• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.2-4
• /
• 2022

매니코어 CPU 시스템에서의 병렬 I/O 는 현재의 리눅스 시스템의 LRU 관리 방법의 한계로 확장성에 문제를 가지고 있다. 본 연구에서는 이 문제를 해결했던 하기 위한 개선된 FinerLRU 를 제안한다. LRU 락을 최대 코어 개수만큼 증가시키고 세분화된 Lock 관리를 통해 버퍼 캐시를 사용하는 파일 시스템의 병렬 I/O 성능을 향상시킨다. 리눅스 5.18.11 에 제안한 방법을 구현하였으며, 64 개의 물리적 코어와 256 개의 논리적 코어를 가지는 Intel Knights Landing 프로세서를 이용한 실험을 통해 두 배 가량의 성능 향상을 얻을 수 있음을 확인하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.50-52
• /
• 2023

ML/DL과 같은 AI의 연구가 HPC 환경에서 수행되면서 데이터 병렬화, 분산 학습 및 대규모 데이터 세트를 처리를 위한 요구사항이 급격히 증가하였다. 또한, 병렬처리 연산에 특화된 가속기 기반 이기종 아키텍처 환경 변화로 I/O 처리에 고대역폭, 저지연의 스토리지 기술을 필요로 하고 있다. 본 논문에서는 고집적의 병렬 컴퓨팅 환경에 고성능 HPC, AI 애플리케이션을 처리하기 위한 티어링 스토리지 기술을 논한다. 나아가 실제 고성능 NVMe 기반의 플래시 티어링 계층 구성에서 액세스 패턴에 따른 데이터 처리 환경을 구축하고 성능을 검증한다. 이로써 다양한 사용자 어플리케이션의 I/O 패턴을 특성에 맞게 지원할 수 있다.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.8
• /
• pp.59-65
• /
• 2016

Performance optimization of applications with massive stream data processing has been performed by considering I/O data coherency problem where a memory is shared between processors and hardware accelerators. A formula for performance analyses is derived based on profiling results of system-level simulations. Our experimental results show that overall performance was improved by 1.40 times on average for various image sizes. Also, further optimization has been performed based on the parameters appeared in the derived formula. The final performance gain was 3.88 times comparing to the original design and we can find that the performance of the design with cacheable shared memory is not always.

• Proceedings of the Korea Contents Association Conference
• /
• 2007.11a
• /
• pp.513-516
• /
• 2007

Because many scientific applications require large data processing, the importance of parallel I/O has been increasingly recognized. Collective I/O is one of the considerable features of parallel I/O and enables application programmers to easily handle their large data volume. In this paper we measure and analyze the performance of original collective I/Os and the subgroup method, the way of using collective I/O of MPI effectively. From the experimental results, we found that the two kinds of subgroup method showed different performance. In terms of collective write operation, the subgroup method caused the performance degradation. However, the subgroup method for collective read showed good performance with small data size.

• The KIPS Transactions:PartC
• /
• v.12C no.4 s.100
• /
• pp.603-608
• /
• 2005

Disk I/O is a major performance bottleneck of web proxy server. Today's most web proxy sowers are design to run on top of a general purpose file system. But general purpose file system can not efficiently handle web cache workload, small files, leading to the performance degradation of entire web proxy servers. In this paper we evaluate the performance potential of raw disk to reduce disk I/O overhead of web proxy servers. To show the performance potential of raw disk, we design a storage management system called Block-structured Storage Management System (BSMS). And we also actually implement web proxy server that incorporate BSMS in Squid. Comprehensive experimental evaluations show that raw disk can be a good solution to improve disk I/O performance significantly for web proxy servers.

• 한국정보통신설비학회:학술대회논문집
• /
• 2006.08a
• /
• pp.77-81
• /
• 2006

Because of the rapid increasing of network user, there are some problems to tolerate the network overhead. Recently, the research and technology of the user-level for high performance and low latency than TCP/IP which relied upon the kernel for processing the messages. For example, there is an Infiniband technology. The Infiniband Trade Association (IBTA) has been proposed as an industry standard for both communication between processing node and I/O devices and for inter-processor communication. It replaces the traditional bus-based interconnect with a switch-based network for connecting processing node and I/O devices. Also Infiniband uses RDMA (Remote DMA) for low latency of CPU and OS to communicate between Remote nodes. In this paper, we develop the SRP (SCSI RDMA Protocol) which is Storage Access Protocol on Infiniband network. And will compare to FC (Fibre Channle) based I-SCSI (Internet SCSI) that it is used to access storage on Etherent Fabric.

• Journal of Powder Materials
• /
• v.25 no.3
• /
• pp.213-219
• /
• 2018

We report a method for preparing rare earth oxides ($Re_xO_y$) from the recycling process for spent Ni-metal hydride (Ni-MH) batteries. This process first involves a leaching of spent Ni-MH powders with sulfuric acid at $90^{\circ}C$, resulting in rare earth precipitates (i.e., $NaRE(SO_4)_2{\cdot}H_2O$, RE = La, Ce, Nd), which are converted into rare earth oxides via two different approaches: i) simple heat treatment in air, and ii) metathesis reaction with NaOH at $70^{\circ}C$. Not only the morphological features but also the crystallographic structures of all products are systematically investigated using field-emission scanning electron microscopy (FESEM) and X-ray diffraction (XRD); their thermal behaviors are also analyzed. In particular, XRD results show that some of the rare earth precipitates are converted into oxide form (such as $La_2O_3$, $Ce_2O_3$, and $Nd_2O_3$) with heat treatment at $1200^{\circ}C$; however, secondary peaks are also observed. On the other hand, rare earth oxides, RExOy can be successfully obtained after metathesis of rare earth precipitates, followed by heat treatment at $1000^{\circ}C$ in air, along with a change of crystallographic structures, i.e., $NaRE(SO_4)_2{\cdot}H_2O{\rightarrow}RE(OH)_3{\rightarrow}RE_xO_y$.

• Advances in concrete construction
• /
• v.3 no.2
• /
• pp.145-159
• /
• 2015

The evolution of nanotechnology provides materials with advance properties. It's a fast growing area of research to introduce the oxide nanoparticles into the cement pastes to improve their performance. The purpose of this paper is to review the effects of oxide nanoparticles (such as $SiO_2$, $TiO_2$, $Fe_2O_3$, $ZnO_2$, $Cr_2O_3$ and $Al_2O_3$) on both of hardened concrete properties (i.e., compressive strength, split tensile strength and flexural strength, water permeability, Abrasion resistance and pore structure of concrete) and fresh concrete properties (i.e., workability and setting time). Graphical representations of all these parameters were presented to facilitate the comparison of the effect of oxide nanoparticles on concrete processing. The paper also introduces some discussion about future work in this direction by identifying some open research area.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2009.11a
• /
• pp.775-776
• /
• 2009

공간데이터를 관리하는 공간 index structure는 대부분 순차처리를 위한 구조를 가지고 있다. 많은 응용분야에서 방대한 양의 공간 데이터는 보조기억장치(예: disk)에 저장이 되어 사용이 되고 공간 index structure의 operation은 I/O에 대한 의존도가 크므로, I/O operation의 병렬처리는 공간 index structure의 질의반응시간을 현저하게 줄일 수 있다. 본 논문에서는 PPR-tree라는 병렬형 공간 index structure를 제안한다.

• The KIPS Transactions:PartA
• /
• v.13A no.6 s.103
• /
• pp.517-524
• /
• 2006

In cluster computers, it is essential to Implement the single I/O space(SIOS) supporting integrated I/O substructure to efficiently process I/O intensive applications. SIOS service provides with global I/O address space to directly access peripherals and hard disks in its own or remote nodes from any node in the cluster computer In this thesis, we propose the implementation method of SIOS in Linux clusters by using only freewares. This method is implemented at device driver level that uses Enhanced Network Block Device(ENBD) and file system level that uses S/W RAID and NFS. The major strengths of this method are easiness of implementation and almost no cost due to using freewares. In addition, since freewares used are open sources, it is possible to apply this method to other platforms with only slight modification. Moreover, experiments show that I/O throughputs are up to 5.5 times higher in write operations and approximately 2.3 times higher in read operations than those of CDD method that uses the device driver developed at kernel level.