Communication and topology-aware load balancing in Charm++ with TreeMatch

Jeannot, Emmanuel; Meneses, Esteban; Mercier, Guillaume; Tessier, François; Zheng, Gengbin

doi:10.1109/cluster.2013.6702666

Cited by 21 publications

(15 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Topology-aware dynamic load balancing is a hot topic [37], [58], which concerns itself with managing change in application behavior and coping with affinity dependence in the input dataset. This requires modification of the affinity modeling from a static model (e.g.…”

Section: Trends and Requirementsmentioning

confidence: 99%

Trends in Data Locality Abstractions for HPC Systems

Unat

Dubey

Hoefler

et al. 2017

IEEE Trans. Parallel Distrib. Syst.

Self Cite

View full text Add to dashboard Cite

Abstract-The cost of data movement has always been an important concern in high performance computing (HPC) systems. It has now become the dominant factor in terms of both energy consumption and performance. Support for expression of data locality has been explored in the past, but those efforts have had only modest success in being adopted in HPC applications for various reasons. However, with the increasing complexity of the memory hierarchy and higher parallelism in emerging HPC systems, locality management has acquired a new urgency. Developers can no longer limit themselves to low-level solutions and ignore the potential for productivity and performance portability obtained by using locality abstractions. Fortunately, the trend emerging in recent literature on the topic alleviates many of the concerns that got in the way of their adoption by application developers. Data locality abstractions are available in the forms of libraries, data structures, languages and runtime systems; a common theme is increasing productivity without sacrificing performance. This paper examines these trends and identifies commonalities that can combine various locality concepts to develop a comprehensive approach to expressing and managing data locality on future large-scale high-performance computing systems.

show abstract

Section: Trends and Requirementsmentioning

confidence: 99%

Trends in Data Locality Abstractions for HPC Systems

Unat

Dubey

Hoefler

et al. 2017

IEEE Trans. Parallel Distrib. Syst.

Self Cite

View full text Add to dashboard Cite

show abstract

“…It is also integrated in the Charm++ programming environment as an ecient load-balancer [11] and in Open MPI [16] to enable rank reordering in virtual topology management routines (e.g. MPI_Dist_graph_create).…”

Section: Treematchmentioning

confidence: 99%

Topology-aware resource management for HPC applications

Georgiou

Jeannot

Mercier

et al. 2017

Proceedings of the 18th International Conference on Distributed Computing and Networking

Self Cite

View full text Add to dashboard Cite

The Resource and Job Management System (RJMS) is a crucial system software part of the HPC stack. It is responsible for eciently delivering computing power to applications in supercomputing environments. Its main intelligence relies on resource selection techniques to nd the most adapted resources to schedule the users' jobs. Improper resource selection operations may lead to poor performance executions and global system utilization along with increase of system fragmentation and jobs starvation. These phenomenas play a role in the increase of the platforms' total cost of ownership and should be minimized. This paper introduces a new topology-aware resource selection algorithm to determine the best choice among the available nodes of the platform based upon their position within the network and taking into account the applications communication matrix. To validate our approach, we integrated this algorithm as a plugin for Slurm, a popular and widespread HPC resource and job management system (RJMS). We validated our plugin with dierent optimization schemes by comparing with the default Slurm algorithm using both emulation of a large-scale platform, and by carrying out experiments in a real cluster.

show abstract

“…A complementary approach to ours is to take into account the communications within the jobs themselves by migrating processes depending on their communication affinity [5]. This approach is rooted in the application, while we are positionning ourselves at the resource and job management system level.…”

Section: Related Workmentioning

confidence: 99%

Communication Models Insights Meet Simulations

Dutot

Poquet

Trystram

2015

Euro-Par 2015: Parallel Processing Workshops

View full text Add to dashboard Cite

It is well-known that taking into account communications while scheduling jobs in large scale parallel computing platforms is a crucial issue. In modern hierarchical platforms, communication times are highly different when occurring inside a cluster or between clusters. Thus, allocating the jobs taking into account locality constraints is a key factor for reaching good performances. However, several theoretical results prove that imposing such constraints reduces the solution space and thus, possibly degrades the performances. In practice, such constraints simplify implementations and most often lead to better results. Our aim in this work is to bridge theoretical and practical intuitions, and check the differences between constrained and unconstrained schedules (namely with respect to locality and node contiguity) through simulations. We have developped a generic tool, using SimGrid as the base simulator, enabling interactions with external batch schedulers to evaluate their scheduling policies. The results confirm that insights gained through theoretical models are ill-suited to current architectures and should be reevaluated.

show abstract

Communication and topology-aware load balancing in Charm++ with TreeMatch

Cited by 21 publications

References 18 publications

Trends in Data Locality Abstractions for HPC Systems

Trends in Data Locality Abstractions for HPC Systems

Topology-aware resource management for HPC applications

Communication Models Insights Meet Simulations

Contact Info

Product

Resources

About