• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.11a
• /
• pp.12-15
• /
• 2020

Data access characteristics can directly affect the efficiency of the system execution. This research is to design an accurate predictor by using historical memory access information, where highly accessible data can be migrated from low-speed storage (SSD/HHD) to high-speed memory (Memory/CPU Cache) in advance, thereby reducing data access latency and further improving overall performance. For this goal, we design a locally weighted linear regression prefetch scheme to cope with irregular access patterns in large graph processing applications for a DARM-PCM hybrid memory structure. By analyzing the testing result, the appropriate structural parameters can be selected, which greatly improves the cache prefetching performance, resulting in overall performance improvement.

• Journal of the Korean Data and Information Science Society
• /
• v.26 no.5
• /
• pp.1035-1045
• /
• 2015

Recent studies in Big Data Analysis are showing promising results, utilizing the main memory for rapid data processing. In-memory computing technology can be highly advantageous when used with high-performing servers having tens of gigabytes of RAM with multi-core processors. The constraint in network in these infrastructure can be lessen by combining in-memory technology with distributed parallel processing. This paper discusses the research in the aforementioned concept applying to a test taxi hailing application without disregard to its underlying RDBMS structure. The application of IMDG technology in the application's backend API without restructuring the database schema yields 6 to 9 times increase in performance in data processing and throughput. Specifically, the change in throughput is very small even with increase in data load processing.

• Smart Media Journal
• /
• v.3 no.1
• /
• pp.33-38
• /
• 2014

Even though the performance of microprocessor is improved continuously, the performance improvement of computing system becomes hard to increase, in order to some drawbacks including increased power consumption. To solve the problem, general-purpose computing on graphics processing units(GPGPUs), which execute general-purpose applications by using specialized parallel-processing device representing graphics processing units(GPUs), have been focused. However, the characteristics of applications related with graphics is substantially different from the characteristics of general-purpose applications. Therefore, GPUs cannot exploit the outstanding computational resources sufficiently due to various constraints, when they execute general-purpose applications. When designing GPUs for GPGPU, memory system is important to effectively exploit the GPUs since typically general-purpose applications requires more memory accesses than graphics applications. Especially, external memory access requiring long latency impose a big overhead on the performance of GPUs. Therefore, the GPU performance must be improved if hierarchical memory architecture which can reduce the number of external memory access is applied. For this reason, we will investigate the analysis of GPU performance according to hierarchical cache architectures in executing various benchmarks.

• JSTS:Journal of Semiconductor Technology and Science
• /
• v.14 no.4
• /
• pp.391-406
• /
• 2014

The performance of General-Purpose computation on Graphics Processing Units (GPGPU) is heavily dependent on the memory access behavior. This sensitivity is due to a combination of the underlying Massively Parallel Processing (MPP) execution model present on GPUs and the lack of architectural support to handle irregular memory access patterns. Application performance can be significantly improved by applying memory-access-pattern-aware optimizations that can exploit knowledge of the characteristics of each access pattern. In this paper, we present an algorithmic methodology to semi-automatically find the best mapping of memory accesses present in serial loop nest to underlying data-parallel architectures based on a comprehensive static memory access pattern analysis. To that end we present a simple, yet powerful, mathematical model that captures all memory access pattern information present in serial data-parallel loop nests. We then show how this model is used in practice to select the most appropriate memory space for data and to search for an appropriate thread mapping and work group size from a large design space. To evaluate the effectiveness of our methodology, we report on execution speedup using selected benchmark kernels that cover a wide range of memory access patterns commonly found in GPGPU workloads. Our experimental results are reported using the industry standard heterogeneous programming language, OpenCL, targeting the NVIDIA GT200 architecture.

• Journal of KIISE:Computer Systems and Theory
• /
• v.34 no.9
• /
• pp.445-458
• /
• 2007

This article presents a novel infrastructure for large-memory database processing using commodity hardware with operating system support. We exploit inexpensive PCs and a high-speed network capable of Remote Direct Memory Access (RDMA) operations to build a new memory hierarchy between fast volatile memory and slow disk storage. The new memory hierarchy guarantees a reasonable response time, and its storage size enables us to run large-memory database systems with little performance degradation. The proposed architecture has two main components: (1) a remote memory system inside the Linux kernel to manage other computers' memory pages efficiently and (2) a remote memory pager responsible for manipulating remote read/write operations on remote memory pages. We insist that the proposed architecture is practical enough to support the rigorous demands of commercial in-memory database systems by demonstrating the performance of publicly available main-memory databases (e.g., MySQL) on our prototyped system. The experimental results show very interesting results from the TPC-C benchmark.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2009.11a
• /
• pp.543-544
• /
• 2009

본 논문은 Linux에서 사용되는 Shared Memory는 동일한 메모리 영역에 여러 개의 프로세스가 접근할 수 있도록 해 주는 기술이다. 본 논문에서는 Shared Memory의 큰 두 갈래 중 커널 단계에서 처리 되는 SVR(System V Release) 형식의 Shared Memory를 다룬다. 본 논문에서는 리눅스 운영체제의 공유 메모리 기능을 Dual Core 시스템에서 동작하도록 구현한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.345-347
• /
• 2023

Rust has gained popularity as a memory safe systems programming language. At the center of its memory safety is a strict memory ownership model with stringent rules enforced by the compiler. This paper aims to shed light on this memory safety model and the role smart pointers play towards its success. We study specific smart pointers, their purposes and contribution to Rust's memory safety. We further explore weaknesses of these smart pointers and their APIs, and provide scenarios under which they may lead to memory vulnerabilities in Rust programs.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.5
• /
• pp.41-51
• /
• 1996

In this paper, anew parallel processing system based on a cluster architecture which provides scalability of a parallel processing system while maintains shared memory multiprocessor characteristics is proposed. In recent days low cost, high performnce microprocessors have led to construction of large scale parallel processing systems. Such parallel processing systems provides large scalability but are mainly used for scientific applications which have large data parallelism. A shared memory multiprocessor system like TICOM is currently used as aserver for the commercial application, however, the shared memory multiprocessor system is known to have very limited scalability. The proposed architecture can support scalability and performance of the parallel processing system while it provides adaptability for the commerical application, hence it can overcome the limitation of the shared memory multiprocessor. The architecture and characteristics of the proposed system shall be described. A proprietary hierarchical crsossbar network is designed for this system, of which the protocol, routing and switching technique and the signal transfer technique are optimized for the proposed architecture. The design trade-offs for the network are described in this paper and with simulation usihng the SES/workbench, it is explored that the network fits to the proposed architecture.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.4
• /
• pp.217-228
• /
• 2022

Currently, there are increasing demands for applying deep neural networks (DNNs) in the embedded domain such as classification and object detection. The DNN processing in embedded domain often requires custom hardware such as NPU for acceleration due to the constraints in power, performance, and area. Processing DNN models requires a large amount of data, and its seamless transfer to NPU is crucial for performance. In this paper, we developed a cycle-accurate NPU simulator to evaluate diverse NPU microarchitectures. In addition, we propose a novel technique for reducing the number of memory accesses when processing convolutional layers in convolutional neural networks (CNNs) on the NPU. The main idea is to reuse data with memory interleaving, which recycles the overlapping data between previous and current input windows. Data memory interleaving makes it possible to quickly read consecutive data in unaligned locations. We implemented the proposed technique to the cycle-accurate NPU simulator and measured the performance with LeNet-5, VGGNet-16, and ResNet-50. The experiment shows up to 2.08x speedup in processing one convolutional layer, compared to the baseline.

• Journal of the Korean Society of Industry Convergence
• /
• v.24 no.1
• /
• pp.69-77
• /
• 2021

The widely used low-cost design methodology for IoT devices is very popular. In such a networked device, memory is composed of flash memory, SRAM, DRAM, etc., and because it processes a large amount of data, memory design is an important factor for system performance. Therefore, each device selects optimized design factors such as function, performance and cost according to market demand. The design of a memory architecture available for low-cost IoT devices is very limited with the configuration of SRAM, flash memory, and DRAM. In order to process as much data as possible in the same space, an architecture that supports parallel processing units is usually provided. Such parallel architecture is a design method that provides high performance at low cost. However, it needs precise software techniques for instruction and data mapping on the parallel architecture. This paper proposes an instruction/data mapping method to support optimized parallel processing performance. The proposed method optimizes system performance by actively using hardware and software parallelism.