• IEMEK Journal of Embedded Systems and Applications
• /
• v.14 no.2
• /
• pp.63-69
• /
• 2019

Recently, ultra-low latency flash storage devices such as Z-SSD and Optane SSD were introduced with the significant technological improvement in the storage devices which provide much faster response time than today's other NVMe SSDs. With such ultra-low latency, $10{\mu}s$, storage devices the cost of context switch could be an overhead during interrupt-driven I/O completion process. As an interrupt-driven I/O completion process could bring an interrupt handling overhead, polling or hybrid-polling for the I/O completion is known to perform better. In this paper, we analyze tail latency problem in a polling process caused by process scheduling in data center environment where multiple applications run simultaneously under one system and we introduce our adaptive polling selection technique which dynamically selects efficient processing method between two techniques according to the system's conditions.

• Journal of information and communication convergence engineering
• /
• v.21 no.1
• /
• pp.9-16
• /
• 2023

The natural language on social network platforms has a certain front-to-back dependency in structure, and the direct conversion of Chinese text into a vector makes the dimensionality very high, thereby resulting in the low accuracy of existing text classification methods. To this end, this study establishes a deep learning model that combines a big data ultra-deep convolutional neural network (UDCNN) and long short-term memory network (LSTM). The deep structure of UDCNN is used to extract the features of text vector classification. The LSTM stores historical information to extract the context dependency of long texts, and word embedding is introduced to convert the text into low-dimensional vectors. Experiments are conducted on the social network platforms Sogou corpus and the University HowNet Chinese corpus. The research results show that compared with CNN + rand, LSTM, and other models, the neural network deep learning hybrid model can effectively improve the accuracy of text classification.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 2003.06a
• /
• pp.660-663
• /
• 2003

This paper proposes a systematic approach for an accurate control of the Terfenol-D actuator taking into account hysteresis, modeled by applying the classical Preisach operator with memory curve. A desired input displacement is calculated by using the hysteresis inverter, which is fed into the actuator. Then the PI compensator corrects the error between the commanded and actual displacements. Experiments with the step responses show that the PI controller settles in 70 ms and the hybrid controller in 20 ms. It means that the concurrent application of two control schemes is effective to control the actuator.

• Journal of KIISE:Databases
• /
• v.37 no.6
• /
• pp.308-314
• /
• 2010

As the price of flash memory continues to drop and the technology of flash SSD controller innovates, high performance flash SSDs with affordable prices flourish in the storage market. Nevertheless, it is hard to expect that flash SSDs will replace harddisks completely as database storage. Instead, the approach to use flash SSD as a cache for harddisks would be more practical, and, in fact, several hybrid storage architectures for flash memory and harddisk have been suggested in the literature. In this paper, we propose a new approach to use flash SSD as an extended buffer for main buffer in database systems, which stores the pages replaced out from main buffer and returns the pages which are re-referenced in the upper buffer layer, improving the system performance drastically. In contrast to the existing approaches to use flash SSD as a cache in the lower storage layer, our approach, which uses flash SSD as an extended buffer in the upper host, can provide fast random read speed for the warm pages which are being replaced out from the limited main buffer. In fact, for all the pages which are missing from the main buffer in a real TPC-C trace, the hit ratio in the extended buffer could be more than 60%, and this supports our conjecture that our simple extended buffer approach could be very effective as a cache. In terms of performance/price, our extended buffer architecture outperforms two other alternative approaches with the same cost, 1) large main buffer and 2) more harddisks.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.46 no.4
• /
• pp.21-30
• /
• 2009

Fusion flash memory such as OneNAND$^{TM}$ is popular as a ubiquitous storage device for embedded systems because it has advantages of NAND and NOR flash memory that it can support large capacity, fast read/write performance and XIP(eXecute-In-Place). Besides, OneNAND$^{TM}$ provides not only advantages of hybrid structure but also multi-block erase function that improves slow erase performance by erasing the multiple blocks simultaneously. But traditional NAND Flash Translation Layer may not fully support it because the garbage collection of traditional FTL only considers a few block as victim block and erases them. In this paper, we propose an Erase Group Flash Translation Layer for improving multi-block erase function. EGFTL uses a superblock scheme for enhancing garbage collection performance and invalid block management to erase multiple blocks simultaneously. Also, it uses clustered hash table to improve the address translation performance of the superblock scheme. The experimental results show that the garbage collection performance of EGFTL is 30% higher than those of traditional FTLs, and the address translation performance of EGFTL is 5% higher than that of Superblock scheme.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.11 no.12
• /
• pp.4997-5006
• /
• 2010

A flash-memory-based SSDs(Solid State Disks) are one of the best media to support portable and desktop computers' storage devices. Their features include non-volatility, low power consumption, and fast access time for read operations, which are sufficient to present flash memories as major database storage components for desktop and server computers. However, we need to improve traditional storage management schemes based on HDD(Hard Disk Drive) and RAID(Redundant array of independent disks) due to the relatively slow or freezing characteristics of write operations of SSDs, as compared to fast read operations. In order to achieve this goal, we propose a new storage management scheme called Hetero-Mirroring based on traditional HDD mirroring scheme. Hetero-Mirroring-based scheme improves RAID-1 operation performance by balancing write-workloads and delaying write operations to avoid SSD freezing. Our test results show that our scheme significantly reduces the write operation overheads and freezing overheads, and improves the performance of traditional SSD-RAID-1 scheme by 18 percent, and the response time of the scheme by 38 percent.

• The Journal of Korean Institute of Next Generation Computing
• /
• v.14 no.5
• /
• pp.40-52
• /
• 2018

Solid state drive (SSD) cache technologies that are used as a second-tier cache between the main memory and hard disk drive (HDD) have been widely studied. The SSD cache requires a new prefetching scheme as well as cache replacement algorithms. This paper presents a prefetching scheme for a storage-class cache using SSD. This prefetching scheme is designed for the storage-class cache and based on a long-term scheduling in contrast to the short-term prefetching in the main memory. Traditional prefetching algorithms just consider only read, but the presented prefetching scheme considers both read and write. An experimental evaluation shows 2.3% to 17.8% of hit rate with a 64GB of SSD and the 4GiB of prefetching size using an I/O trace of 14 days. The proposed prefetching scheme showed significant improvement of cache hit rate and can be easily implemented in storage-class cache systems.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.16 no.4
• /
• pp.203-209
• /
• 2023

Harmony search(HS) does not use the evaluation of individual harmony when referring to HM when constructing a new harmony, but particle swarm optimization(PSO), on the contrary, uses the evaluation value of individual particles and the evaluation value of the population to find a solution. However, in this study, we tried to improve the performance of the algorithm by finding and identifying similarities between HS and PSO and applying the particle improvement process of PSO to HS. To apply the PSO algorithm, the local best of individual particles and the global best of the swam are required. In this study, the process of HS improving the worst harmony in harmony memory(HM) was viewed as a process very similar to that of PSO. Therefore, the worst harmony of HM was regarded as the local best of a particle, and the best harmony was regarded as the global best of swam. In this way, the performance of the HS was improved by introducing the particle improvement process of the PSO into the HS harmony improvement process. The results of this study were confirmed by comparing examples of optimization values for various functions. As a result, it was found that the suggested HS-PSO was much better than the existing HS in terms of accuracy and consistency.

• Journal of Korean Society of Coastal and Ocean Engineers
• /
• v.4 no.1
• /
• pp.34-44
• /
• 1992

A numerical model using Hybrid Element Method(HEM) is presented for the prediction of wave agitations in a harbor which are induced by the intrusion and transformation of incident short-period waves. A linear mild-slope equation including bottom friction is used as the governing equation and a partial absorbing boundary condition is used on solid boundaries. Functional derived in the present paper is based on the Chen and Mei(1974)'s concept which uses finite element net in the inner region and analytical solution of Helmholtz equation in the outer region. Final simultaneous equations are solved using the Gaussian Elimination Method. The model appears to be reasonably good from the comparison of numerical calculation with hydraulic experimental results of short-wave diffraction through a breakwater gap(Pos and Kilner, 1987). The problem of requring large computational memory could be overcome using 8-noded isoparametric elements.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.795-802
• /
• 2003

To live autonomously, intelligent agents such as robots or virtual characters need ability that recognizes given environment, and learns and chooses adaptive actions. So, we propose an action selection/learning mechanism in intelligent agents. The proposed mechanism employs a hybrid system which integrates a behavior-based method using the reinforcement learning and a cognitive-based method using the symbolic learning. The characteristics of our mechanism are as follows. First, because it learns adaptive actions about environment using reinforcement learning, our agents have flexibility about environmental changes. Second, because it learns environmental factors for the agent's goals using inductive machine learning and association rules, the agent learns and selects appropriate actions faster in given surrounding and more efficiently in extended surroundings. Third, in implementing the intelligent agents, we considers only the recognized states which are found by a state detector rather than by all states. Because this method consider only necessary states, we can reduce the space of memory. And because it represents and processes new states dynamically, we can cope with the change of environment spontaneously.