graph is a DAG. Very recently, Hermann et al [14] presented the first multi-level partitioner for DAGs. The algorithm finds matchings such that the contracted graph remains acyclic and uses an algorithm comparable to Fiduccia-Mattheyses algorithm [11] for refinement.…”

Section: Preliminariesmentioning

confidence: 99%

Evolutionary multi-level acyclic graph partitioning

Moreira¹,

Popp²,

Schulz

2018

Proceedings of the Genetic and Evolutionary Computation Conference

View full text Add to dashboard Cite

Directed graphs are widely used to model data flow and execution dependencies in streaming applications. This enables the utilization of graph partitioning algorithms for the problem of parallelizing computation for multiprocessor architectures. However due to resource restrictions, an acyclicity constraint on the partition is necessary when mapping streaming applications to an embedded multiprocessor. Here, we contribute a multi-level algorithm for the acyclic graph partitioning problem. Based on this, we engineer an evolutionary algorithm to further reduce communication cost, as well as to improve load balancing and the scheduling makespan on embedded multiprocessor architectures. Practical MotivationComputer vision and imaging applications have high demands for computational power. However, these applications often need to run on embedded devices with severely limited compute resources and a tight thermal budget. This requires the use of specialized hardware and a programming model that allows to fully utilize the compute resources.The context of this research is the development of specialized processors at Intel Corporation for advanced imaging and computer vision. In particular, our target platform is a heterogeneous multiprocessor architecture that is currently used in Intel processors. Several VLIW processors with vector units are available to exploit the abundance of data parallelism that typically exists in imaging algorithms. The architecture is designed for low power and typically has small local program and data memories. To cope with memory constraints, it is necessary to break the application, which is given as a directed dataflow graph, into smaller blocks that are executed one after another. The quality of this partitioning has a strong impact on performance. It is known that the problem is NP-complete [22] and that there is no constant factor approximation algorithm for general graphs [22]. Therefore heuristic algorithms are used in practice.We contribute (a) a new multi-level approach for the acyclic graph partitioning problem, (b) based on this, a coarse-grained distributed evolutionary algorithm, (c) an objective function that improves load balancing on the multiprocessor architecture and (d) an evaluation on a large set of graphs and a real application. Our focus is on solution quality, not algorithm running time, since these partitions are typically computed once before the application is compiled. The rest of the paper is organized as follows: we present all necessary background information on the application graph and hardware in Section 2 and then briefly introduce the notation and related work in Section 3. Our new multi-level approach is described in Section 4. We illustrate the evolutionary algorithm that provides multi-level recombination and mutation operations, as well as a novel fitness function in Section 5. The experimental evaluation of our algorithms is found in Section 6. We conclude in Section 7.

show abstract

Mapping tree‐shaped workflows on systems with different memory sizes and processor speeds

Kulagina

Meyerhenke

Benoît

2023

Concurrency and Computation

View full text Add to dashboard Cite

Directed acyclic graphs are commonly used to model scientific workflows, by expressing dependencies between tasks, as well as the resource requirements of the workflow. As a special case, rooted directed trees occur in several applications, for instance in sparse matrix computations. Since typical workflows are modeled by large trees, it is crucial to schedule them efficiently, so that their execution time (or makespan) is minimized. Furthermore, it is usually beneficial to distribute the execution on several compute nodes, hence increasing the available memory, and allowing us to parallelize parts of the execution. To exploit the heterogeneity of modern clusters in this context, we investigate the partitioning and mapping of tree‐shaped workflows on two types of target architecture models: in AM1, each processor can have a different memory size, and in AM2, each processor can also have a different speed (in addition to a different memory size). We design a three‐step heuristic for AM1, which adapts and extends previous work for homogeneous clusters [Gou C, Benoit A, Marchal L. Partitioning tree‐shaped task graphs for distributed platforms with limited memory. IEEE Trans Parallel Dist Syst 2020; 31(7): 1533–1544]. The changes we propose concern the assignment to processors (accounting for the different memory sizes) and the availability of suitable processors when splitting or merging subtrees. For AM2, we extend the heuristic for AM1 with a two‐phase local search approach. Phase A is a swap‐based hill climber, while (the optional) Phase B is inspired by iterated local search. We evaluate our heuristics for AM1 and AM2 with extensive simulations, and we demonstrate that exploiting the heterogeneity in the cluster significantly reduces the makespan, compared to the state of the art for homogeneous processors.

show abstract

Acyclic Partitioning of Large Directed Acyclic Graphs

Cited by 30 publications

References 13 publications

A Scalable Clustering-Based Task Scheduler for Homogeneous Processors Using DAG Partitioning

A Scalable Clustering-Based Task Scheduler for Homogeneous Processors Using DAG Partitioning

Evolutionary multi-level acyclic graph partitioning

Mapping tree‐shaped workflows on systems with different memory sizes and processor speeds

Contact Info

Product

Resources

About