A Tool for Bottleneck Analysis and Performance Prediction for GPU-Accelerated Applications

Madougou, Souley; Vărbănescu, Ana Lucia; Laat, Cees de; Nieuwpoort, Rob V. van

doi:10.1109/ipdpsw.2016.198

Cited by 12 publications

(8 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, the complexity of scientific and technological concerns, as well as the automation of industrial and economic operations, necessitated a highly sophisticated use of computer resources [1][2][3][4]. This has prompted the development of a number of technologies, ranging from the creation of more powerful CPUs, GPUs, TPUs, and other hardware components to the construction of computer clusters capable of working on the same problem in parallel [5][6][7].…”

Section: Related Workmentioning

confidence: 99%

Deploying Efficiently Modern Applications on Cloud

2022

View full text Add to dashboard Cite

This study analyses some of the leading technologies for the construction and configuration of IT infrastructures to provide services to users. For modern applications, guaranteeing service continuity even in very high computational load or network problems is essential. Our configuration has among the main objectives of being highly available (HA) and horizontally scalable, that is, able to increase the computational resources that can be delivered when needed and reduce them when they are no longer necessary. Various architectural possibilities are analysed, and the central schemes used to tackle problems of this type are also described in terms of disaster recovery. The benefits offered by virtualisation technologies are highlighted and are bought with modern techniques for managing Docker containers that will be used to build the back-end of a sample infrastructure related to a use-case we have developed. In addition to this, an in-depth analysis is reported on the central autoscaling policies that can help manage high loads of requests from users to the services provided by the infrastructure. The results we have presented show an average response time of 21.7 milliseconds with a standard deviation of 76.3 milliseconds showing excellent responsiveness. Some peaks are associated with high-stress events for the infrastructure, but the response time does not exceed 2 s even in this case. The results of the considered use case studied for nine months are presented and discussed. In the study period, we improved the back-end configuration and defined the main metrics to deploy the web application efficiently..

show abstract

Section: Related Workmentioning

confidence: 99%

Deploying Efficiently Modern Applications on Cloud

2022

View full text Add to dashboard Cite

show abstract

Section: Related Workmentioning

confidence: 99%

Deploying Efficiently Modern Applications on Cloud

Perri¹,

Simonetti²,

Gervasi³

2022

Preprint

View full text Add to dashboard Cite

This study analyses some of the leading technologies for the construction and configuration of IT infrastructures to provide services to users. For modern applications, guaranteeing service continuity even in very high computational load or network problems is essential. Our configuration has among the main objectives of being highly available (HA) and horizontally scalable, that is, able to increase the computational resources that can be delivered when needed and reduce them when they are no longer necessary. Various architectural possibilities are analysed, and the central schemes used to tackle problems of this type are also described in terms of disaster recovery. The benefits offered by virtualisation technologies are highlighted and are bought with modern techniques for managing Docker containers that will be used to build the back-end of a sample infrastructure related to a use-case we have developed. In addition to this, an in-depth analysis is reported on the central autoscaling policies that can help manage high loads of requests from users to the services provided by the infrastructure. The results we have presented show an average response time of 21.7 milliseconds with a standard deviation of 76.3 milliseconds showing excellent responsiveness. Some peaks are associated with high-stress events for the infrastructure, but the response time does not exceed 2 seconds even in this case. The results of the considered use case studied for nine months are presented and discussed. In the study period, we improved the back-end configuration and defined the main metrics to deploy the web application efficiently.

show abstract

“…Statistical and machine learning methods [25,52,31] are often implemented as separate frameworks that are highly automated and require additional training so that the neural network correctly recognises patterns and behaviour of the kernel code. The network learns the kernel's instructions composition given a set of input data and is then able to reason about the performance of subsequent kernel executions for different launch configurations.…”

Section: Related Workmentioning

confidence: 99%

Modelling and simulation of GPU processing in the MERPSYS environment

Gajger

Czarnul

2018

SCPE

View full text Add to dashboard Cite

In this work, we evaluate an analytical GPU performance model based on Little's law, that expresses the kernel execution time in terms of latency bound, throughput bound, and achieved occupancy.We then combine it with the results of several research papers, introduce equations for data transfer time estimation, and finally incorporate it into the MERPSYS framework, which is a general-purpose simulator for parallel and distributed systems.The resulting solution enables the user to express a CUDA application in a MERPSYS editor using an extended Java language and then conveniently evaluate its performance for various launch configurations using different hardware units.We also provide a systematic methodology for extracting kernel characteristics, that are used as input parameters of the model.The model was evaluated using kernels representing different traits and for a large variety of launch configurations.We found it to be very accurate for computation bound kernels and realistic workloads, whilst for memory throughput bound kernels and uncommon scenarios the results were still within acceptable limits.We have also proven its portability between two devices of the same hardware architecture but different processing power.Consequently, MERPSYS with the theoretical models embedded in it can be used for evaluationof application performance on various GPUs and used for performance prediction and e.g. purchase decision making.

show abstract

A Tool for Bottleneck Analysis and Performance Prediction for GPU-Accelerated Applications

Cited by 12 publications

References 17 publications

Deploying Efficiently Modern Applications on Cloud

Deploying Efficiently Modern Applications on Cloud

Deploying Efficiently Modern Applications on Cloud

Modelling and simulation of GPU processing in the MERPSYS environment

Contact Info

Product

Resources

About