Abstract:Summary
Containers are gaining popularity over virtual machines as they provide the advantages of virtualization with the performance of near bare metal. The uniformity of support provided by Docker containers across different cloud providers makes them a popular choice for developers. Evolution of microservice architecture allows complex applications to be structured into independent modular components making them easier to manage. High‐performance computing (HPC) applications are one such application to be d… Show more
“…Maliszewski et al [24] investigated the performance of scientific workloads with single or multi-tenant instances in a single node, where each tenant held its independent application among other tenants. Jha et al have studied HPC microservices in different container environments [17,18]. Their work includes flexible deployments for HPC applications on a single node, from running a single or multiple applications in a single container, to running multiple containers each holding a single application.…”
Section: Related Workmentioning
confidence: 99%
“…However, there is a lack of research on multi-container deployment solutions for a singletenant multi-process HPC workload. Unlike the multi-container deployments holding workloads for multiple tenants [16][17][18]24], using multiple containers to package a single-tenant multi-process/thread HPC workload refers to partitioning the processes or threads belonging to each application into different containers, obtaining in that way a finer-grained deployment. Whereas few works include experiments with different container granularity [8,34,35], none of them provide a deep understanding of the impact of such multi-container deployments on the performance of HPC workloads, which considers different containerization technologies, container grain sizes, and hardware platforms.…”
The High-Performance Computing (HPC) community has recently started to use containerization to obtain fast, customized, portable, flexible, and reproducible deployments of their workloads. Previous work showed that deploying an HPC workload into a single container can keep bare-metal performance. However, there is a lack of research on multi-container deployments that partition the processes belonging to each application into different containers. Partitioning HPC applications have shown to improve their performance on virtual machines by allowing them to be set affinity to a NUMA (Non-Uniform Memory Access) domain. Consequently, it is essential to understand the performance implications of distinct multi-container deployment schemes for HPC workloads, focusing on the impact of the container granularity and its combination with processor and memory affinity. This paper presents a systematic performance comparison and analysis of multi-container deployment schemes for HPC workloads on a single-node platform, which considers different containerization technologies (including Docker and Singularity), two different platform architectures (UMA and NUMA), and two application subscription modes (exactly-subscription and over-subscription). Our results indicate that finer-grained multi-container deployments, on one side, can benefit the performance of some applications with low inter-process communication, especially in over-subscribed scenarios and when combined with affinity but, on the other side, they can incur some performance degradation for communicationintensive applications when using containerization technologies that deploy isolated network namespaces.
“…Maliszewski et al [24] investigated the performance of scientific workloads with single or multi-tenant instances in a single node, where each tenant held its independent application among other tenants. Jha et al have studied HPC microservices in different container environments [17,18]. Their work includes flexible deployments for HPC applications on a single node, from running a single or multiple applications in a single container, to running multiple containers each holding a single application.…”
Section: Related Workmentioning
confidence: 99%
“…However, there is a lack of research on multi-container deployment solutions for a singletenant multi-process HPC workload. Unlike the multi-container deployments holding workloads for multiple tenants [16][17][18]24], using multiple containers to package a single-tenant multi-process/thread HPC workload refers to partitioning the processes or threads belonging to each application into different containers, obtaining in that way a finer-grained deployment. Whereas few works include experiments with different container granularity [8,34,35], none of them provide a deep understanding of the impact of such multi-container deployments on the performance of HPC workloads, which considers different containerization technologies, container grain sizes, and hardware platforms.…”
The High-Performance Computing (HPC) community has recently started to use containerization to obtain fast, customized, portable, flexible, and reproducible deployments of their workloads. Previous work showed that deploying an HPC workload into a single container can keep bare-metal performance. However, there is a lack of research on multi-container deployments that partition the processes belonging to each application into different containers. Partitioning HPC applications have shown to improve their performance on virtual machines by allowing them to be set affinity to a NUMA (Non-Uniform Memory Access) domain. Consequently, it is essential to understand the performance implications of distinct multi-container deployment schemes for HPC workloads, focusing on the impact of the container granularity and its combination with processor and memory affinity. This paper presents a systematic performance comparison and analysis of multi-container deployment schemes for HPC workloads on a single-node platform, which considers different containerization technologies (including Docker and Singularity), two different platform architectures (UMA and NUMA), and two application subscription modes (exactly-subscription and over-subscription). Our results indicate that finer-grained multi-container deployments, on one side, can benefit the performance of some applications with low inter-process communication, especially in over-subscribed scenarios and when combined with affinity but, on the other side, they can incur some performance degradation for communicationintensive applications when using containerization technologies that deploy isolated network namespaces.
“…Because Spark executor exchanges task results with each other for fault tolerance, the inter-container communication latency of containers on the same host would be negligible. In order to constrain the performance degradation caused by the resource competition of these containers [ 21 ], we propose a task scheduling mechanism (TSM) to change the containers’ affinities with the hosts based on a weighted average policy. The TSM adjusts the pod affinities to a host after referencing the ratio between the number of completed and failed tasks at the runtime.…”
Remotely-sensed satellite image fusion is indispensable for the generation of long-term gap-free Earth observation data. While cloud computing (CC) provides the big picture for RS big data (RSBD), the fundamental question of the efficient fusion of RSBD on CC platforms has not yet been settled. To this end, we propose a lightweight cloud-native framework for the elastic processing of RSBD in this study. With the scaling mechanisms provided by both the Infrastructure as a Service (IaaS) and Platform as a Services (PaaS) of CC, the Spark-on-Kubernetes operator model running in the framework can enhance the efficiency of Spark-based algorithms without considering bottlenecks such as task latency caused by an unbalanced workload, and can ease the burden to tune the performance parameters for their parallel algorithms. Internally, we propose a task scheduling mechanism (TSM) to dynamically change the Spark executor pods’ affinities to the computing hosts. The TSM learns the workload of a computing host. Learning from the ratio between the number of completed and failed tasks on a computing host, the TSM dispatches Spark executor pods to newer and less-overwhelmed computing hosts. In order to illustrate the advantage, we implement a parallel enhanced spatial and temporal adaptive reflectance fusion model (PESTARFM) to enable the efficient fusion of big RS images with a Spark aggregation function. We construct an OpenStack cloud computing environment to test the usability of the framework. According to the experiments, TSM can improve the performance of the PESTARFM using only PaaS scaling to about 11.7%. When using both the IaaS and PaaS scaling, the maximum performance gain with the TSM can be even greater than 13.6%. The fusion of such big Sentinel and PlanetScope images requires less than 4 min in the experimental environment.
O Kafka é uma plataforma de mensageria e streaming que segue um modelo produtor-consumidor. Visando garantir a entrega de mensagens, o Kafka apresenta um mecanismo de Reconhecimento Positivo (ack). Apesar de existirem três níveis distintos de configuração padrão para ack, os mesmos apresentam restrições de confiabilidade ou desempenho durante transmissões em redes instáveis, obrigando os usuários a priorizar um destes requisitos. Este trabalho propõe o Reliable Ack (rAck), uma configuração para transmissão de mensagens baseada no monitoramento, identificação e recuperação de mensagens em caso de perda. Experimentos foram conduzidos com o objetivo de comparar a configuração rAck com os níveis padrão de ack do Kafka, permitindo a observação de que a configuração proposta é viável para transmissões de mensagens em redes suscetíveis a perdas de pacotes, recuperando mensagens e apresentando desempenho satisfatório.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.