• The Journal of Bigdata
• /
• v.9 no.1
• /
• pp.101-113
• /
• 2024

Federated learning has garnered attention as an efficient method for training machine learning models in a distributed environment while maintaining data privacy and security. This study proposes a novel FedRFBagging algorithm to optimize the performance of random forest models in such federated learning environments. By dynamically adjusting the trees of local random forest models based on client-specific data characteristics, the proposed approach reduces communication costs and achieves high prediction accuracy even in environments with numerous clients. This method adapts to various data conditions, significantly enhancing model stability and training speed. While random forest models consist of multiple decision trees, transmitting all trees to the server in a federated learning environment results in exponentially increasing communication overhead, making their use impractical. Additionally, differences in data distribution among clients can lead to quality imbalances in the trees. To address this, the FedRFBagging algorithm selects only the highest-performing trees from each client for transmission to the server, which then reselects trees based on impurity values to construct the optimal global model. This reduces communication overhead and maintains high prediction performance across diverse data distributions. Although the global model reflects data from various clients, the data characteristics of each client may differ. To compensate for this, clients further train additional trees on the global model to perform local optimizations tailored to their data. This improves the overall model's prediction accuracy and adapts to changing data distributions. Our study demonstrates that the FedRFBagging algorithm effectively addresses the communication cost and performance issues associated with random forest models in federated learning environments, suggesting its applicability in such settings.

• Journal of KIISE
• /
• v.43 no.5
• /
• pp.524-530
• /
• 2016

The desire to access data faster and the growth of next-generation memories such as non-volatile memories, contribute to the development of research on memory file systems. It is recommended that memory mapped file I/O, which has less overhead than read-write I/O, is utilized in a high-performance memory file system. Memory mapped file I/O, however, brings a page table overhead, which becomes one of the big overheads that needs to be resolved in the entire file I/O performance. We find that same overheads occur unnecessarily, because a page table of a file is removed whenever a file is opened after being closed. To remove the duplicated overhead, we propose the mapping cache, a technique that does not delete a page table of a file but saves the page table to be reused when the mapping of the file is released. We demonstrate that mapping cache improves the performance of traditional file I/O by 2.8x and web server performance by 12%.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.41 no.11
• /
• pp.1472-1480
• /
• 2016

When disasters such as earthquake, flood, typhoon, fire and terrorism are occurred a video management system is required which can shoot video on the disaster site, and send it to a server located in the command center that can grasp at a glance the site conditions. In this paper, to know the performance elements information needed to construct the video management system. we presented a method of configuring an ad hoc network based on Wi-Fi direct communication. And we also evaluated the performance through built of environment for the video management system. The evaluated performance test items are 'packet flow tests according to the video compression ratios and the image sizes', 'packet flow tests according to the distance and traffic tests', 'packet flow tests according to whether LOS or not', and 'performance test of scenarios'.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.7
• /
• pp.675-683
• /
• 2000

For real time services of multimedia applications, storage systems should provide data without degrading their performance. Since it is very important to recover data immediately at a disk failure, the disk recovery system is required. This paper presents a hybrid recovery scheme which prevents degraded the performance on a single disk failure at RAID level 5 architecture until the failed disk is replaced with a new disk. The proposed scheme is very economical compared to previous spare disk schemes because it does not use extra disks. The performance of the proposed scheme is evaluated and analyzed with that of the RAID level 5 for various requested sizes through the simulation. The results show that the performance of the proposed scheme is improved up to 20 percents at the failure mode and 80 percents at the reconfigured mode. After a multimedia server system has been built with a RAID controller and hard disks, the data recovery performance of the propose scheme are compared with the results of simulation.

• Journal of KIISE:Computer Systems and Theory
• /
• v.37 no.6
• /
• pp.339-347
• /
• 2010

Inter-process communication (IPC), which is a technique that enables exchanging data among multiple processes, is commonly used not only in user applications but also in system processes. For this reason, the performance of IPC highly influences the performance of whole computer system. Especially, heavy overload on a single server process caused by IPC requests from multiple client processes, easily results overall slowdown of IPC response time. Here, to deal with the problem stated above, the time-slice donation technique which is adapted in L4 microkernel is analyzed and enhanced for reducing latency of IPC response time and implemented on linux kernel for actual performance evaluation. While trying to maintain the additional overhead as least as possible, the experiment shows that the use of this technique enhances the performance of IPC multiple times of existing technique under certain circumstances.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38B no.12
• /
• pp.976-983
• /
• 2013

The targets for the traditional IT management solutions has been mostly systems (server) and networks. Such that management system for system and network, SMS and NMS respectively, used be the IT management solutions. Now the speed of hardware improvement is faster than ever, the performance improvement scope is shifting from hardware towards software. Therefore, it is not arguable that the business success depends on the efficient and optimized application performance management. There are many different approaches are developed in the application performance management, but in this study, API-based logging system that most efficiently visualizes the correlation of call and response between Components and enables active performance management for monitoring real-time applications operation status and resolving the problems and failures quickly.

• Journal of KIISE:Computer Systems and Theory
• /
• v.30 no.9
• /
• pp.445-453
• /
• 2003

As cluster systems used for high performance computing consist of large number of running servers, one has to solve the low availability problems occurred by the high chance of the server failures. To handle the problems, it is necessary to have the precise definition of available performance of cluster systems that represents availability and performability of the systems simultaneously. Previous research results that mention availability issues lack for concerning system performance such as waiting time and response time in their availability definition. In this paper, we propose a new availability metric for (n, k)-way cluster systems which compose of n primary servers and k backup servers. With the metric, the change of system performance according to arrival rates is captured and the waiting time of a request can be kept below to a certain level. Using various system operating parameters, we calculate availability and downtime of cluster systems along with waiting tine deadline.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.14 no.2
• /
• pp.415-420
• /
• 2010

This paper describes the design overview of shared file system $SANique^{TM}$ and analyzes the performance evaluation results of I/O intensive stress test based on various cluster file system architectures. Especially, we illustrate the performance analysis for the comparison results between the $SANique^{TM}$ and the Linux file system EXT3 system that is used to generally in Unix world. In order to perform our evaluation, Oracle 10g database system is operated on the top of cluster file system, and we developed the various kinds of testing tools which are compiled by ESQL/C from Oracle. Three types of architectures are used in this performance evaluation. Those are the cluster file system $SANique^{TM}$, EXT3 and the combined architecture of $SANique^{TM}$ and EXT3. In this paper, we present that the results of $SANique^{TM}$ outperforms other cluster file systems in the overhead for providing the true sharing over the connecting server nodes.

• Journal of the Korea Society of Computer and Information
• /
• v.18 no.12
• /
• pp.35-43
• /
• 2013

This paper suggests a way to implement a system that broadcasts compressed JPEG images obtained from a webcam to multiple users on their smartphones. The system was implemented to maintain a suitable image quality so that things are identifiable while using the minimal amount of data, in order to deliver as many frames as possible to as many users as possible. Also, the suggested way was applied and various tests were done, including on: the performance of the server that provides image information; the performance of the client that receives the image information and displays it on the smartphone; and the max. number of simultaneous users supported by the system. When a broadcasting system is implemented using webcams and smartphones, the results of this paper can be used in estimating the suitable system parameters depending on network performance, including the max. number of simultaneous clients supported, the client smartphone performance required, and the number of frames that can be transmitted per second.

• The Journal of the Korea Contents Association
• /
• v.8 no.10
• /
• pp.20-26
• /
• 2008

The quality of stream remaking is decided the load of a server and Jitter through the traffic of the transmission path between end to end. In order to improve these problems in this paper, I propose tree construction method of low-delay-level-multicast. In this case which the network congestion will be occurred by streaming quality, I also propose the technique which dynamically changes the transmission path. This technique first constructs the overlay structure for relaxing the overload of server. Secondly, in order to decrease Jitter of client, it makes upload bandwidth and low latency balanced. In the evaluation of the performance, this paper showed better enhancement of about $15%{\sim}24%$ than P2CAST[4] in the simulation about node average join count, average bandwidth, service request refusal ratio, RTT measurement of nodes, and node average join count by defect ratio.