How file access patterns influence interference among cluster applications

Kuo, Chih-Song; Shah, Abad; Nomura, Akihiro; Matsuoka, Satoshi; Wolf, Felix

doi:10.1109/cluster.2014.6968743

Cited by 19 publications

(14 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Performance observed when accessing a PFS is highly dependent on the way this access is performed, i.e. on the application's access pattern [3], [4]. For instance, higher performance is achieved when accessing sequentially positioned portions of a file in large requests as opposed to accessing small sparse portions.…”

Section: Introductionmentioning

confidence: 99%

Collective I/O Performance on the Santos Dumont Supercomputer

Carneiro

Bez

Boito

et al. 2018

2018 26th Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP)

View full text Add to dashboard Cite

The historical gap between processing and data access speeds causes many applications to spend a large portion of their execution on I/O operations. From the point of view of a large-scale, expensive, supercomputer, it is important to ensure applications achieve the best I/O performance to promote an efficient usage of the machine. In this paper, we evaluate the I/O infrastructure of the Santos Dumont supercomputer, the largest one from Latin America. More specifically, we investigate the performance of collective I/O operations. By conducting an analysis of a scientific application that uses the machine, we identify large performance differences between the available MPI implementations. We then further study the observed phenomenon using the BT-IO and IOR benchmarks, in addition to a custom microbenchmark. We conclude that the customized MPI implementation by Bull (used by more than 20% of the jobs) presents the worst performance for small collective write operations. Our results are being used to help the Santos Dumont users to achieve the best performance for their applications. Additionally, by investigating the observed phenomenon, we provide information to help improve future MPI-IO collective write implementations.

show abstract

Section: Introductionmentioning

confidence: 99%

Collective I/O Performance on the Santos Dumont Supercomputer

Carneiro

Bez

Boito

et al. 2018

2018 26th Euromicro International Conference on Parallel, Distributed and Network-Based Processing (PDP)

View full text Add to dashboard Cite

show abstract

“…In our previous work, we analyzed write access patterns and their effect on interference [6]. In this work, we expand on the topic with a more realistic checkpointing pattern, evaluate how the interference potential depends on the number of processes the application runs with, and confirm our findings with a larger set of production codes.…”

Section: Introductionmentioning

confidence: 54%

“…On systems with limited metadata resources such patterns can quickly create a bottleneck at scale. Compared to our previous work [6], we have updated the open-write-close pattern to create a new file in each iteration, mimicking the checkpointing pattern in a more realistic fashion.…”

Section: File Access Patternsmentioning

confidence: 99%

“…First, we verified the interference trend for micro-benchmarks against applications, and later confirmed it for application vs application. In our previous study [6], we considered two typical I/O-intensive HPC applications, OpenFOAM and MadBench2. Here, we extend the work by also evaluating our approach with HACCIO, a code-writing large checkpoints, against the micro-benchmarks and the other two applications.…”

Section: Applicationsmentioning

confidence: 99%

See 1 more Smart Citation

How File-access Patterns Influence the Degree of I/O Interference between Cluster Applications

Shah

Kuo

Nomura

et al. 2019

JSFI

View full text Add to dashboard Cite

On large-scale clusters, tens to hundreds of applications can simultaneously access a parallel file system, leading to contention and, in its wake, to degraded application performance. In this article, we analyze the influence of file-access patterns on the degree of interference. As it is by experience most intrusive, we focus our attention on write-write contention. We observe considerable differences among the interference potentials of several typical write patterns. In particular, we found that if one parallel program writes large output files while another one writes small checkpointing files, then the latter is slowed down when the checkpointing files are small enough and the former is vice versa. Moreover, applications with a few processes writing large output files already can significantly hinder applications with many processes from checkpointing small files. Such effects can seriously impact the runtime of real applications-up to a factor of five in one instance. Our insights and measurement techniques offer an opportunity to automatically classify the interference potential between applications and to adjust scheduling decisions accordingly.

show abstract

“…Although PFSs provide a high I/O parallelism, some workloads face low performance because of their access patterns. Some optimization approaches exist, for instance, for workloads with small() and non‐contiguous() I/O requests, which are well‐known examples of problematic access patterns.…”

Section: Introductionmentioning

confidence: 70%

A coarse‐grained page cache aware multivariate analytical model for the storage performance of a parallel file system

Inacio

Dantas

2017

Concurrency and Computation

View full text Add to dashboard Cite

Summary Huge requirements of emerging Big Data applications combined with the performance impairment of current I/O subsystems pose a great challenge on data storage, management, and access performance. In order to design efficient storage systems, a clear understanding about how factors in the I/O path affect the performance of a data‐intensive application is of utmost importance. This paper reports our ongoing research toward addressing this issue, presenting a coarse‐grained page cache aware multivariate analytical model for the performance of write operations in a parallel file system. The proposed model was developed to reflect the performance behavior observed in an extensive experimental effort, in which the impact of 14 parameters in the response time and throughput of the OrangeFS was investigated. More than one million experiments were carried out using four distinct computing infrastructures, providing a detailed performance characterization. Additionally, a thorough evaluation of the proposed model, covering more than 14 000 scenarios, is reported, discussing both qualitative and quantitative aspects. Evaluation results indicate that the model succeeded in representing the behavior of the parallel file system performance, achieving a Mean Absolute Percentage Error of 39.94%.

show abstract

How file access patterns influence interference among cluster applications

Cited by 19 publications

References 16 publications

Collective I/O Performance on the Santos Dumont Supercomputer

Collective I/O Performance on the Santos Dumont Supercomputer

How File-access Patterns Influence the Degree of I/O Interference between Cluster Applications

A coarse‐grained page cache aware multivariate analytical model for the storage performance of a parallel file system

Contact Info

Product

Resources

About