• JSTS:Journal of Semiconductor Technology and Science
• /
• v.17 no.4
• /
• pp.514-523
• /
• 2017

In this paper, we propose Adaptive Writeback-aware Cache management (AWC) to prolong the lifetime of non-volatile main memory systems by reducing the number of writebacks. The last-level cache in AWC is partitioned into Least Recently Used (LRU) segment and LRU using Dirty block Precedence (DP-LRU) segment. The DP-LRU segment evicts clean blocks first for giving reuse opportunity to dirty blocks. AWC can also determine the efficient size of DP-LRU segment for reducing the number of writebacks according to memory access patterns of programs. In the performance evaluation, we showed that AWC reduced the number of writebacks up to 29% and 46%, and saved the energy of a main memory system up to 23% and 49% in a single-core and multi-core, respectively. AWC also reduced the runtime by 1.5% and 3.2% on average compared to previous cache managements for non-volatile main memory systems, in a single-core and a multi-core, respectively.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.7
• /
• pp.45-51
• /
• 2019

Modern file systems have journaling mechanism to maintain their stored state consistently even under unexpected system crashes or disasters. However, the journaling makes I/O throughput lower. This performance degradation comes from the ordering mechanism between the data buffer and metadata buffer and two-staged buffer writing. Especially, if the data buffer and metadata buffer are journalled at the same time, then it incurs significant performance degradation due to the two-staged writing. That shows the trade-off relation-ship between I/O performance and system reliability. In this paper, we propose RFJ: a reliable and fast jour-naling mechanism to deal with this trade-off relationship. We propose an ordering enforced writeback journaling mode and selective journaling mechanism. The Ordering enforced writeback journaling mode achieves low I/O latency and the selective journaling mechanism achieves high reliability. The experimental result shows that the performance of RFJ is almost 5x faster than the journal mode of Ext3 file system but it still supports the same reliability with the journal mode.

• Journal of the Korea Society of Computer and Information
• /
• v.23 no.4
• /
• pp.1-8
• /
• 2018

In this paper, we propose an adaptive partition cache replacement policy and evaluate the performance of our scheme using various monitoring ratios to help lifetime extension of non-volatile main memory systems without performance degradation. The proposal combines conventional LRU (Least Recently Used) replacement policy and Early Eviction Zone (E2Z), which considers a dirty bit as well as LRU bits to select a candidate block. In particular, this paper shows the performance of non-volatile memory using various monitoring ratios and determines optimized monitoring ratio and partition size of E2Z for reducing the number of writebacks using cache hit counter logic and hit predictor. In the experiment evaluation, we showed that 1:128 combination provided the best results of writebacks and runtime, in terms of performance and complexity trade-off relation, and our proposal yielded up to 42% reduction of writebacks, compared with others.

• Annual Conference of KIPS
• /
• 2014.04a
• /
• pp.33-36
• /
• 2014

최근 스마트폰이나 태블릿 PC 등의 모바일 디바이스가 상용화 되어감에 따라 그 안에서 핵심적인 처리기능을 담당하는 프로세서의 코어 수가 점차적으로 늘어나고 있다. 많은 수의 코어를 효율적으로 사용하기 위해 여러 가지 메커니즘이 구현되어 있으나, 단일 프로세스를 순차적으로 실행하는 경우 여전히 성능에서의 한계가 존재한다. 병렬화 되어 있지 않은 프로세스의 경우, Amdahl's Law[1]에 따르면 순차적으로 실행을 할 수 밖에 없는 부분이 존재하고, 이 부분은 하나의 코어에서만 실행되기 때문에 많은 연산 자원들이 낭비되는 현상이 발생한다. 본 논문은 다중 코어 환경에서 이러한 잉여자원을 효과적으로 사용하기 위해 Back-end Fusion 이라는 구조를 제안하여 프로세서의 성능 향상을 위한 연구를 진행하였다. Back-end Fusion 이란, 연산 처리를 담당하는 back-end 부분(execution unit, writeback 단계 등)을 필요에 따라 코어 간에 동적으로 재구성하여 성능을 향상시키는 메커니즘이다. 이 재구성된 프로세서의 back-end 를 효율적으로 사용하기 위해, 종속성과 로드 밸런스 등을 고려한 인스트럭션 분배 알고리즘을 함께 제안한다. Intel 사의 x86 Instruction Set Architecture(ISA)를 기반으로 한 시뮬레이터를 이용하여 Back-end Fusion 프로세서의 성능을 측정 해 본 결과 기존의 단일 코어 프로세서에 비해 평균 32.2%의 성능 향상을 확인할 수 있었다.