Ko, Young-Woong

Jin, Hyun-Wook

Jung, Jin-Hwan

Choi, Jin-Hee

Byun, Sang-Seon

Ryu, Eun-Seok

Hwang, Jae-Hyun

Yoo, See-Hwan

Hong, Cheol-Ho

Ahn, Sung Won

Park, Hyunchan

Kim, Young-Pil

Lee, Jung-Hwan

Lee, Chiyoung

Kim, Sewon

Yang, Gyeongsik

Lee, Kyungwoon

Yoo, Yeonho

Ryu, Kyung-Sang

Kim, Tae-Sung

Oh, Seok-Keun

Kwon, Soon-Cheol

Kim, Il-Joong

Lim, Min-Hee

Heo, Yoon-Jeong

Yang, Soon-Seop

Cha, Gyu-Il

Kim, Hyo-Seon

Heo, Jae-Seok

Jeon, Seung-Hyub

Kim, Young-Ho

Park, Ji-Hyun

Park, Jeong-Ae

Kim, Il-Jin

Cho, Gook-Hyang

Ah, Jae-Yong

Kim, Min-Ja

Kim, Jin-Ug

Han, Jong-Min

Yun, Min-Hong

Jung, Jae-Gyu

Lee, ShinHyoung

Song, In-Jun

Kim, Kyung-Yun

Son, Jae-Woong

Hwang, Jung-Sub

Han, Seung-Jin

Kang, Jin-Kyu

Woo, Soo-Chang

Choi, Sun-Oh

Lee, Young-Rim

Jo, Soon-Haeng

Lee, Junseok

Park, Chan-Woong

Park, Miri

Kim, Tae-Kyoung

Jung, Jae-Woo

Lee, Tae-Hoon

Kang, Seung-Yeop

Jo, Jae-Hyun

Kwak, Keun-Hwan

Park, Dae-Young

Kim, Do-Jung

Ahn, Woo-Geun

Lim, Gyu-Bo

Kwon, Gi-Dung

Choi, Jae-Won

Kim, Sae-ra

Kang, Eun-Mi

Kwon, Yong-Jin

Yoo, Sung-Bae

Jang, Youngjun

Kwon, Jaehwan

Jo, Hanchan

Park, Sun-Kyo

Yun, Jong-Hee

Jeon, Moo-Woong

Kim, Beom-Joon

Bae, Hee-jae

Ahn, Chang-jin

Uh, Yoo-Kyung

Lee, Sang-Beom

Oh, Seung-Hoon

Kim, Young-Hyun

Park, Hae-Jung

Lee, Kyoung-Woon

Park, Geun-Yeong

Kang, Dae-Myeong

Moon, Tae-Ho

Oh, Suk-Young

Park, Joon-Young

Lim, Jea-Min

Ko, Young-Geun

Lee, Sang-Hun

Kang, Dong-Hwa

Kim, Hyoung-Tae

Yoo, Hyung-Sic

Ko, Hyun-Seok

Yu, So-Mang

Operating Systems Lab in Korea University was established in 1995.
Our research group aims to construct
next-generation computer systems
achieving
high-performance, robustness, and security.
The target systems are deep learning systems, cloud platform, network OS, hypervisors etc.
과학기술정보통신부 지정
SW스타랩
과학기술정보통신부,
한국과학기술기획평가원 선정
국가 R&D 중 ICT분야
유일한 “최우수” 성과

한국어 설명이 없습니다. Operating Systems Lab in Korea University was established in 1995.
Our research group aims to construct
next-generation computer systems
achieving
high-performance, robustness, and security.
The target systems are deep learning systems, cloud platform, network OS, hypervisors etc.
과학기술정보통신부 지정
SW스타랩
과학기술정보통신부,
한국과학기술기획평가원 선정
국가 R&D 중 ICT분야
유일한 “최우수” 성과를 번역합니다.

Making Sense of Job Preemption for Distributed Deep Learning Acceleration

Prediction-based GPU Sharing for Distributed Training

DEVICE AND METHOD FOR REAL-TIME AND LIGHTWEIGHT GPU WORKLOAD TRACING AND BOTTLENECK ANALYSIS

서버리스 워크로드의 QoS 보장을 위한 저지연 컨테이너 기법 비교분석

NVIDIA DALI 실행 모드에 따른 JPEG 이미지 전처리 성능 비교 분석

GPU 자원 구성 장치 및 방법

GPU Memory Prediction for Multimodal Model Training

다중 GPU 기반 학습 장치 및 방법

프라이빗 데이터 기반의 딥러닝 모델 특화학습을 위한 컨소시엄 러닝 시스템

One-way delay estimation and its application

Resource analysis of blockchain consensus algorithms in hyperledger fabric

DAoT: Dynamic and energy-aware authentication for smart home appliances in Internet of Things

PARFAIT: A new scheduler framework supporting heterogeneous Xen-ARM schedulers

An Intelligent Search Method for query optimization by semijoins

Eliminating the Performance Anomaly of 802.11 b

On the resource management of kubernetes

qCon: QoS-aware network resource management for fog computing

VADI: GPU virtualization for an automotive platform

Method for adjusting a transmission rate to obtain the optimum transmission rate in a mobile ad hoc network environment

Secure device access for automotive software

Prediction of the resource consumption of distributed deep learning systems

Exploiting NIC architectural support for enhancing IP-based protocols on high-performance networks

Libera for programmable network virtualization

Scalable ROI algorithm for H. 264/SVC-based video streaming

Mobivmm: a virtual machine monitor for mobile phones

An intelligent search

Impact of protocol overheads on network throughput over high-speed interconnects: measurement, analysis, and improvement

Network monitoring for SDN virtual networks

Comprehensive prediction models of control traffic for SDN controllers

Chuck Yoo

Delay estimation is a difficult problem in computer networks. Accurate one-way delay estimation is crucial because it serves a very important role in network performance and thus application design. RTT (Round Trip Time) is often used as an approximation of the delay, but because it is a sum of the forward and reverse delays, the actual one-way delay cannot be estimated accurately from RTT. To estimate one-way delay accurately, this paper proposes a new scheme that analytically derives one-way delay, forward and reverse delay, respectively. We show that the performance of TCP can improve dramatically in asymmetric networks using our scheme. A key contribution of this paper is that our one-way delay estimation is much more accurate than RTT estimation so that TCP can quickly find the network capacity in the slow start phase. Since RTT is the sum of the forward and reverse delays, our scheme can be …

In the blockchain network, the consensus algorithm is used to tolerate node faults with data consistency and integrity, so it is vital in all blockchain services. Previous studies on the consensus algorithm have the following limitations: 1) no resource consumption analysis was done, 2) performance analysis was not comprehensive in terms of blockchain parameters (e.g., number of orderer nodes, number of fault nodes, batch size, payload size), and 3) practical fault scenarios were not evaluated. In other words, the resource provisioning of consensus algorithms in clouds has not been addressed adequately. As many blockchain services are deployed in the form of blockchain-as-a-service (BaaS), how to provision consensus algorithms becomes a key question to be answered. This study presents a kernel-level analysis for the resource consumption and comprehensive performance evaluations of three major …

The Internet of Things (IoT) is beneficial to individuals, business, and society because it creates a wide range of value-added services by interconnecting diverse devices and information objects. However, as IoT devices have resource constraints to employ powerful security mechanisms, they are vulnerable to sophisticated security attacks such as the main-in-the-middle (MITM) attack. Therefore, an intelligent authentication mechanism that considers both resource constraints and security is required. In this paper, we suggest a dynamic and energy-aware authentication scheme for the Internet of Things (DAoT). DAoT uses a feedback control scheme to dynamically select an energy-efficient authentication policy. With DAoT, IoT devices with limited resources can be safely interconnected because DAoT finds and adopts the best cost-effective authentication mechanism.

In recent consumer electronics devices, virtualization is widely adopted for diverse reasons such as enhanced reliability, stronger security and better user customizability. Recent CE virtual machines require not only diverse functionalities of GPOS, but also real-time performance of RTOS. However, the current hypervisor-based virtual machine monitors cannot sufficiently support both RTOS and GPOS at the same time. The reason is that the current hypervisor scheduler is biased to only one kind of guest OS and it cannot support heterogeneous schedulers. Therefore, this paper proposes a new scheduler architecture, PARFAIT for both RT and GP guest OSs. Our new scheduler enables to guarantee CPU bandwidth for RTOS and provides fairness among GPOSs using hierarchical structure.

Query optimization strategies based on the reduction of the referenced relations by means of semijoins have received considerable attention. The limitations of such strategies have to do with computational e5ciency (very large search space of semijoin reduction sequences), optimality of the solution (when heuristics are used), and generality of the class of queries allowed (eg, simple queries, chain queries, tree queries). We consider the problem of finding an optimal semijoin sequence that fully reduces a given tree query. We present a new method that “intelligently” navigates the space of all semijoin sequences and returns an optimal solution. We report on experiments that show that our method performs very efficiently; on average, less than five percent of the search space is searched before an optimal solution is found. Other advantages of the method are: 1) ease of implementation, 2) generality of the cost model considered, and 3) ability to handle tree queries with arbitrary target lists.

In this paper, we propose a mechanism to eliminate the performance anomaly of IEEE 802.11b. Performance anomaly happens when nodes that have different transmission rates are in the same wireless cell. All the nodes in the cell might experience the same throughput even though their transmission rates are different because DCF of WLAN provides equal probability of channel access, but it does not guarantee the equal utilization of the wireless channel among the nodes. To reduce such a performance anomaly, we adjust the frame size proportionally depending on the bit rate. Additionally, our scheme eliminates the performance anomaly in multi-hop case. Simulation study shows that our scheme achieves an improvement in the aggregate throughput and the fairness.

Kubernetes is the most popular container orchestration platform that enables users to create and run multiple containers in cloud environments. Kubernetes offers resource management to isolate the resource usage of containers on a host server because performance isolation is an important factor in terms of service quality. This paper investigates whether the resource management of Kubernetes is sufficient to isolate the performance of containers. This is different from previous studies that mostly focuses on efficient resource management rather than the study on performance interference. We evaluate the performance interference 1) between CPU-intensive and networkintensive containers, and 2) between multiple network-intensive containers. Our evaluation results show that containers experience performance degradation by 50% due to the co-located containers even under the resource management of …

Fog computing is a new computing paradigm that employs computation and network resources at the edge of a network to build small clouds, which perform as small data centers. In fog computing, lightweight virtualization (e.g., containers) has been widely used to achieve low overhead for performance-limited fog devices such as WiFi access points (APs) and set-top boxes. Unfortunately, containers have a weakness in the control of network bandwidth for outbound traffic, which poses a challenge to fog computing. Existing solutions for containers fail to achieve desirable network bandwidth control, which causes bandwidth-sensitive applications to suffer unacceptable network performance. In this paper, we propose qCon, which is a QoS-aware network resource management framework for containers to limit the rate of outbound traffic in fog computing. qCon aims to provide both proportional share scheduling and bandwidth shaping to satisfy various performance demands from containers while implementing a lightweight framework. For this purpose, qCon supports the following three scheduling policies that can be applied to containers simultaneously: proportional share scheduling, minimum bandwidth reservation, and maximum bandwidth limitation. For a lightweight implementation, qCon develops its own scheduling framework on the Linux bridge by interposing qCon’s scheduling interface on the frame processing function of the bridge. To show qCon’s effectiveness in a real fog computing environment, we implement qCon in a Docker container infrastructure on a performance-limited fog device—a Raspberry Pi 3 Model B board.

Modern vehicles are evolving with more electronic components than ever before (In this paper, “vehicle” means “automotive vehicle.” It is also equal to “car.”) One notable example is graphical processing unit (GPU), which is a key component to implement a digital cluster. To implement the digital cluster that displays all the meters (such as speed and fuel gauge) together with infotainment services (such as navigator and browser), the GPU needs to be virtualized; however, GPU virtualization for the digital cluster has not been addressed yet. This paper presents a Virtualized Automotive DIsplay (VADI) system to virtualize a GPU and its attached display device. VADI manages two execution domains: one for the automotive control software and the other for the in-vehicle infotainment (IVI) software. Through GPU virtualization, VADI provides GPU rendering to both execution domains, and it simultaneously displays their …

A method for adjusting a transmission rate in the transmis sion control protocol (TCP) under a mobile ad hoc network environment, including calculating a difference value (Diff) between the current transmission rate (cur bw) and the previous transmission rate (base bw), setting up upper and lower thresholds by calculating the standard variation (base bw Var) of the previous transmission rate, and updating the previous transmission rate to the current transmission rate depending upon a comparison result of Diff and the size of the upper and lower thresholds.

Current automotive software is evolving to integrate extension software with control software. Software integration increases the complexity of software and can cause system fault that threats safety of automobile. To solve this problem, extension software should be isolated from control software. However, this isolation alone is not sufficient to prevent extension software from faulting control software due to the lack of secure device access. In this paper, we propose a secure automotive software platform that has secure device access method with TrustZone. Our method restricts a direct access of the extension software and supports multicore processors. Measured overhead of our platform shows less than 1% degradation, and the maximum bandwidth of device access is achieved up to 5MB/s.

The prediction of the resource consumption for the distributed training of deep learning models is of paramount importance, as it can inform a priori users how long their training would take and also enable users to manage the cost of training. Yet, no such prediction is available for users because the resource consumption itself varies significantly according to "settings" such as GPU types and also by "workloads" like deep learning models. Previous studies have aimed to derive or model such a prediction, but they fall short of accommodating the various combinations of settings and workloads together. This study presents Driple that designs graph neural networks to predict the resource consumption of diverse workloads. Driple also designs transfer learning to extend the graph neural networks to adapt to differences in settings. The evaluation results show that Driple can effectively predict a wide range of workloads …

While a number of user-level protocols have been developed to reduce the gap between the performance capabilities of the physical network and the performance actually available, their compatibility issues with the existing sockets-based applications and IP-based infrastructure has been an area of major concern. To address these compatibility issues while maintaining a high performance, a number of researchers have been looking at alternative approaches to optimize the existing traditional protocol stacks. Broadly, previous research has broken up the overheads in the traditional protocol stack into four related aspects, namely: (i) compute requirements and contention, (ii) memory contention, (iii) I/O bus contention and (iv) system resources’ idle time. While previous research dealing with some of these aspects exists, to the best of our knowledge, there is no work which deals with all these issues in an integrated …

Current network virtualization allows tenants to have their own virtual networks. However, new demands to "program" virtual networks at a finer granularity have arisen as tenants want the ability to provision and control switches and links in their virtual networks. This study proposes a new concept called the p-NIaaS model. The p-NIaaS model enables tenants to program their own packet processing logic and monitor network status from any virtual network infrastructure, which is not possible with current network virtualization. This article presents the Libera network hypervisor that implements the p-NIaaS model. Libera overcomes the shortcomings of existing network hypervisors such as scalability, VM migration support, and flexibility. The evaluation shows that the Libera hypervisor is highly scalable and effectively supports VM migration. We also present the overheads of Libera. Libera incurs up to 11 percent …

H.264/SVC has been widely adopted to smart CE devices for network-adaptive video transmission. However, current H.264/SVC does not support sufficiently high subjective QoS (Quality of Service) because its scalability methods provide only coarse-grained network adaptivity. To address this problem, we propose a Scalable ROI (SROI) algorithm that applies the ROI (Region of Interests) concept to SVC (Scalable Video Codec). While existing methods use a fixed size of enhancement layers, our method utilizes only ROI area as an enhancement layer so that it can improve subjective QoS with low transmission overhead. Through some experiments, we show that our method can increase PSNR (Peak Signal-to-Noise Ratio) by 4~5 dB in ROI areas.

Mobile phones have evolved into complex systems as they have more and more new applications built-in. As a result, they are less reliable and less secure than before. Virtual Machine Monitors (VMM) or hypervisors have been introduced to help the reliability and security of mobile phones but the existing research does not completely address three issues critical to mobile phones: real-time support, resource limitation, and power efficiency. In this paper we propose building a new VMM called MobiVMM for mobile phones to deal with these issues. MobiVMM enables real-time support using priority based scheduling and a pseudo-polling mechanism. Resource and power efficiency is achieved through light-weight design and implementation, highly customized guest operating systems, and a virtual hardware abstraction layer. We present our design considerations and report some preliminary experimental results …

Although extremely high-speed interconnects are available today, the traditional protocol stacks such as TCP/IP and UDP/IP are not able to utilize the maximum network bandwidth due to inherent overheads in the protocol stacks. Such overheads are a big obstacle for high-performance computing applications to exploit high-speed interconnects in cluster environments. To address this issue, many researchers have been presenting analyses of protocol overheads and suggesting a number of optimization approaches to harness the TCP/IP suite over high-speed interconnects. However, to the best of our knowledge, there is no study that analyzes and optimizes the protocol overheads thoroughly in an integrated manner. In this paper, we exploit a set of protocol optimization mechanisms in an integrated manner while dealing with the full spectrum of the protocol layers from the transport layer to the physical layer. To …

This paper proposes V-Sight, a network monitoring framework for software-defined networking (SDN)-based virtual networks. Network virtualization with SDN (SDN-NV) makes it possible to realize programmable virtual networks; so, the technology can be beneficial to cloud services for tenants. However, to the best of our knowledge, although network monitoring is a vital prerequisite for managing and optimizing virtual networks, it has not been investigated in the context of SDN-NV. Thus, virtual networks suffer from non-isolated statistics between virtual networks, high monitoring delays, and excessive control channel consumption for gathering statistics, which critically hinders the benefits of SDN-NV. To solve these problems, V-Sight presents three key mechanisms: 1) statistics virtualization for isolated statistics, 2) transmission disaggregation for reduced transmission delay, and 3) pCollector aggregation for …

In SDN, as the control channel becomes a performance bottleneck, modeling the control channel traffic is important. Such a model is useful in predicting the control channel traffic for network provisioning. However, previously proposed models are quite limited in that they assume only the forwarding function of a specific controller for their models. To overcome the limitations, first, this paper analyzes the control traffic by seven functions (including forwarding function) of a controller. Then, we build a seven-function model to predict control channel usage and evaluate the prediction accuracy that achieves as high as 94%. Note that previous models did not have any quantitative evaluation. Our model is built with the Open Network Operating System (ONOS) controller and extended to Floodlight and POX controllers. We show that the extended model also achieves similar prediction accuracy (95%). Furthermore, we …

Operating Systems Lab., Korea University (고려대 운영체제연구실)

교수님 소개

유혁

연구실 소개

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Making Sense of Job Preemption for Distributed Deep Learning Acceleration

Prediction-based GPU Sharing for Distributed Training

DEVICE AND METHOD FOR REAL-TIME AND LIGHTWEIGHT GPU WORKLOAD TRACING AND BOTTLENECK ANALYSIS

연구실 보유 특허

대학별 입시끈

우리 연구실 모집

교수님 소개

유혁

연구실 키워드

알럼나이가 간 회사

연구실 지원 방법

대우 조건

연도별 논문 인용수

저널 & 학회 논문 출판 수

연구실 논문 목록

Making Sense of Job Preemption for Distributed Deep Learning Acceleration

Prediction-based GPU Sharing for Distributed Training

DEVICE AND METHOD FOR REAL-TIME AND LIGHTWEIGHT GPU WORKLOAD TRACING AND BOTTLENECK ANALYSIS

연구실 보유 특허

최근 3년간 연구실적 높은 연구실

대학별 입시끈

우리 연구실 모집