Gluster 파일시스템의 파일 크기에 따른 I/O 처리 분석

성한울(Hanul Sung),하수진(Sujin Ha),구동훈(Donghun Koo),방지우(Jiwoo Bang),염헌영(HeonYoung Yeom),엄현상(Hyeonsang Eom) 한국정보과학회 2017 한국정보과학회 학술발표논문집 Vol.2017 No.12

Ring bus 경쟁상황을 피하기 위한 CPU-GPU 간의 스케줄러

성한울(Hanul Sung),엄현상(Hyeonsang Eom),염헌영(HeonYoung Yeom) 한국정보과학회 2014 한국정보과학회 학술발표논문집 Vol.2014 No.6

내장형 GPU 환경에서 GPU 프로세스의 성능 보장을 위한 캐시 분할 방식

성한울(Hanul Sung),엄현상(Hyeonsang Eom),염헌영(HeonYoung Yeom) 한국정보과학회 2013 한국정보과학회 학술발표논문집 Vol.2013 No.11

내장형 GPU 환경에서 CPU-GPU 간의 공유 캐시에서의 캐시 분할 방식의 필요성

성한울(Hanul Sung),엄현상(Hyeonsang Eom),염헌영(HeonYoung Yeom) 한국정보과학회 2014 정보과학회 컴퓨팅의 실제 논문지 Vol.20 No.9

최근 전력의 한계 때문에 많은 트랜지스터를 모두 이용할 수 없는 ‘다크실리콘’ 문제가 발생했다. 이 문제를 효율적으로 해결하기 위하여 CPU(Central processing unit)와 GPU(Graphic processing unit)를 함께 사용하여 분산처리하기 시작했다. 최근에는 CPU(Central processing unit)와 GPU(Graphic processing unit)가 메모리와 Last Level Cache를 공유하는 내장형 GPU 프로세서(Integrated graphic processing unit processor)가 등장했다. 하지만 CPU 프로세스와 GPU 프로세스가 LLC(Last level cache)로 접근하기 위한 어떠한 규칙이 없기 때문에, 동시에 CPU 프로세스와 GPU 프로세스 수행될 때 LLC(Last level cache)를 차지하기 위한 경쟁이 일어나 성능 저하가 발생한다. 본 논문에서는 캐시 접근 빈도가 큰 여러 개의 프로세스들이 수행됨에 따라 캐시 오염이 발생한 상황에서 GPU 프로세스의 성능 보장을 위하여 GPU 프로세스만을 위한 고정된 Last Level Cache 공간을 주는 캐시 분할방식이 필요함을 증명하고 캐시를 분할하기 위한 페이지 컬러링 기법을 소개하고 디자인한다. Recently, Distributed computing processing begins using both CPU(Central processing unit) and GPU(Graphic processing unit) to improve the performance to overcome darksilicon problem which cannot use all of the transistors because of the electric power limitation . There is an integrated graphics processor that CPU and GPU share memory and Last level cache(LLC). But, There is no LLC access rules between CPU and GPU, so if GPU and CPU processes run together at the same time, performance of both processes gets worse because of the contention on the LLC. This Paper gives evidence to prove the need of the Cache Partitioning and is mentioned about the cache partitioning design using page coloring to allocate the L3 Cache space only for the GPU process to guarantee GPU process performance.

비트코인 가격 예측을 위한 LSTM 모델의 Hyper-parameter 최적화 연구

김준호,성한울 한국융합학회 2022 한국융합학회논문지 Vol.13 No.4

비트코인은 정부나 금융기관에 의존되어 있지 않은 전자 거래를 지향하며 만들어진 peer-to-peer 방식의 암호화폐이다. 비트코인은 최초 발행 이후 거대한 블록체인 금융 시장을 생성했고, 이에 따라 기계 학습을 이용한 비트코인 가격 데이터를 예측하는 연구들이 활발해졌다. 그러나 기계 학습 연구의 비효율적인 Hyper-parameter 최적화 과정이 연구 진행에 있어 비용적인 측면을 악화시키고 있다. 본 논문은 LSTM(Long Short-Term Memory) 층을 사용하는 비트코인 가격 예측 모델에서 가장 대표적인 Hyper-parameter 중 Timesteps, LSTM 유닛의 수, 그리고 Dropout 비율의 전체 조합을 구성하고 각각의 조합에 대한 예측 성능을 측정하는 실험을 통해 정확한 비트코인 가격 예측을 위한 Hyper-parameter 최적화의 방향성을 분석하고 제시한다. Bitcoin is a peer-to-peer cryptocurrency designed for electronic transactions that do not depend on the government or financial institutions. Since Bitcoin was first issued, a huge blockchain financial market has been created, and as a result, research to predict Bitcoin price data using machine learning has been increasing. However, the inefficient Hyper-parameter optimization process of machine learning research is interrupting the progress of the research. In this paper, we analyzes and presents the direction of Hyper-parameter optimization through experiments that compose the entire combination of the Timesteps, the number of LSTM units, and the Dropout ratio among the most representative Hyper-parameter and measure the predictive performance for each combination based on Bitcoin price prediction model using LSTM layer.

A Study on Parallel Programming Approach for MMO Game Server

이정훈,성한울 (사)한국컴퓨터게임학회 2023 한국컴퓨터게임학회논문지 Vol.36 No.3

Massively Multiplayer Online Role-Playing Game(MMORPG) attract thousand of concurrent users, leading to a surge in server requests. The massive requests results in contention on the game server, causing unexpected latency issues that undermine user experience. Delayed latency leads to user dissatisfaction and, ultimately, user churn. In this paper, we propose a multi-threaded game server design that provides stable response time by utilizing computing resources. And we study the performance of the design under various environments by configuring the approaches we implemented. The server consists of a socket system that helps communicate between clients and servers, a task system that handles changes in the status of user, and a synchronization system that allows users to receive the same game information. To prevent serious contention, both lock and lock-free algorithms for process synchronization are applied, and some parallel programming approaches such as visual processing are introduced.

Adaptive Game Environment Design based on Player Behavior Pattern for Difficulty Adjustment

이수환,성한울 (사)한국컴퓨터게임학회 2023 한국컴퓨터게임학회논문지 Vol.36 No.3

User churn in games often arises due to inadequate game difficulty. To address this, non-player characters (NPCs) has been utilized to modulate difficulty according to individual game skill. Nevertheless, the effectiveness of solely NPC-based adjustments is limited since game difficulty is influenced by both NPCs and environmental factors. This paper introduces a novel method for dynamically tailoring game difficulty by adjusting in-game environments based on player behavior patterns in top-down shooter game. Through analysis of diverse user game play data, we find that factors within the game environment, such as the distribution of enemy characters and the arrangement of terrain, have a substantial influence on the level of difficulty. Furthermore, it has been observed that behavioral patterns of players show variations according to changes in the game environment. Using these analytical result, we devise an artificial neural network model that configures an environment that suit player behavior patterns. With the model, we figure out the user player pattern and control the difficulty dynamically by changing the environment factors. Through the experiments, we show that our method provides an appropriate level of difficulty for users to prevent user churn.

강화학습을 이용한 대전 슈팅 게임의 플랫폼 형태에 따른 레벨 디자인 영향 분석

김준호,성한울 (사)한국컴퓨터게임학회 2023 한국컴퓨터게임학회논문지 Vol.36 No.4

As multi-agent reinforcement learning advance, it has been utilized for level design in game with multiple players and complex interactive environments. Despite the significant impact that the platform's design has on game level design, player-related metrics including behavior, skill level, and skill composition have been the primary focus of most reinforcement learning studies. In this paper, we study the impact of platform design on playing experience by utilizing multi-agent reinforcement learning. we use the Unity Engine to create various platform designs based on a 2vs2 shooting game and visual sensor visibility and structural complexity are taken into consideration for analysis. With Unity ML-Agents Toolkit, we construct a reinforcement learning environment based on the MA-POCA algorithm and the self-play method. Though the agent, we analyze changes in play experience with win rate, the number of episode, draw rate, and the Elo rating value. With the study, we figure out that visibility and complexity of platform design exhibit minimal impact on win rate balance in self-play method, but does influence the total number of episodes, draw rate, and the increase in Elo rating value.

향상된 유닛 생산과 게임 플레이 경험을 위한 스타크래프트 II 강화학습 AI

구교승,성한울 (사)한국컴퓨터게임학회 2023 한국컴퓨터게임학회논문지 Vol.36 No.4

The built-in AI in StarCraft II follows a predefined pattern of behavior patterns. This makes it easy for users to understand the AI's strategy without spending a lot of time, posing a challenge in maintaining user interest. To solve this problem, reinforcement learning AI has been developed for StarCraft II. But since theses studies focus on the win rate to train the agent, it provides monotonous strategies and misses the fun of the game. In this paper, we propose a player-like AI to improve the fun of the game by producing various units and adaptive strategies. To achieve this, we use reinforcement learning to train the agent the official counter list of StarCraft II and reward it based on the information of the opponent's scouting to make the agent change its strategy flexibly. The training is based on the "Zerg" race on the Simple64 map, and the opponent is the "Protoss" race. Based on the trained data, we evaluated the diversity of unit production, strategic unit production, etc. In addition, we had users play with the AI of existing work and the proposed AI, respectively. We then evaluated the fun and difficulty of the AI and its similarity to the player using a questionnaire. As a result, users rated the proposed AI higher than the fixed-strategy AI fin terms of fun, difficulty, and similarity.

Immunogenicity of MenACWY-CRM in Korean Military Recruits: Influence of Tetanus-Diphtheria Toxoid Vaccination on the Vaccine Response to MenACWY-CRM

김경효,김한울,박인호,유수성,유희태,오인수,성필수,신의철 연세대학교의과대학 2016 Yonsei medical journal Vol.57 No.6

The quadrivalent meningococcal conjugate vaccine (MenACWY-CRM) has been introduced for military recruits in Korea since 2012. This study was performed to evaluate the immunogenicity of MenACWY-CRM in Korean military recruits. In addition, the influence of tetanus-diphtheria toxoids (Td) vaccination on the vaccine response to MenACWY-CRM was analyzed. A total of 75 military recruits were enrolled. Among them, 18 received a dose of MenACWY-CRM only (group 1), and 57 received Td three days before MenACWY-CRM immunization (group 2). The immunogenicity of MenACWY-CRM was compared between the two groups. The serum bactericidal activity with baby rabbit complement was measured before and three weeks after immunization against serogroups A, C, W-135, and Y. The geometric mean titers (GMTs) against four serogroups were significantly increased in both groups after immunization. Compared to group 2, group 1 exhibited significantly higher vaccine responses in several aspects:post-immune GMTs against serogroup A and C, seroresponse rates against serogroup A, and a fold increases of titers against serogroup A, C, and Y. MenACWY-CRM was immunogenic against all vaccine-serogroups in Korean military recruits. Vaccine response to MenACWY-CRM was influenced by Td administered three days earlier.