• The KIPS Transactions:PartA
• /
• v.14A no.4
• /
• pp.235-244
• /
• 2007

Caching and prefetching policies have been used in most of computer systems to compensate speed differences between primary memory and secondary storage devices. In this paper, we design and implement a Flash Cache Core Module(FCCM) on the YAFFS which operates on a file system layer for NAND flash memory. The FCCM is independent of the underlying kernel in order to support its stability and compatibility. Also, we implement the Dirty-Last memory replacement technique considering the characteristics of flash memory, and the waiting queue for pages to be prefetched according to page hit. The FCCM reduced the number of I/Os and the amount of prefetched pages by maximum 55%(20% on average) and maximum 55%(24% on average), respectively, comparing with caching and prefetching policies of Linux.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2016.01a
• /
• pp.5-7
• /
• 2016

본 논문에서는 블록 생성 번호를 활용한 동적 마모도 평준화 기법을 제안한다. 지금까지 제안된 동적 마모도 평준화 기법들은 콜드 블록을 판별하기 위해 경과 시간을 사용하고 있다. 하지만 저장장치의 데이터 접근은 일정한 시간 간격으로 이루어지는 것이 아니기 때문에 이와 같은 경과 시간을 사용하는 방식은 데이터에 대한 블록 접근 정보가 왜곡될 수 있는 단점이 있다. 이러한 단점을 해결하기 위해, 본 논문에서 제안하는 기법은 블록을 할당할 때 블록 순차 번호를 테이블에 저장하고 이를 이용하여 블록의 접근 빈도를 판별한다. 실험에서 제한하는 기법은 기존의 CB, CAT 기법과 비교하여 최대 11% 수명이 향상됨을 확인하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2004.05a
• /
• pp.1667-1670
• /
• 2004

x86기반 임베디드 리눅스 시스템의 저장장치로 단일-칩 플래시 디스크인 DOC(DiskOnChip) 시스템이 많이 사용되고 있다. 본 연구에서는 DOC 스스로 부팅하도록 부팅이미지, 커널이미지, 루트파일시스템을 설치하는 과정을 설명한다. DOC는 자체 기능으로 에러탐색 및 수정기능과 파일시스템으로 TrueFFS가 인터페이스로 동작한다. 또한, 구성된 DOC 저장 시스템에 GUI 구현할 수 있도록 Qt-E 계층을 추가하여 시스템 개발자는 단지 어플리케이션을 설치함으로써 쉽게 임베디드시스템을 구성할 수 있다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.11a
• /
• pp.69-72
• /
• 2020

Log Structured Merged Tree(LSM-Tree)구조를 사용하여 빠른 데이터 쓰기 성능을 보유한 RocksDB에는 쓰기 증폭과 공간 증폭 현상이 발생한다. 쓰기 증폭은 과도한 쓰기 연산을 유발하여 데이터 처리 성능 저하와 플래시 메모리 기반 장치의 수명 저하를 초래하며, 공간 증폭은 데이터 저장 공간 점유로 인한 저장 공간 부족 문제를 야기한다. 본 논문에서는 쓰기 증폭과 공간 증폭 완화를 위해 RocksDB 의 성능에 영향 주는 주요 파라미터를 추출하고, 기계학습 기법인 랜덤 포레스트를 사용하여 추출한 파라미터가 쓰기 증폭과 공간 증폭에 미치는 영향을 분석하였다. 실험결과 쓰기 증폭과 공간 증폭에 영향을 많이 주는 주요 요소를 선별하였고 다른 파라미터에 대비해서 성능 격차가 61.7% 더 나타낸 것을 발견하였다.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.10
• /
• pp.1-14
• /
• 2020

In this paper, we propose to solve the problem of increasing system memory usage due to an increase in the number of mapping information management when using a NAND flash memory-based storage device in an existing sector-based file system. The proposed technique is to store only mapping information in page units based on index blocks and manage them in block units. To this end, the proposed technique uses a sequential offset for storing and managing a plurality of mapping information in one page in a small block, and a reverse offset for a spare page corresponding to a change in mapping information in the block. Through this, the proposed technique has the advantage that the number of block-unit deletions is less than that of the existing technique, and the system memory usage required for mapping information management is low. Reduced by about 32%.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2017.10a
• /
• pp.322-325
• /
• 2017

Due to the rapid increase in data, various studies are being conducted to efficiently manage the data. In 2025, the total amount of data will increase to more than 163 ZB, and more than a quarter of the data will be a real-time data. As mass storage devices is changed from HDD to SSD, SSD needs own way to manage their data effectively. In this paper, we study the SSD system structure and deduplication management methods of data management related to Flash-SSD. We also propose an application level user file system using deduplication. It is anticipated that it saves storage capacity and minimize reducing performance by unnecessary traffic.

• Journal of the Korea Society of Computer and Information
• /
• v.26 no.11
• /
• pp.1-9
• /
• 2021

3D-NAND flash memory provides high capacity per unit area by stacking 2D-NAND cells having a planar structure. However, due to the nature of the lamination process, there is a problem that the frequency of error occurrence may vary depending on each layer or physical cell location. This phenomenon becomes more pronounced as the number of write/erase(P/E) operations of the flash memory increases. Most flash-based storage devices such as SSDs use ECC for error correction. Since this method provides a fixed strength of data protection for all flash memory pages, it has limitations in 3D NAND flash memory, where the error rate varies depending on the physical location. Therefore, in this paper, pages and layers with different error rates are classified into clusters through the K-means machine learning algorithm, and differentiated data protection strength is applied to each cluster. We classify pages and layers based on the number of errors measured after endurance test, where the error rate varies significantly for each page and layer, and add parity data to stripes for areas vulnerable to errors to provides differentiate data protection strength. We show the possibility that this differentiated data protection policy can contribute to the improvement of reliability and lifespan of 3D NAND flash memory compared to the protection techniques using RAID-like or ECC alone.

• Journal of Digital Contents Society
• /
• v.12 no.2
• /
• pp.141-150
• /
• 2011

Recently the Solid State Drive (SSD) is widely used for storage system of various mobile devices. In this case, existing buffer replacement algorithms based on the hard disk do not consider characteristics of flash memory, so it caused performance degradation of the system. This paper proposes a novel buffer replacement policy called ABM (Asymmetric Buffer Management) policy. ABM policy separates read and write buffer space and applies different replacement unit and replacement algorithm for each buffer. In addition, write buffer delay scheme and dynamic size adaptation algorithm is applied for better performance. ABM outperforms other replacement policies, especially ABM-LRU-CLC shows 32% better performance than normal LRU policy.

• Journal of Digital Contents Society
• /
• v.12 no.2
• /
• pp.241-252
• /
• 2011

During the last decades, other parts of operating systems, storage devices, and media are changed steadily, whereas filesystem is changed little. As data is grown bigger, the number of files to be managed also increases in geometrically. Researches about new filesystem schemes are being done widely to support these files efficiently. In web document search area, there are many researches about finding meaningful documents using semantic search. Many researches tried to apply these schemes, which is been proven in web document search previously, to filesystems. But they've focused only on higher layer of filesystem, that is not related seriously to storage media. Therefore they're not well tuned to physical characteristics of new flash memory based SSD which has different features against traditional HDD. We enhance log structured filesystem, that is already well known to work better in SSD, by putting semantic search scheme to and with multi logging point.

• Journal of KIISE
• /
• v.44 no.8
• /
• pp.767-773
• /
• 2017

In a distributed system environment, such as a cloud database, the tail latency needs to be kept short to ensure uniform quality of service. In this paper, through experiments on a Cassandra database, we show that long tail latency is caused by a lack of memory space because the database cannot receive any request until free space is reclaimed by writing the buffered data to the storage device. We observed that, since the performance of the storage device determines the amount of time required for writing the buffered data, the performance degradation of Solid State Drive (SSD) due to garbage collection results in a longer tail latency. We propose a garbage collection synchronization technique, called SyncGC, that simultaneously performs garbage collection in the java virtual machine and in the garbage collection in SSD concurrently, thus hiding garbage collection overheads in the SSD. Our evaluations on real SSDs show that SyncGC reduces the tail latency of $99.9^{th}$ and, $99.9^{th}-percentile$ by 31% and 36%, respectively.