• Annual Conference of KIPS
• /
• 2020.11a
• /
• pp.84-87
• /
• 2020

최근, 응용 프로그램의 이식성, 확장성, 낮은 오버헤드 및 관리의 용이성 등을 제공하는 컨테이너 기술을 고성능 컴퓨팅 (high performance computing, HPC) 환경에 접목하려는 다양한 연구들이 진행되고 있다. 본 논문에서는 멀티 프로그램화된 환경, 즉, 컨테이너 기반의 다수개의 HPC 워크로드들이 동시에 실행되는 환경에서 멀티 프로그래밍 수준, 통신 패턴 및 비율에 따른 HPC 워크로드들의 성능 특성을 분석하고, HPC 워크로드가 실행되는 동일한 컨테이너 그룹에 속한 컨테이너들의 스케쥴링 시간 부조화가 데이터 교환 지연 시간을 증가시키고 그 결과 응용 성능을 크게 저하시킬 수 있음을 확인한다. 또한 HPC 워크로드가 수행되는 동일 그룹 컨테이너들의 CPU 점유 가능값(CPU Shares)을 동적으로 조절하는 휴리스틱을 제안, 적용함으로써, HPC 워크로드의 성능(통신소비시간 최대 약 42.5%, 워크로드 실행시간 최대 약 23.6% 감소)을 크게 향상시킬 수 있음을 확인한다.

• 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.

• The KIPS Transactions:PartD
• /
• v.11D no.4
• /
• pp.747-754
• /
• 2004

Database administrators should be aware of performance characteristics of database systems in order to manage database system effectively. The usages of system resources in database systems could be quite different under database workloads. The objective of this paper is to identify and analyze performance characteristics of database systems in different workloads, which could help database tuners tune database systems Under the TPC-C and TPC-W workloads, which represent typical workloads of online transaction processing and electronic commerce respectively, we investigated usage types of resource that are determined by fourteen performance indicator, and are behaved in response to changes of four tuning parameters (data buffer, private memory, I/O process, shared memory). Eight out of the fourteen performance indicators cleary show the performance differences under the workloads. Changes of data buffer parameter give a influences to database system. The tuning parameter that affects the system performance significantly is the database buffer size in the both workloads.

• Annual Conference of KIPS
• /
• 2017.04a
• /
• pp.50-52
• /
• 2017

클라우드 컴퓨팅 시장의 지속적 성장과 가상화의 인기로 세계적 VDI 시장은 꾸준한 성장률을 보이고 있다. 또한 의료, 교육, 금융 등의 폭넓은 분야에서 VDI 서비스가 활용될 전망이다. 하지만 기존 VDI 서비스는 고정적인 자원 할당으로 사용자 워크로드 맞춤형 자원이 제공되지 못하는 문제점이 있다. 따라서 본 논문에서는 기존 VDI에 비해 실행속도가 빠른 컨테이너의 장점을 살려 VDI를 컨테이너화 하고, 사용자 워크로드 맞춤형으로 자원을 분배하기 위해 VDI 컨테이너 자원 사용량 데이터로 K-means 알고리즘을 통한 군집 분석 기반의 워크로드 분류 방법을 제시하였다.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.29 no.6
• /
• pp.1305-1317
• /
• 2019

Although encryption for data protection is essential in the big data platform environment of public institutions and corporations, much performance verification studies on encryption algorithms considering actual big data workloads have not been conducted. In this paper, we analyzed the performance change of AES, ARIA, and 3DES for each of six workloads of big data by adding data and nodes in MongoDB environment. This enables us to identify the optimal block-based cryptographic algorithm for each workload in the big data platform environment, and test the performance of MongoDB by testing various workloads in data and node configurations using the NoSQL Database Benchmark (YCSB). We propose an optimized architecture that takes into account.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.19 no.1
• /
• pp.111-116
• /
• 2019

Recently, GPU expands application domains from graphic processing to various kinds of parallel workloads. However, current GPU systems focus on the maximization of each workload's parallelism through simplified control rather than considering various workload characteristics. This paper classifies the resource usage characteristics of GPU workloads into computing-bound, memory-bound, and dependency-latency-bound, and quantifies the fine-grained bottleneck for efficient workload allocation. For example, we identify the exact bottleneck resources such as single function unit, double function unit, or special function unit even for the same computing-bound workloads. Our analysis implies that workloads can be allocated together if fine-grained bottleneck resources are different even for the same computing-bound workloads, which can eventually contribute to efficient workload allocation in GPU.

• Journal of Digital Contents Society
• /
• v.19 no.2
• /
• pp.335-341
• /
• 2018

There exists a limit in efficient processing of mixed workloads that database markets requires in recent years since existing database systems utilize a static page storage model. In this paper we propose a dynamic page storage model that can reflect the characteristics of mixed workloads. We also describe how to extract optimized column groups from given mixed workloads and how to construct pages dynamically. Finally, we show in our experiments that the proposed model is more efficient than the existing model in processing given mixed workloads.

• KIPS Transactions on Computer and Communication Systems
• /
• v.2 no.11
• /
• pp.461-470
• /
• 2013

This paper proposes high speed kernel data collection method for analysis of memory workload, using technique of direct access to process's memory management structure. The conventional analysis tools have a slower data collection speed and they are lack of scalability due to collection only formalized memory information. The proposed method collects kernel data much faster than the conventional methods using technique of direct collect to process's memory information, page table, page structure in the memory management structure, and it can collect data which user wanted. We collect memory management data of the running process, and analyze its memory workload.

• KIPS Transactions on Computer and Communication Systems
• /
• v.8 no.10
• /
• pp.245-250
• /
• 2019

The use of NAND flash memory has increased rapidly due to the development of mobile fields. However, NAND flash memory has a limited lifespan, so studies are underway to predict its lifespan. Workload is one of the factors that significantly affect the life of NAND flash memory, and workload analysis studies in mobile environments are insufficient. In this paper, we analyze the distribution of workload in the mobile environment by collecting traces generated by using Android-based smartphones. The collected traces can be divided into three groups of hotness. Also they are distributed in the form of heavy tails. We fit this to the Pareto, Lognormal, and Weibull distributions, and Traces are closest to the Pareto distribution.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.10c
• /
• pp.303-306
• /
• 2006

데이터베이스 시스템을 효율적으로 운영하기 위하여 데이터베이스 관리자는 시스템의 자원 사용과 응용 프로그램에 의한 워크 로드의 특징을 알아야 한다. 워크 로드 분석을 위해 테이블, 리소스, 튜닝 방법론 등 여러 연구가 진행되어 왔으나 워크 로드를 형성하는 역할이 특정 테이블에만 집중되어 있는 현상에 대해서는 연구된 적이 없었다. 본 논문에서는 운영 시스템의 테이블 간의 연관 관계를 도출해 보고 연관 관계를 가지는 테이블 들이 워크 로드에 참여하는 유형과 횟수를 분석하는 워크 로드 분석 도구를 제안하고 대형 CRM 분석 시스템에 적용하여 데이터베이스 시스템의 워크로드를 분석해 본다.