• Journal of Digital Contents Society
• /
• v.19 no.6
• /
• pp.1161-1167
• /
• 2018

The development of machine learning has enabled machines to perform delicate tasks that only humans could do, and thus many companies have introduced machine learning based translators. Existing translators have good performances but they have problems in number translation. The translators often mistranslate numbers when the input sentence includes a large number. Furthermore, the output sentence structure completely changes even if only one number in the input sentence changes. In this paper, first, we optimized a neural machine translation model architecture that uses bidirectional RNN, LSTM, and the attention mechanism through data cleansing and changing the dictionary size. Then, we implemented a number-processing algorithm specialized in number translation and applied it to the neural machine translation model to solve the problems above. The paper includes the data cleansing method, an optimal dictionary size and the number-processing algorithm, as well as experiment results for translation performance based on the BLEU score.

• Journal of Digital Contents Society
• /
• v.16 no.4
• /
• pp.535-543
• /
• 2015

Recently, NAND flash based storage devices are being used as a storage device in various fields through hiding the limitations of NAND flash memory and maximizing the advantages. In particular, those storage devices contain a software layer called Flash Translation Layer(FTL) to hide the "erase-before-write" characteristics of NAND flash memory. FTL includes the metadata for managing the data requested from host. That metadata is stored in internal memory because metadata is very frequently accessed data for processing the requests from host. Thus, if the power-loss occurs, all data in memory is lost. So metadata management scheme is necessary to store the metadata periodically and to load the metadata in the initialization step. Therefore we proposed the scheme which satisfies the core requirements for metadata management and efficient operation. And we verified the efficiency of proposed scheme by experiments.

• Journal of KIISE:Computing Practices and Letters
• /
• v.14 no.2
• /
• pp.201-205
• /
• 2008

The flash memory is widely used as a main storage of embedded devices. It is adopted as a storage of database as growing the capacity of the flash memory. We run TPC-C benchmark on various FTL algorithms. But, the database shows poor performance on flash memory because the characteristic of I/O requests is full random. In this paper, we show the performance of all existing FTL algorithms is very poor. Especially, the FTL algorithm known as good at small mobile equipment shows worst performance. In addition, the chip-inter leaving which is a technique to improve the performance of the flash memory doesn't work well. In this paper, we inform you the reason that we need a new FTL algorithm and the direction for the database in the future.

• Journal of KIISE:Computing Practices and Letters
• /
• v.16 no.3
• /
• pp.371-375
• /
• 2010

SSD is a storage medium based on NAND Flash memory. Because of its short latency, low power consumption, and resistance to shock, it's not only used in PC but also in server computers. Most SSDs use FTL to overcome the erase-before-overwrite characteristic of NAND flash. There are several types of FTL, but page mapped FTL shows better performance than others. But its usefulness is limited because of its large memory footprint for the mapping table. For example, 64MB memory space is required only for the mapping table for a 64GB MLC SSD. In this paper, we propose a novel caching scheme for the mapping table. By using the mapping-table-meta-data we construct a fully associative cache, and translate the address within O(1) time. The simulation results show more than 80 hit ratio with 32KB cache and 90% with 512KB cache. The overall memory footprint was only 1.9% of 64MB. The time overhead of cache miss was measured lower than 2% for most workload.

• Journal of KIISE
• /
• v.44 no.11
• /
• pp.1130-1137
• /
• 2017

The multi-leveling technique that stores multiple bits in a single memory cell has significantly improved the density of NAND flash memory along with shrinking processes. However, because of the side effects of the multi-leveling technique, the average write performance of MLC NAND flash memory is degraded more than twice that of SLC NAND flash memory. In this paper, we introduce existing cross-layer optimization techniques proposed to improve the performance of MLC NAND flash-based storages, and propose a new integration technique that overcomes the limitations of existing techniques by exploiting their complementarity. By fully exploiting the performance asymmetry in MLC NAND flash devices at the flash translation layer, the proposed technique can handle many write requests with the performance of SLC NAND flash devices, thus significantly improving the performance of NAND flash-based storages. Experimental results show that the proposed technique improves performance 39% on average over individual techniques.

• Journal of Digital Contents Society
• /
• v.11 no.1
• /
• pp.91-97
• /
• 2010

Flash memory has been used by many digital devices for data storage, exploiting the advantages of non-volatility, low power, stability, and so on, with the help of high integrity, large capacity, and low price. As the fast growing popularity of flash memory, the density of it increases so significantly that its entire address mapping table becomes too big to be stored in SRAM. This paper proposes the associated page address cache with an efficient table management scheme for hybrid flash translation layer mapping. For this purpose, all tables are integrated into a map block containing entire physical page tables. Simulation results show that the proposed scheme can save the extra memory areas and decrease the searching time with less 2.5% of miss ratio on PC workload and can decrease the write overhead by performing write operation 33% out of total writes requested.

• The Journal of Society for e-Business Studies
• /
• v.15 no.4
• /
• pp.101-121
• /
• 2010

This paper presents an intelligent prefetching technique that significantly improves performance of hybrid fash-disk storage, a combination of flash memory and hard disk. Since flash memory embedded in a hybrid device is much faster than hard disk in terms of I/O operations, it can be utilized as a 'cache' space to improve system performance. The basic strategy for prefetching is to utilize sequential pattern mining, with which we can extract the access patterns of objects from historical access sequences. We use two techniques for enhancing the performance of hybrid storage with prefetching. One of them is to modify a FAST algorithm for mapping the flash memory. The other is to extend the unit of prefetching to a block level as well as a file level for effectively utilizing flash memory space. For evaluating the proposed technique, we perform the experiments using the synthetic data and real UCC data, and prove the usability of our technique.

• Journal of the Korea Society of Computer and Information
• /
• v.22 no.7
• /
• pp.1-7
• /
• 2017

As the capacity of NAND flash module increases, the amount of RAM increases for caching and maintaining the FTL mapping information. In order to reduce the amount of mapping information managed in the RAM, a demand-based address mapping method stores the entire mapping information in the flash and some valid mapping information in the form of cache in the RAM so that the RAM can be used efficiently. However, when cache miss occurs, it is necessary to read the mapping information recorded in the flash, so overhead occurs to translate the address. If the RAM space is not enough, the cache hit ratio decreases, resulting in greater overhead. In this paper, we propose a method using two tables called TPMT(Translation Page Mapping Table) and SMT(Segmented Translation Page Mapping Table) to utilize both temporal locality and spatial locality more efficiently. A performance evaluation shows that this method can improve the cache hit ratio by up to 30% and reduces the extra translation operations by up to 72%, compared to the TPM scheme.

• ETRI Journal
• /
• v.33 no.5
• /
• pp.731-740
• /
• 2011

Reconfigurable computing using a field-programmable gate-array (FPGA) device has become a promising solution in system design because of its power efficiency and design flexibility. To bring the benefit of FPGA to many application programmers, there has been intensive research about automatic translation from high-level programming languages (HLL) such as C and C++ into hardware. However, the large gap of syntaxes and semantics between hardware and software programming makes the translation challenging. In this paper, we introduce a new approach for the translation by using the widely used GCC compiler. By simply adding a hardware description language (HDL) backend to the existing state-of- the-art compiler, we could minimize an effort to implement the translator while supporting full features of HLL in the HLL-to-HDL translation and providing high performance. Our translator, called GCC2Verilog, was implemented as the GCC's cross compiler targeting at FPGAs instead of microprocessor architectures. Our experiment shows that we could achieve a speedup of up to 34 times and 17 times on average with 4-port memory over PICO microprocessor execution in selected EEMBC benchmarks.

• Journal of Internet Computing and Services
• /
• v.12 no.5
• /
• pp.63-72
• /
• 2011

Binary translation is a kind of the emulation method which converts a binary code compiled on the particular instruction set architecture to the new binary code that can be run on another one. It has been mostly used for migrating legacy systems to new architecture. In recent, binary translation is used for instrumenting programs without modifying source code, because it enables inserting additional codes dynamically, For general application, there already exists some instrumentation software using binary translation, such as dynamic binary analyzers and virtual machine monitors. On the other hand, in order to be benefited from binary translation in kernel-level, a few issues, which include system performance, memory management, privileged instructions, and synchronization, should be treated. These matters are derived from the structure of the kernel, and the difference between the kernel and user-level application. In this paper, we present a scheme to apply binary translation and dynamic instrumentation on kernel. We implement it on Linux kernel and demonstrate that kernel-level binary translation adds an insignificant overhead to performance of the system.