Cloud computing manages system resources such as processing, storage, and networking by providing users with multiple virtual machines (VMs) as needed. It is one of the rapidly growing fields that come with huge computational power for scientific workloads. Currently, the scientific community is ready to work over the cloud as it is considered as a resource-rich paradigm. The traditional way of executing scientific workloads on cloud computing is by using virtual machines. However, the latest emerging concept of containerization is growing more rapidly and gained popularity because of its unique features. Containers are treated as lightweight as compared to virtual machines in cloud computing. In this regard, a few VMs/containers-associated problems of performance and throughput are encountered because of middleware technologies such as virtualization or containerization. In this paper, we introduce the configurations of VMs and containers for cloud-based scientific workloads in order to utilize the technologies to solve scientific problems and handle their workloads. This paper also tackles throughput and efficiency problems related to VMs and containers in the cloud environment and explores efficient resource provisioning by combining four unique methods: hyperthreading (HT), vCPU cores selection, vCPU affinity, and isolation of vCPUs. The HEPSCPEC06 benchmark suite is used to evaluate the throughput and efficiency of VMs and containers. The proposed solution is to implement four basic techniques to reduce the effect of virtualization and containerization. Additionally, these techniques are used to make virtual machines and containers more effective and powerful for scientific workloads. The results show that allowing hyperthreading, isolation of CPU cores, proper numbering, and allocation of vCPU cores can improve the throughput and performance of virtual machines and containers.

Data are important and ever growing in data-intensive scientific environments. Such research data growth requires data storage systems that play pivotal roles in data management and analysis for scientific discoveries. Redundant Array of Independent Disks (RAID), a well-known storage technology combining multiple disks into a single large logical volume, has been widely used for the purpose of data redundancy and performance improvement. However, this requires RAID-capable hardware or software to build up a RAID-enabled disk array. In addition, it is difficult to scale up the RAID-based storage. In order to mitigate such a problem, many distributed file systems have been developed and are being actively used in various environments, especially in data-intensive computing facilities, where a tremendous amount of data have to be handled. In this study, we investigated and benchmarked various distributed file systems, such as Ceph, GlusterFS, Lustre and EOS for data-intensive environments. In our experiment, we configured the distributed file systems under a Reliable Array of Independent Nodes (RAIN) structure and a Filesystem in Userspace (FUSE) environment. Our results identify the characteristics of each file system that affect the read and write performance depending on the features of data, which have to be considered in data-intensive computing environments.

Deep learning is a growing technique used to solve complex artificial intelligence (AI) problems. Large-scale deep learning has become a significant issue as a result of the expansion of datasets and the complexity of deep learning models. For training large-scale models, the cloud can be used as a distributed HPC (high-performance computing) tool with benefits in cost and flexibility. However, one of the major performance barriers in distributed deep learning in a distributed HPC environment is the network. The performance is often limited by heavy traffic like many stochastic gradient descent transfers for distributed communication. There are many network studies in distributed deep learning to solve these problems, but most research only focuses on improving communication performance and applying new methods or algorithms like overlapping parameter synchronization to minimize communication delay rather than considering the actual network. In this paper, we are focusing on the actual network, especially in a distributed HPC environment. In such an environment, if cluster nodes are assigned to different zones/regions which means a set of an appropriate number of distributed HPC nodes when performing distributed deep learning tasks, performance degradation due to network delay may occur. The proposed network optimization algorithm ensures that distributed work is placed in the same zone as much as possible to reduce network delay. Furthermore, scoring using network monitoring tools like loss, delay, and throughput is applied to select the optimal node within the zone. Our proposal has been validated on the Kubernetes platform, an open source orchestrator for the automatic management and deployment of micro-services. The performance of distributed deep learning is improved through the proposed scheduler.

Here, we introduce the current stage and future directions of the wireless infrastructure of the Korea Research Environment Open NETwork (KREONET), a representative national research and education network in Korea. In 2018, ScienceLoRa, a pioneering wireless network infrastructure for scientific applications based on low-power wide-area network technology, was launched. Existing in-service applications in monitoring regions, research facilities, and universities prove the effectiveness of using wireless infrastructure in scientific areas. Furthermore, to support the more stringent requirements of various scientific scenarios, ScienceLoRa is evolving toward ScienceIoT by employing high-performance wireless technology and distributed computing capability. Specifically, by accommodating a private 5G network and an integrated edge computing platform, ScienceIoT is expected to support cutting-edge scientific applications requiring high-throughput and distributed data processing.

Large scientific experimental facilities currently are generating a tremendous amount of data. In recent years, the significant growth of scientific data analysis has been observed across scientific research centers. Scientific experimental facilities are producing an unprecedented amount of data and facing new challenges to transfer the large data sets across multi continents. In particular, these days the data transfer is playing an important role in new scientific discoveries. The performance of distributed scientific environment is highly dependent on high-performance, adaptive, and robust network service infrastructures. To support large scale data transfer for extreme-scale distributed science, there is the need of high performance, scalable, end-to-end, and programmable networks that enable scientific applications to use the networks efficiently. We worked on the AmoebaNet solution to address the problems of a dynamic programmable network for bulk data transfer in extreme-scale distributed science environments. A major goal of the AmoebaNet project is to apply software-defined networking (SDN) technology to provide “Application-aware” network to facilitate bulk data transfer. We have prototyped AmoebaNet’s SDN-enabled network service that allows application to dynamically program the networks at run-time for bulk data transfers. In this paper, we evaluated AmoebaNet solution with real world test cases and shown that how it efficiently and dynamically can use the networks for bulk data transfer in large-scale scientific environments.