Understanding Performance Interference in Next-Generation HPC Systems

Mondragon, Oscar H.; Bridges, Patrick G.; Levy, Scott; Ferreira, Kurt Brian; Widener, Patrick

doi:10.1109/sc.2016.32

Cited by 19 publications

(7 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Martinasso et al [36] develop a congestion-aware performance model for PCIe communication to study the impact of PCIe topology. Mondragon et al use both simulation and modeling technique to profile next-generation interference sources and performance of the HPC benchmarks [12]. Yang et al [37] performance modeling the applications by running kernels on the target platform and then conduct the prediction cross-platform based on relative performance between the target platforms.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Using hardware counter-based performance model to diagnose scaling issues of HPC applications

Ding

Song

et al. 2018

Neural Comput & Applic

View full text Add to dashboard Cite

Performance diagnosing for HPC applications can be extremely difficult due to their complicated performance behaviors. One hand, developers used to identify the potential performance bottlenecks by conducting detailed instrumentation, which may introduce significant performance overheads or even performance deviations. On the other hand, developers can only conduct small numbers of application runs for profiling the performance with the limitations on both computing resources and time duration. Meanwhile, the performance bottlenecks of HPC applications may vary with the degree of parallelism. To address these challenges, our paper proposes a systematic performance diagnosing method focusing on building an accurate and interpretable performance model with performance counters. Our method is able to diagnose the HPC application scaling issues by predicting its runtime and performance behaviors in different functions. After applying this modeling method on three real-world HPC applications, HOMME, CICE and OpenFoam, our evaluations show that our diagnosing method based on the performance model has the ability to diagnose the potential scaling issues, which is typically missed by the traditional performance diagnosing method and achieves about 10% prediction errors in a scale of 4096 MPI ranks on two problem sizes.Neural Computing and Applications (2019) 31:1563-1575 https://doi.org/10.1007/s00521-018-3496-z( 0123456789().,-volV) (0123456789(). ,-volV)

show abstract

Section: Related Workmentioning

confidence: 99%

“…Statistical model [12,13] tries to overcome the disadvantages by predicting the scaling performance according to a large number of sampled application runs without digging into the source codes. However, it introduces a large number of application runs to train the performance model for achieving satisfying accuracy.…”

Section: Introductionmentioning

confidence: 99%

Using hardware counter-based performance model to diagnose scaling issues of HPC applications

Ding

Song

et al. 2018

Neural Comput & Applic

View full text Add to dashboard Cite

show abstract

“…In CC environments, sharing computational resources such as memory and processors creates performance impacts when workloads are run simultaneously. The proper allocation of resources can mitigate these impacts [17].…”

Section: Tier 1: Node Allocation In the Cloud Using Genetic Programmingmentioning

confidence: 99%

An Intelligent Approach to Resource Allocation on Heterogeneous Cloud Infrastructures

Marquez¹,

Mondragon²,

González³

2021

Applied Sciences

Self Cite

View full text Add to dashboard Cite

Cloud computing systems are rapidly evolving toward multicloud architectures supported on heterogeneous hardware. Cloud service providers are widely offering different types of storage infrastructures and multi-NUMA architecture servers. Existing cloud resource allocation solutions do not comprehensively consider this heterogeneous infrastructure. In this study, we present a novel approach comprised of a hierarchical framework based on genetic programming to solve problems related to data placement and virtual machine allocation for analytics applications running on heterogeneous hardware with a variety of storage types and nonuniform memory access. Our approach optimizes data placement using the Hadoop File System on heterogeneous storage devices on multicloud systems. It guarantees the efficient allocation of virtual machines on physical machines with multiple NUMA (nonuniform memory access) domains by minimizing contention between workloads. We prove that our solutions for data placement and virtual machine allocation outperform other state-of-the-art approaches.

show abstract

“…Here, a sensitive application is one in which an MPI synchronizing collective call (e.g., MPI_Barrier, MPI_Allreduce, MPI_Allgather, and so on) is called at least once per iteration in an iterative simulation or algorithm. Previous works() have shown that interference sources have potentially greater impact on applications with fine‐grained parallelism (i.e., applications with shorter per‐iteration intervals). For example, in Seelam et al, authors show that jitter can generate slowdowns as high as 8% for applications with computation intervals of 100 ms and over 16% of slowdowns for applications with 10‐ms computation intervals at 32 K CPUs.…”

Section: Importance Of Time Agreementmentioning

confidence: 99%

An evaluation of the state of time synchronization on leadership class supercomputers

Jones

Ostrouchov

Koenig

et al. 2017

Concurrency and Computation

Self Cite

View full text Add to dashboard Cite

SummaryWe present a detailed examination of time agreement characteristics for nodes within extreme-scale parallel computers. Using a software tool we introduce in this paper, we quantify attributes of clock skew among nodes in three representative high-performance computers sited at three national laboratories. Our measurements detail the statistical properties of time agreement among nodes and how time agreement drifts over typical application execution durations.We discuss the implications of our measurements, why the current state of the field is inadequate, and propose strategies to address observed shortcomings. KEYWORDSclock synchronization, large-scale systems, system software, time service INTRODUCTIONThe trend towards increasing node counts in high-performance computing (HPC) is motivating a move toward greater levels of concurrency in HPC systems. Today's software environment is now being called on to produce new solutions for emerging issues including managing system power, resilience, and performance characteristics. The distributed algorithms that underlie such services operate much more efficiently in the presence of tightly synchronized clocks. For example, tightly synchronized clocks benefit well-known gang scheduling techniques and complex consensus algorithms. To illustrate the point, such time synchronization enables more aggressive assumptions about communication and synchronization patterns, the removal of unnecessary locks, and a wide range of other applications. Clock-based techniques are already frequently deployed in cloud and data center distributed systems for precisely these reasons.We examined the time synchronization on some of the world's fastest and most powerful machines. These leadership-class systems employ high-end hardware connected by an extremely low-latency, low-jitter, interconnect in a carefully controlled environment, in contrast to widely distributed cloud-based systems based on commodity hardware and networks. Because of this, we assumed that these systems would have more stable, predictable hardware clocks, and close base time agreement using only standard time synchronization systems like Network Time Protocol (NTP). We did not believe that the complex hardware and software techniques used to provide time synchronization in wide-area systems would be necessary in leadership systems.Our results demonstrate that the actual time uncertainty for leadership-class machines is often unexpectedly large, in some cases over 600 milliseconds despite network latencies of less than two microseconds. Building on this, we set out to thoroughly quantify the magnitude of the time synchronization challenge in leadership-class systems. This study shows that the current time protocol in use, NTP, is not suitable for providing the level of time synchronization necessary for important system software tasks such as coordinated scheduling. Based on this, we conclude

show abstract

Understanding Performance Interference in Next-Generation HPC Systems

Cited by 19 publications

References 36 publications

Using hardware counter-based performance model to diagnose scaling issues of HPC applications

Using hardware counter-based performance model to diagnose scaling issues of HPC applications

An Intelligent Approach to Resource Allocation on Heterogeneous Cloud Infrastructures

An evaluation of the state of time synchronization on leadership class supercomputers

Contact Info

Product

Resources

About