• KNOM Review
• /
• v.23 no.1
• /
• pp.43-50
• /
• 2020

As General Purpose Graphics Processing Unit (GPGPU) recently plays an essential role in high-performance computing, several cloud service providers offer GPU service. Most cluster orchestration platforms in a cloud environment using containers allocate the integer number of GPU to jobs and do not allow a node shared with other jobs. In this case, resource utilization of a GPU node might be low if a job does not intensively require either many cores or large size of memory in GPU. GPU virtualization brings opportunities to realize kernel concurrency and share resources. However, performance may vary depending on characteristics of applications running concurrently and interference among them due to resource contention on a node. This paper proposes GPU container co-execution framework with multiple server creation and execution based on Kubernetes, container orchestration platform for measuring interference which may be occurred by sharing GPU resources. Performance changes according to scheduling policies were investigated by executing several jobs on GPU. The result shows that optimal scheduling is not possible only considering GPU memory and computing resource usage. Interference caused by co-execution among applications is measured using the framework.

• Journal of Internet Computing and Services
• /
• v.8 no.5
• /
• pp.21-33
• /
• 2007

An overlay network is a virtual network which is constructed on top of a physical computer network. A node in the overlay network is connected through virtual or logical links, where each link corresponds to a path of the links in the underlying physical network. Overlay networks are suitable for sharing heterogeneous resources in distributed environments, However, overlay networks are limited for achieving reliable communication that failure detection in overlay networks is a very important issue. In this paper, we review conditions of conventional failure detection and propose a new approach to failure detection and resilience which can be applied to HRing (Hierarchical Ring) overlay networks. The proposed method consists of the failure detection and the failure resilience phases. Because it utilizes the characteristics of the HRing overlay network for failure detection, it can reduce unnecessary network traffic and provide better scalability and flexibility. We also analyzed and evaluated the performance of the proposed approach through simulations.

• Journal of the Korean Society for information Management
• /
• v.15 no.3
• /
• pp.151-173
• /
• 1998

National Union Catalog presupposes an information infrastructure which permits users to identify and locate materials of all the domestic libraries and to share materials among libraries. Recently, there have been developed national union catalog projects in many countries using Z39.50 protocol implementation. In Korea, multi-union catalogs based on the types of libraries, i.e., public, academic, and special libraries, have been constructed separately. The virtual national union catalog which is a decentralized electronically accessible catalog unified by linking the databases of several institutions using Z39.50 is preferable to a union catalog implemented by a single centralized database. In this paper, hybrid union catalog is suggested with mixtures of centralized union catalogs and Z39.50-based distributed search systems for the development of a national union catalog. It is also suggested that the roles of a national library and centers for union catalog, and relationships between two types of institution for making sound ground for implementing the virtual Korean union catalog.

• The Korean Journal of Applied Statistics
• /
• v.27 no.7
• /
• pp.1171-1185
• /
• 2014

Rhipe that integrates R and Hadoop environment, made it possible to process and analyze massive amounts of data using a distributed processing environment. In this paper, we implemented multiple regression analysis using Rhipe with various data sizes of actual data and simulated data. Experimental results for comparing the computing speeds of pseudo-distributed and fully-distributed modes for configuring Hadoop cluster, showed fully-distributed mode was more fast than pseudo-distributed mode and computing speeds of fully-distributed mode were faster as the number of data nodes increases. We also compared the performance of our Rhipe with stats and biglm packages available on bigmemory. The results showed that our Rhipe was more fast than other packages owing to paralleling processing with increasing the number of map tasks as the size of data increases.

• Journal of the Korea Society of Computer and Information
• /
• v.19 no.7
• /
• pp.59-70
• /
• 2014

This paper propose a new wireless multi-hop routing protocol that takes the hierarchical mesh of the hybrid WMN into account. WMN that is possible to provide various applications of wireless networks still has many open issues that should be solved despite the studies carried out over a decade. Especially, in routing protocol area, a problem degrading the routing efficiency by applying one of the routing protocols, which are designed for the MANET, to the hybrid WMN be solved above all. For the improvement of the routing performance, both good routing protocol and metric are essential. However, the recent studies are only concentrated in routing metric by use of the cross-layer design. Therefore, this paper is dedicated to the routing protocol that is essential for the performance of the routing but needed more studies. The proposed protocol in this paper is reactive, and designed to reorganize the hybrid WMN with several pseudo domains, and carry out domain-based route decision. By the simulation result for the performance analysis of the proposed protocol, the average delay for the route decision was decreased by 43% compared to AODV that is the typical reactive protocol.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.18 no.11
• /
• pp.1689-1700
• /
• 1993

WDM(Wavelength Division Multiplexing) based-on fixed wavelengths is a new means of utilizing the bandwidth of optical fibers. In this Paper, an improved virtual topology called "(a, $\beta$,${\gamma}$) ShuffleNet " is introdced for designing large-scale WDM switches. The proposed one is an architecture created by vertically stacking x planes of a ($\beta$,${\gamma}$) ShuffleNet in parallel via $\beta$r nodes called "bridge nodes" so that N-by-N(N=(x*$\beta$${\gamma}$*${\gamma}$) switching is achieved based on the self-routing algorithm for each ($\beta$,${\gamma}$) ShuffleNet. With the topological parallelism, in contrast to the conventional virtual topologies, the diameter of 2${\gamma}$ hops can be fixed and high utilization and performance are provided while N increases. Such a scalability characteristic allows to design a growable broadband switch. As for the delay, we show that the traffic locality, due to the topological feature. result in low delay characteristics.lay characteristics.

• The KIPS Transactions:PartC
• /
• v.11C no.7 s.96
• /
• pp.937-950
• /
• 2004

In this paper, we present the design and implementation of new execution environment named Customizable Architecture for Flexible Execution Environment(CAFEs) that supports the mechanism of protocol customization and service referencing. We introduce a new concept as micro protocols and active services to enhance software reusability. Micro protocol represents a specific algorithm or functionality of existing network protocol and the active service is in charge of binding legacy application and releasing the active network oriented services. The pro-posed active network execution environment is made up of three parts, virtual network system, protocol and service manager, event engine. First, the virtual network system is used to connect each active nodes using virtual network channels which are based on multiple existing protocol layers. Second, the protocol and service manager is responsible for composing micro protocols and active services to develop new network proto-col and service easily. Finally, the event engine is used to detect the automatic transition of system components using event delegation model. To verify the CAFTs, we have an experiment about the delivery of web contents which are suitable for the user's terminals in the wireless Internet environment. As a result. we are able to obtain the availability of developed execution environment.

• The Transactions of the Korea Information Processing Society
• /
• v.5 no.5
• /
• pp.1320-1332
• /
• 1998

We propose Multiph which is a virtual topology for a reliable group communication platform on dynamic high speed network. We also define group communication primitives for transmission of messages and group management. Multiph consists of a set of nodes which are process groups and a set of edges which are the connection between nodes providing the route of messages. Processes can send and receive messages overlapped in a several multicast groups. Overlapped process organizes non-overlapped logical group, which called meta-group. Multiph guarantees ordering of messages that received at the destination group. It has no extra node that decreases system performance. Although the group has dynamic group change to leave and add process, it does not reconstruct the existing platform structure.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.1B
• /
• pp.45-54
• /
• 2003

In the Protocol-Independent Multicast Sparse Mode(PIM-SM), we need a careful selection scheme for the Rendezvous Point(RP) which influences much on the QoS due to the delay among multicast group members and packet loss The QoS based RP selection schemes choose RPs which satisfy the restriction conditions such as bandwidth, delay, and other QoS parameters In this paper, we propose a new RP selection scheme which is a variation of the group-based RP selection scheme The new algorithm, MCT(Maximum Cross Tree), is implemented by taking advantages of the topology-based selection scheme and the group-based selection scheme as well In order to verify the proposed algorithm, we first measure the multicast traffic data of the Virtual Conference System implemented on the IPv6 network via KOREN, then expand the results to two types of network models and analyze the performance by computer simulation.

• Proceedings of the Korean Information Science Society Conference
• /
• 2010.06c
• /
• pp.185-190
• /
• 2010

최근 위치 인식 기술과 휴대 장치의 발달로 인해 이동하는 객체를 기반으로 하는 위치 기반 서비스(Location Based Service, LBS)의 관심이 점점 증가하고 있고 그에 관련된 연구들이 활발하게 진행되고 있다. 이동 객체의 응용은 빈번하게 변경되는 이동객체의 위치정보를 효과적으로 처리할 수 있는 색인구조를 필요로 한다. 위치정보를 색인하기 위해 R-tree 기반의 색인들이 제안되었다. 하지만 R-tree는 변경보다는 검색 연산에 초점이 맞추어진 색인구조이기 때문에 잦은 변경을 다루어야 하는 이동객체 환경에 적합하지 못하다. 최근 이러한 객체의 빠른 위치 변경을 지원하는 그리드 기반의 색인 구조가 제안되었다. 하지만 셀의 객체 점유율에 따라 검색 속도가 저하되는 단점은 여전히 해결되지 못하고 있다. 이러한 단점은 객체들이 특정 영역에 몰리는 경우 또는 그리드의 해상도를 잘못 지정한 경우 더욱 부각된다. 본 논문에서는 이러한 단점을 해결하기 위해 가상 셀 기반의 색인 구조를 제안한다. 데이터 페이지에 객체의 점유율을 보장하기 위해 여러 개의 인접한 셀들의 데이터를 한 데이터 페이지에 함께 저장한다. 공간 채움 곡선을 기반으로 순서화된 셀들로 셀의 인접성을 결정한다. 또한 공간 채움 곡선의 차수를 동적으로 지정하여 객체가 집중된 셀에 대해서는 셀의 단위 크기를 작게 지정한다. 뿐만 아니라 셀을 표현하기 위한 식별자를 위해 비트를 이용한 표현식을 제안하였다. 이로 인해 노드의 팬아웃을 증가시켰고, 저장공간을 절약하였다. 실험을 통해서 제안하는 색인 기법의 우수성을 증명하였다.