• Journal of Platform Technology
• /
• v.11 no.4
• /
• pp.71-79
• /
• 2023

In this paper, we analyze the characteristics of machine learning workloads and, based on them, propose a distributed in-memory caching technique to improve the performance of machine learning workloads. The core of machine learning workload is model training, and model training is a computationally intensive task. Performing machine learning workloads in a Kubernetes-based cloud environment in which the computing framework and storage are separated can effectively allocate resources, but delays can occur because IO must be performed through network communication. In this paper, we propose a distributed in-memory caching technique to improve the performance of machine learning workloads performed in such an environment. In particular, we propose a new method of precaching data required for machine learning workloads into the distributed in-memory cache by considering Kubflow pipelines, a Kubernetes-based machine learning pipeline management tool.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2009.04a
• /
• pp.1116-1119
• /
• 2009

최근 그룹 통신 분야에서 전통적으로 대표되는 IP 멀티캐스트의 대안기술로서 호스트기반 멀티캐스트 및 P2P시스템은 기존의 IP 멀티캐스트의 단점을 극복하고자 하는 연구이다. 이러한 호스트기반 멀티캐스트는 사용자 중심의 서비스를 제공함으로써 다양한 네트워크 환경에서 누구나 사용자들이 효율적으로 데이터를 공유하게 해주고 있다. 본 논문에서는 이러한 협동 기반의 호스트 멀티캐스트를 사용하여 미디어 서비스를 제공하는 VOD 시스템 전반에 관한 연구를 수행하고 이러한 VOD의 서비스를 고도화 하고자 한다. 보다 구체적으로는 패칭 기법을 사용하여 사용자간의 전달 지연을 최소화 할 수 있는 효율적인 자료 전송 경로를 설정 방법과 주문형 비디오 서버를 위한 현실적인 호스트 기반 멀티캐스트 트리 구성 모델을 제시하고자 한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2009.04a
• /
• pp.942-945
• /
• 2009

유한차분시간영역 방법은 전자기파 관련 분야의 전산모사에 많이 사용되는 수치해석기법이다. 이 방법을 이용하여 구현한 전산모사 프로그램은 많은 계산 자원 필요로 하기 때문에 병렬 계산 환경을 이용하게 되는 경우가 많다. 병렬 계산 환경에서 전산모사를 수행할 경우, 병렬로 수행되는 각 프로세스 간의 통신 속도와 네트워크의 지연 시간은 계산의 병목 현상을 초래하여 전체적인 성능을 저하시키는 원인이 된다. 따라서, 본 논문에서는 MPI의 지속 통신 메커니즘을 이용하여 병렬 프로세스 간 동기화 속도를 증가시킴으로써 유한차분시간영역 전산모사 프로그램에서의 MPI 통신 성능의 향상을 꾀하고, 그 결과를 그래프로 도시하였다. 또한 기존의 양방향 통신과 단방향 통신 메커니즘을 사용했을 때의 성능과 비교/분석하여, 병렬 유한차분시간영역 전산모사 프로그램에 있어서 지속 통신 메커니즘의 장/단점을 제시하고, 그 효용성에 관해 논의한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2008.11a
• /
• pp.1195-1198
• /
• 2008

본 논문은 다중 Tier 상에서 온라인 서비스 대량 데이타 처리를 빠르고 정확하게 클라이언트에 전달하는 기법을 제안한다. Tier 가 많은 온라인 서비스상에서 대량의 데이타를 빠르게 처리하는 데에는 많은 어려움이 있다. Tier 간 지연 시간의 최소화, 네트워크 대역폭를 고려한 트란잭션(Transaction)의 적절한 분할 통신, 이 기종간의 데이타 변환 시 처리속도 개선 등이 해결해야 할 주요한 요건이라고 할 수 있다. 하지만 이러한 문제들이 해결된다고 해서 괄목할 만한 성능의 개선은 쉽게 나타나지 않는다. 그 이유는 바로 Partial Query에 의한 데이타 통신이 꾸준히 반복 발생하기 때문이다. 온라인 서비스의 특성상 대량 데이타는 많은 사용자의 효율적인 트란잭션 처리를 위하여 분할(Partial) 처리되어 통신하는 방식을 기준으로 사용하고 있다. 이러한 방식을 준수 하기 위해서는 데이타 사이즈에 비례하는 반복의 증가가 불가피하다. 그래서 반복 횟수를 줄이는데 포커스를 두고 온라인 서비스 대량 데이타 처리에 대한 성능 데스트를 진행한 결과 반복이 최소화 될수록 성능은 최대한으로 유지되며, 다른 어떤 기술적인 요소를 개선하는 것보다 큰 효과를 볼 수 있음을 알 수 있었다.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.46 no.3
• /
• pp.75-88
• /
• 2009

This paper presents a video quality assessment method based on quality degradation factors of real-time multimedia streaming services. The video quality degradation is caused by video source compression and network states. In this paper, we propose a blocky metric on an image domain to measure quality degradation by video compression. In this paper, the proposed boundary strength index for the blocky metric is defined by ratio of the variation of two pixel values adjacent to $8{\times}8$ block boundary and the average variation at several pixels adjacent to the two boundary pixels. On the other hand, unnatural image movement caused by network performance deterioration such as jitter and delay factors can be observed. In this paper, a temporal-Jerkiness measurement method is proposed by computing statistics of luminance differences between consecutive frames and play-time intervals between frames. The proposed final Perceptual Video Quality Metric (PVQM) is proposed by consolidating both blocking strength and temporal-jerkiness. To evaluate performance of the proposed algorithm, the accuracy of the proposed algorithm is compared with Difference of Mean Opinion Score (DMOS) based on human visual system.

• The KIPS Transactions:PartC
• /
• v.19C no.3
• /
• pp.191-208
• /
• 2012

Proxy Mobile IPv6 (PMIPv6) is designed to provide network-based mobility management support to an MN without any involvement of the MN in the mobility related signalling, hence, the proxy mobility entity performs all related signalling on behalf of the MN. The new principal functional entities of PMIPv6 are the local mobility anchor (LMA) and the mobile access gateway (MAG). In PMIPv6, all the data traffic sent from the MN gets routed to the LMA through a tunnel between the LMA and the MAG, but it still has the single point of failure (SPOF) and bottleneck state of traffic. To solve these problems, various approaches directed towards PMIPv6 performance improvements such as route optimization proposed. But these approaches add additional signalling to support MN's mobility, which incurs extra network overhead and still has difficult to apply to multiple-LMA networks. In this paper, we propose a improved route optimization in PMIPv6-based multiple-LMA networks. All LMA connected to the proxy internetworking gateway (PIG), which performs inter-domain distributed mobility control. And, each MAG keeps the information of all LMA in PMIPv6 domain, so it is possible to perform fast route optimization. Therefore, it supports route optimization without any additional signalling because the LMA receives the state information of route optimization from PIG.

• Journal of KIISE:Information Networking
• /
• v.36 no.1
• /
• pp.12-23
• /
• 2009

Sensor networks deployed in battlefields to support military operations or deployed in natural habitats to monitor the rare wildlifes must take account of protection of the location of valuable assets(i.e., soldiers or wildlifes) from an adversary or malicious tracing as well as the security of messages in transit. In this paper we propose a routing protocol GSLP(GPSR-based Source-Location Privacy) that is capable of enhancing the location privacy of an active source node(i.e., message-originating node) in the presence of multiple dormant sources(i.e., nodes lying nearby an asset whose location needs to be secured). Extended is a simple, yet scalable, routing scheme GPSR(greedy perimeter stateless routing) to select randomly a next-hop node with a certain probability for randomizing paths and to perform perimeter routing for detouring dormant sources so that the privacy strength of the active source, defined as safety period, keeps enhanced. The simulation results obtained by increasing the number of dormant sources up to 1.0% of the total number of nodes show that GSLP yields increased and nearly invariant safety periods, while those of PR-SP(Phantom Routing, Single Path), a notable existing protocol for source-location privacy, rapidly drop off as the number of dormant sources increases. It turns out that delivery latencies of GSLP are roughly less than two-fold of the shortest path length between the active source and the destination.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.9
• /
• pp.11-20
• /
• 2014

For Web-based applications in IoT environment, IETF CoRE WG has standardizing the CoAP. One of limitations of CoAP is that CoAP standard does not consider the mobility management of the CoAP sensor node. In this paper, we propose the mobility management protocol of CoAP sensor node by considering the characteristics of the constrained network. The proposed mobility management protocol supports for Web client to be transmitted the sensing data from CoAP node reliably while the CoAP sensor moves into different wireless networks. To do this, we designed the architecture with the separate IP address management of CoAP sensor node and presented the mobility management protocol, which includes the holding and binding mode, in order to provide the reliable transmission. Finally, the numerical analysis and simulation with NS2 tool have been done for the performance evaluation in terms of the handover latency and packet loss with comparing the proposed mobility management protocol with other the existing mobility management protocols. The performance result shows that the proposed mobility management can provide the transmission of sensing data without the packet loss comparing with the existing mobility management protocol reliably.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.46 no.4
• /
• pp.87-93
• /
• 2009

Wireless mesh networks (WMNs) are emerging technologies that provide ubiquitous environments and wireless broadband access. The aggregate capacity of WMNs can be improved by minimizing the effect of channel interference. The IEEE 802.11b/g standard which is mainly used for the network interface technology in WMNs provides 3 multiple channels. We must consider the channel scanning delay and the channel dependency problem to effectively assign channels in like these multi-channel WMNs. This paper proposes a cluster-based channel assignment (CB-CA) algorithm for multi-channel WMNs to solve such problems. The CB-CA does not perform the channel scanning and the channel switching through assigning co-channel to the inter-cluster head (CH) links. In the CB-CA, the communication between the CH and cluster member (CM) nodes uses a channel has no effect on channels being used by the inter-CH links. Therefore, the CB-CA can minimize the interference within multi-channel environments. Our simulation results show that CB-CA can improve the performance of WMNs.

• Journal of The Korean Association For Science Education
• /
• v.38 no.4
• /
• pp.565-574
• /
• 2018

The purpose of this study is to investigate the features of science classes of Science Core Schools (SCSs) perceived by students. 654 students from 14 SCSs were surveyed with two open-ended questions on the features of science classes. The students' responses were analyzed with NetMiner 4.5, in terms of the centrality (of betweenness and of degree) analysis and the community analysis. The results of the research are as follows: (1) the science classes of SCSs were perceived by students to be of the environment of free questioning, active participation and communication, caring teacher, more science experiments and advanced contents, and knowledge sharing; (2) science classes in SCSs were perceived to be different from those of ordinary high schools because SCSs provide more opportunities for science-related special courses (like project work, advanced science subjects), extra-curricular activities, inquiry and research activities, school supports, hard-working classroom environment, longer studying hours, R&E and club activities. The students' perceptions of SCS science classes appear to be in line with the characteristics of 'good' science lessons from previous studies. The SCS project itself and the features of SCS science classes would help us to see how we introduce educational innovations into actual schools.