• 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.

• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.11
• /
• pp.319-328
• /
• 2023

Unlike cloud computing, edge computing technology analyzes and judges data close to devices and users, providing advantages such as real-time service, sensitive data protection, and reduced network traffic. EdgeX Foundry, a representative open source of edge computing platforms, is an open source-based edge middleware platform that provides services between various devices and IT systems in the real world. EdgeX Foundry provides a service for handling camera devices, along with a service for handling existing sensed data, which only supports simple streaming and camera device management and does not store or process image data obtained from the device inside EdgeX. This paper presents a technique that can store and process image data inside EdgeX by applying some of the services provided by EdgeX Foundry. Based on the proposed technique, a service pipeline for object detection services used core in the field of autonomous driving was created for experiments and performance evaluation, and then compared and analyzed with existing methods.

• Genomics & Informatics
• /
• v.18 no.1
• /
• pp.8.1-8.10
• /
• 2020

The explosive growth of next-generation sequencing data has resulted in ultra-large-scale datasets and ensuing computational problems. In Korea, the amount of genomic data has been increasing rapidly in the recent years. Leveraging these big data requires researchers to use large-scale computational resources and analysis pipelines. A promising solution for addressing this computational challenge is cloud computing, where CPUs, memory, storage, and programs are accessible in the form of virtual machines. Here, we present a cloud computing-based system, Bio-Express, that provides user-friendly, cost-effective analysis of massive genomic datasets. Bio-Express is loaded with predefined multi-omics data analysis pipelines, which are divided into genome, transcriptome, epigenome, and metagenome pipelines. Users can employ predefined pipelines or create a new pipeline for analyzing their own omics data. We also developed several web-based services for facilitating downstream analysis of genome data. Bio-Express web service is freely available at https://www. bioexpress.re.kr/.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.10 no.12
• /
• pp.2187-2192
• /
• 2006

This paper presents a method of constructing the universal multiple processing element unit(UMPEU) by De Bruijn Graph. The second method is as following. First, we propose transformation operators in order to construct the De Bruijn UMPEU using properties of graph. Second, we construct the transformation table of De Bruijn graph using above transformation operators. Finally we construct the De Bruijn graph using transformation table. The proposed UMPEU be able to construct the De Bruijn graph for any prime number and integer value of finite fields. Also the UMPEU is applied to fault-tolerant computing system, pipeline class. parallel processing network, switching function and its circuits.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.7
• /
• pp.1-7
• /
• 2022

Recently, as the use of applications such as big data programs and machine learning programs that are driven while generating large amounts of data in the program itself becomes common, the existing main memory alone lacks memory, making it difficult to execute the program quickly. In particular, the need to derive results more quickly has emerged in a situation where it is necessary to analyze whether the entire sequence is genetically altered due to the outbreak of the coronavirus. As a result of measuring performance by applying large-capacity data to a computing system equipped with a self-developed memory pool MOCA host adapter instead of processing large-capacity data from an existing SSD, performance improved by 16% compared to the existing SSD system. In addition, in various other benchmark tests, IO performance was 92.8%, 80.6%, and 32.8% faster than SSD in computing systems equipped with memory pool MOCA host adapters such as SortSampleBam, ApplyBQSR, and GatherBamFiles by task of workflow. When analyzing large amounts of data, such as electrical dielectric pipeline analysis, it is judged that the measurement delay occurring at runtime can be reduced in the computing system equipped with the memory pool MOCA host adapter developed in this research.

• Journal of KIISE:Computing Practices and Letters
• /
• v.13 no.5
• /
• pp.293-299
• /
• 2007

Intelligent pipeline inspection gauges (PIGs) are inspection vehicles that move along within a (gas or oil) pipeline and acquire signals (also called sensor data) from their surrounding rings of sensors. By analyzing the signals captured in intelligent PIGs, we can detect pipeline defects, such as holes and curvatures and other potential causes of gas explosions. There are two major data access patterns apparent when an analyzer accesses the pipeline signal data. The first is a sequential pattern where an analyst reads the sensor data one time only in a sequential fashion. The second is the repetitive pattern where an analyzer repeatedly reads the signal data within a fixed range; this is the dominant pattern in analyzing the signal data. The existing PIG software reads signal data directly from the server at every user#s request, requiring network transfer and disk access cost. It works well only for the sequential pattern, but not for the more dominant repetitive pattern. This problem becomes very serious in a client/server environment where several analysts analyze the signal data concurrently. To tackle this problem, we devise a fast in-memory cache manager, called T-Cache, by considering pipeline sensor data as multiple time-series data and by efficiently caching the time-series data at T-Cache. To the best of the authors# knowledge, this is the first research on caching pipeline signals on the client-side. We propose a new concept of the signal cache line as a caching unit, which is a set of time-series signal data for a fixed distance. We also provide the various data structures including smart cursors and algorithms used in T-Cache. Experimental results show that T-Cache performs much better for the repetitive pattern in terms of disk I/Os and the elapsed time. Even with the sequential pattern, T-Cache shows almost the same performance as a system that does not use any caching, indicating the caching overhead in T-Cache is negligible.

• Proceedings of the IEEK Conference
• /
• 2002.07a
• /
• pp.98-101
• /
• 2002

This paper reports a parallel architecture in a HMM based speech recognition system for a low-power LSI design. The proposed architecture calculates output probability of continuous HMM (CHMM) by using concurrent and pipeline processing. They enable to reduce memory access and have high computing efficiency. The novel point is the efficient use of register arrays that reduce memory access considerably compared with any conventional method. The implemented system can achieve a real time response with lower clock in a middle size vocabulary recognition task (100-1000 words) by using this technique.

• The Journal of Information Technology
• /
• v.12 no.2
• /
• pp.39-47
• /
• 2009

Residue Number System is used for the purpose of increasing the speed of processing in the many application parts of Image Processing, Computer Graphic, Neural Computing, Digital Signal Processing etc, since it has the characteristic of parallelism and no carry propagation at each moduli. DRNS has the twice RNS Conversion, it is used to decreases the size of the operator in RNS. But it has a week point on the Second Residue to First Residue Conversion time. So, in this paper SRTFR(Second Residue to First Residue) Converter using MRC(Mixed Radix Conversion) is designed to decrease the size of RTB(Residue to Binary) Converter. Since the proposed SRTFR Converter using MRC(Mixed Rdix Convertion) has a pipeline processing. Also, modular operation is applied to at each partitioned SAM(Subtraction and Addition) and MA(Multiplication and addition). In the following study, the more effective design on MA is needed.

• Proceedings of the IEEK Conference
• /
• 2002.07b
• /
• pp.959-962
• /
• 2002

This paper presents a method of constructing the universal multiple processing element unit(UMPEU) based on De Bruijn Graph. The proposed method is as following. Firstly we propose transformation operators in order to construct the De Bruijn graph using properties of graph. Secondly we construct the transformation table of De Bruijn graph using above transformation operators. Finally we construct the De Bruijn graph using transformation table. The proposed UMPEU is capable of constructing the De Bruijn geraph for any prime number and integer value of finite fields. Also the UMPEU is applied to fault-tolerant computing system, pipeline class, parallel processing network, switching function and its circuits.

• The Bulletin of The Korean Astronomical Society
• /
• v.43 no.1
• /
• pp.39.2-39.2
• /
• 2018

GW170817/GRB170817A establishes a double neutron star merger as the progenitor of a short gamma-ray burst, starting 1.7 s post-coalescence. GRB170817A represents prompt or continuous emission from a newly formed hyper-massive neutron star or black hole. We report on a deep search for broadband extended gravitational-wave emission in spectrograms up to 700 Hz of LIGO O2 data covering this event produced by butterfly filtering comprising a bank of templates of 0.5 s. A detailed discussion is given of signal-to-noise ratios in image analysis of spectrograms and confidence levels of candidate features. This new pipeline is realized by heterogeneous computing with modern graphics processor units (GPUs). (Based on van Putten, M.H.PM., 2017, PTEP, 093F01.)