Accurate analytical performance model of communications in MPI applications

Martínez, D.; Dominguez, Jessica A.; Pena, Tomás F.; Rivera, Francisco F.; Blanco, Vicente

doi:10.1109/ipdps.2009.5161175

Cited by 10 publications

(8 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Subsequently, based on the LogP model, models such as LoPC [10], LoGPC [11], LoGPG [12], LogGPO [13], and LoOgGP [14] were developed to account for network congestion. Unlike the model in this paper, these models do not consider routing protocols to model congestion and do not model the traffic on individual links.…”

Section: The Dragonfly Interconnectmentioning

confidence: 99%

Maximizing Throughput on a Dragonfly Network

Jain

Bhatelé

et al. 2014

SC14: International Conference for High Performance Computing, Networking, Storage and Analysis

View full text Add to dashboard Cite

Abstract-Interconnection networks are a critical resource for large supercomputers. The dragonfly topology, which provides a low network diameter and large bisection bandwidth, is being explored as a promising option for building multi-Petaflop/s and Exaflop/s systems. Unlike the extensively studied torus networks, the best choices of message routing and job placement strategies for the dragonfly topology are not well understood. This paper aims at analyzing the behavior of an interconnect based on the dragonfly topology for various routing strategies, job placement policies, and application communication patterns. Our study is based on a novel model that predicts traffic on individual links for direct, indirect, and adaptive routing strategies. We analyze results for individual communication patterns and some common parallel job workloads. The predictions presented in this paper are for a 100+ Petaflop/s prototype machine with 92,160 highradix routers and 8.8 million cores.

show abstract

Section: The Dragonfly Interconnectmentioning

confidence: 99%

Maximizing Throughput on a Dragonfly Network

Jain

Bhatelé

et al. 2014

SC14: International Conference for High Performance Computing, Networking, Storage and Analysis

View full text Add to dashboard Cite

show abstract

“…Numerous works exist that model the performance of a particular MPI implementation on a particular system [1,15,16,19,20]. Other works focus on modeling particular aspects of MPI, such as collective communication [11,15,18].…”

Section: Previous Work and A General Approach To Modelingmentioning

confidence: 99%

“…The LogGP model provides the parameter o to model the per-message overhead; however, this might not be sufficient if the overhead grows with the message size. An additional parameter, for example, O for the overhead per byte, could be introduced to capture those effects [15]. Such a model and its derivation are well understood, and we omit details for brevity.…”

Section: The Deficiency Of Current Point-to-point Modelsmentioning

confidence: 99%

Toward Performance Models of MPI Implementations for Understanding Application Scaling Issues

Hoefler

Gropp

Thakur

et al. 2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Designing and tuning parallel applications with MPI, particularly at large scale, requires understanding the performance implications of different choices of algorithms and implementation options. Which algorithm is better depends in part on the performance of the different possible communication approaches, which in turn can depend on both the system hardware and the MPI implementation. In the absence of detailed performance models for different MPI implementations, application developers often must select methods and tune codes without the means to realistically estimate the achievable performance and rationally defend their choices. In this paper, we advocate the construction of more useful performance models that take into account limitations on network-injection rates and effective bisection bandwidth. Since collective communication plays a crucial role in enabling scalability, we also provide analytical models for scalability of collective communication algorithms, such as broadcast, allreduce, and all-to-all. We apply these models to an IBM Blue Gene/P system and compare the analytical performance estimates with experimentally measured values. MotivationPerformance modeling of parallel applications leads to an understanding of their running time on parallel systems. To develop a model for an existing application or algorithm, one typically constructs a dependency graph of computations and communications from the start of the algorithm (input) to the end of the algorithm (output). This application model can then be matched to a machine model in order to estimate the run time of the algorithm on a particular architecture.Performance models can be used to make important early decisions about algorithmic choices. For example, to compute a three-dimensional Fast Fourier Transformation (3d FFT), one can either use a one-dimensional decomposition where each process computes full planes (2d FFTs) or a two-dimensional decomposition where each process computes sets of pencils (1d FFTs). If we assume

show abstract

“…In this case, two linear regressions are needed to obtain the overhead and gap parameters of the LoOgGP model (o, O, g and G) [12]:…”

Section: Assessment Of Logp-based Modelsmentioning

confidence: 99%

“…Experimental results on different systems [12] show that a given message size range can be split into intervals in which T o (s) and T g (s) simultaneously presents a linear dependency with the size of the message. The boundaries of these intervals can be matched with architectural features such as buffer sizes, MTU size or the switch between transport protocols.…”

Section: Assessment Of Logp-based Modelsmentioning

confidence: 99%

Automatic parameter assessment of logp-based communication models in MPI environments

Martínez

Blanco

Dominguez

et al. 2010

Procedia Computer Science

Self Cite

View full text Add to dashboard Cite

Communications in modern interconnection networks become a complex issue to deliver the highest performance to the user. In MPI environments, this complexity leads to different communication protocols for different message sizes. The characterization of these different behaviors is important and useful for software developers and network designers. This paper presents an automatic method to obtain a characterization of the communication behavior of a specific MPI environment using LogP-based models. This procedure automatically detects the message sizes where the communication behavior changes due to the influence of different architectural features and protocols. Thereby, the range of message sizes is split in different intervals, in a transparent way to the user. Then, using LogP-based models, each interval is characterized by its own set of parameters. The procedure is based on the parameterized roundtrip-time low-overhead microbenchmark. The detection of different communication behaviors and the assessment of the parameters of the LogP-based models were implemented in an easy to use statistical environment. Real MPI environments were characterized using our proposal.

show abstract

Accurate analytical performance model of communications in MPI applications

Cited by 10 publications

References 10 publications

Maximizing Throughput on a Dragonfly Network

Maximizing Throughput on a Dragonfly Network

Toward Performance Models of MPI Implementations for Understanding Application Scaling Issues

Automatic parameter assessment of logp-based communication models in MPI environments

Contact Info

Product

Resources

About