Earticle

마이크로프로세서에 비해 상대적으로 느린 메모리의 속도를 극복하고, 시스템의 성능을 향상시키기 위하여 캐시가 도입되어 사용되고 있다. 이러한 캐시의 성능을 더욱 향상시키기 위하여 많은 노력들이 이루어지고 있으며 그러한 노력 중 하나로 최적화된 교체 알고리즘을 찾기 위한 연구들이 이루어지고 있다. 하지만 이러한 연구의 대부분은 상위 레벨 캐시에 집중되어 있다. 용량이 상대적으로 큰 최하위 레벨 캐시의 경우 상위 캐시에서와는 다른 특성들 을 가질 수 있다. 그렇기 때문에 상위 레벨 캐시와 똑같은 교체 알고리즘을 적용할 경우 최적의 성능을 보이지 못 하는 경우가 많다. 본 논문에서는 이러한 LLC (Last Level Cache)의 특성에 초점을 맞추어 그에 적합한 그룹 기 반 캐시 교체 알고리즘을 제안한다. 본 논문에서 제안하는 알고리즘에서는, 큰 배열에 속한 데이터가 다시 사용될 가능성이 높은 점을 감안하여, 데이터 그룹을 기반으로 교체될 캐시 라인을 선택한다. 본 논문에서는 제안하는 그 룹 기반 교체 알고리즘을 LLC에 적용하여 실험해보았으며, 이때의 캐시 실패율이 LRU 알고리즘과 비교하여 8.5% 낮게 나타났다. 이때, 그룹을 정의하는 방법에 따라 성능에 차이를 보일 수 있으며, 이에 대한 연구는 추후연구로 남겨둔다.

Caches were proposed to overcome the slow access speed of the memories compared to the microprocessors, and improve the performance of the system. In order to maximize the performance of the cache, many studies have been done. Among them, many of the researches focused on finding the optimal cache replacement algorithm. However, most of the studies were mainly focusing on the high level caches. The last level cache with relatively larger capacity, may have some unique properties different from the higher level caches. Therefore, using the same replacement algorithm that high level caches use, does not always result in best performance. In this paper, we propose a group based cache replacement algorithm that focuses on the properties of the LLC. In our proposed algorithm, the victim cache is selected based on data group, assuming that array data are more likely to be reused. We evaluated our proposed group based replacement algorithm on the LLC against the LRU algorithm. Cache miss rate of our proposed algorithm was 8.5% lower compared to the LRU algorithm. The result may vary depending on the methods for determining groups. We leave this as future work.

본 연구에서는 멀티 혐기배양기의 잔존 산소 농도 저감을 위한 3단계 자동 가스 치환 제어 알고리즘을 설계하고, PLC 기반 시퀀스 제어 모듈과 산업용 고정밀 DAQ 모듈로 구성된 이중 계층 통합 제어 플랫폼에 구현하였다. 배양실 내 산소(이후 O2) 농도 감소 동특성을 1차 선형 희석 모델로 수학적 모델링하여 Stage별 최적 SOL 파라미터를 설계하였으며, O2 농도에 따라 Stage A(O2>1%), B(0.5%≤O2≤1%), C(O2<0.5%)로 자동 전환되는 상태 전이 구조와 PLC 하드웨어 기반 압력 우선순위 안전 인터락을 구현하였다. 시뮬레이션 예측값(Stage B→C 달성 20분) 대비 실증값(20.7분)의 오차율 3.5%로 모델 타당성을 정량 확인하였으며, N=10 반복 정밀도(σ=0.008%)를 확인하였으며, 자체 종합 성능 평가 44항목 중 37항목 Pass(84.1%), Fail 0건, 미평가(N/A) 7항목의 결과를 얻었다. 본 연구에서 검토한 선행연구의 범위에서는 산소농도 구간별 차등 제어와 시뮬레이션-실증 정량 비교를 함께 제시한 사례가 확인되지 않았으며, 이는 혐기 배양기 제어 시스템 국산화의 기초 자료를 제공한다.

This study designed a three-stage automatic gas replacement control algorithm for residual oxygen reduction in a multi-anaerobic incubator, implemented on a PLC-based sequence control module integrated with an industrial high-precision DAQ module. A first-order linear dilution model was used to derive optimal SOL parameters per stage, enabling automatic transitions among Stage A (O2>1%), B (0.5–1%), and C (O2<0.5%), backed by a PLC-based pressure-priority safety interlock. Simulation predicted the B→C transition at 20 min versus 20.7 min measured on the prototype (3.5% error), validating the model. N=10 repeatability trials achieved σ=0.008%. The 44-item comprehensive performance evaluation resulted in 37 items passed (84.1%), 0 failed, and 7 items not yet evaluated (N/A). To the best of the authors' knowledge based on the reviewed literature, this study offers a differentiated multi-stage gas control approach with quantitative simulation-prototype validation, providing foundational data for domestic anaerobic incubator control development.

자동차를 기반으로 하는 애드 혹 네트워크 중 하나인 Vehicular Ad Hoc Network(VANET)은 자율 주행차와 커넥티드 카 등의 기술 개발과 더불어 많은 연구가 진행 중인 주제 중 하나이다. VANET을 구현하는 경우 기존의 TCP/IP를 이용한 연구도 진행하고 있지만, Content Centric Networking (CCN)을 이용한 VANET over CCN (VCCN) 관련 연구도 다양한 장점으로 인해 높은 실현 가능성을 보여주고 있다. 본 논문은 CCN 노드들이 관리하는 전송관련 정보 테이블 중에서 Content Store(CS)에 할당된 공간이 모두 사용되어 새로운 항목을 추가할 경우 기존 의 일부 항목을 교체하는 경우에 사용되는 다양한 교체 알고리즘의 성능을 평가하고 특정 네트워크 상황에서의 전송 특성을 분석하였다. 모의실험 결과, LRU 교체 알고리즘의 성능이 다른 세 가지 교체 알고리즘 보다 우수하였다. 또 한 CS 의 크기가 작더라도 전송 성능이 일정 수준을 유지하였으며, CS의 크기가 클수록 전송 성능은 그에 비례하여 상승하였다. 네트워크가 혼잡한 경우 또는 전송 노드 사이의 거리가 먼 경우 데이터 전송 성능도 감소하였다.

VANET (Vehicular Ad Hoc Network), an example of an ad hoc vehicular networks, becomes one of the popular research areas together with the self-driving cars and the connected cars. In terms of the VANET implementation, the traditional TCP/IP protocol stack could be applied to VANET. Recently, CCN (Content Centric Networking) shows better possibility to apply to VANET, called VCCN (VANET over CCN). CCN maintains several data tables including CS (Content Store) which keeps track of the currently requested content segments. When the CS becomes full and new content should be stored in CS, a replacement algorithm is needed. This paper compares and contrasts four replacement algorithms. In addition, it analyzes the transmission characteristics in diverse network conditions. According to the simulation results, LRU replacement algorithm shows better performances than the remaining three algorithms. In addition, even the size of CS is small, the network maintains a reasonable transmission performance. As the CS size becomes larger, the transmission rate increases proportionally. The transmission performance decreases when the network is crowded as well as the number of transmission hops becomes large.

Caching is one of the major steps in system designing to reach a higher performance in operating systems, databases and World Wide Web. High performance processors need memory systems with a proper access time, but still there is a big gap between performances of processors and memory systems. Virtual memory management and hierarchical memory models play an important role in system performance. In these architectures caching replacement policy defines the enhancement factor of the memory system and could modify the efficiency of the system. Different caching policies have different effects on the system performance. Because of the highlight role of replacement policies in the systems, there have been lots of work and proposed algorithms to overcome the problem of performance gap between processor and memory. Most of these policies are the enhancement of the Least-Recently-Used (LRU) and Least-Frequently-Used (LFU) schemes. Although most of the proposed schemes could solve the defects of the LRU and LFU, but they have lots of overhead and are difficult to implement. The most profit of LRU and LFU is their simple implementation. This article proposes an adaptive replacement policy which has low overhead on system and is easy to implement. This model is named Weighting Replacement Policy (WRP) which is based on ranking of the pages in the cache according to three factors. Whenever a miss occurs, a page with the lowest rank point is selected to be substituted by the new desired page. The most advantage of this model is it’s similarity to both LRU and LFU, which means it has the benefits of both (i.e. in cases like loops in which LRU fails, it will switch to LFU). Simulations show that this algorithm performs better than LRU and LFU. In addition, it performs similarly to LRU in the worst cases. The new approach can be applied as a replacement policy in virtual memory systems and web caching.

In this paper we proposed a recovery and replacement algorithm mainly used to increase or strengthen the life of a sensor node when they complete the battery power. In generally wireless sensor network consists of lot of sensor nodes, hundred(s) or thousand(s) of nodes form a network to transfer the data from source node to sink node. The sensor node lose their battery power when it is used for long period of time, so it is difficult to recharge nodes of the sensor and also main in wireless sensor network is reduce the data loss, energy consumption and recovery and replace the effected nodes. So here we proposed a recovery and replacement algorithm which is a combination of genetic algorithm and grade diffusion algorithm. Here each node contains either two values those are binary values 0, 1. 1 means battery life greater than the threshold values otherwise it will be 0. This approach is better than existed algorithms, the number of faults decreases, reduces the data loss up to 98% and reduce energy consumption to 32%.

OSGi is a middleware standard for home gateways, designed for smart home applications. OSGi models services as separate components, called bundles. Smart home applications might differ in their importance. For example, home security system is more important than Internet game. Bundles collaborate to provide the required service. This paper proposes a bundle replacement algorithm that takes into account the priority of the bundle and the interdependence between different bundles. Thus, given a home gateway that hosts several applications with different priorities and arbitrary dependencies among them. When the home gateway runs out of memory, which bundles will be stopped or kicked out of memory to start a new service. Because of the bundle dependencies, traditional memory management algorithms might not be efficient. Efficient replacement algorithm should stop the least important and a small number of bundles. The proposed algorithm takes into consideration the priority of the bundle or application and dependencies between different bundles, in addition to the amount of memory occupied by each service. We implement the proposed algorithms and performed many experiments to evaluate its performance and execution time. We used best fit and worst fit as yardstick to show the effectiveness of the proposed algorithms. The proposed algorithms are implemented as a part of the OSGi framework(Open Service Gateway initiative).

The concept of caching is a fundamental feature in modern computing architectures and, has no doubt, found wide applications in diverse areas. Performance evaluation of systems is functionally related to how caching is implemented on a given computing platform, a metric influenced by the cache replacement policy. This paper describes an online learning-induced, self-adjusting cache management strategy with low overhead and scan-resistant characteristics that outperforms the LRU replacement algorithm using adaptation to balance between workload frequency and recency patterns.

Regression testing remains one of the most laborious and costly software testing activities in software maintenance. Its laboriousness and costing increases with increasing number of test suits: need to changed re-testing strategy. The strategy should identify and eliminate redundant test cases in order to minimize the test cases from test suites. The aim of regression testing is to reduce the retesting effort and achieve the adequate testing coverage. An important research problem, in this context, is the new developed algorithm is needed to reduce the retesting effort, cost and time without disturbance of thoroughness. The traditional optimal page replacement algorithm is used for managing the virtual memory. Our approach has been showing that the novel optimal page replacement algorithm reducing the redundant test cases during retesting of modified object oriented program. Our testing approach is on unit level. The logic based transitive relationship model achieves the adequate testing coverage to the modified object oriented program. Hybridization of clock page replacement-counting based algorithm is used for making the logic based transition relationship model. An Illustrative example has been presented to establish the effectiveness of our methodology coupled with logic based transitive relationship model.

To improve these unscientific methods, I would like to design a system that automatically tells replacement time of engine oil by fuzzy algorithm. There are ‘time series analysis’ and ‘cause and effect’ analysis in quantitative method. Time series analysis is to forecast the value by choosing one variable with the basis of historical data. All methods of demanding forecasting belong to time series analysis. Cause and effect analysis is the method, which makes a model by the relation between past values of other variables and the particular variable when the variable is influenced by other variables. In the thesis to replace engine oil in optimum time and to be aware of right replacement time of engine oil automatically.