• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.12 no.3
• /
• pp.1380-1386
• /
• 2011

The request of interrupt accompanies a context switching. If the interrupt is frequently requested, this overhead of context switching can reduce seriously the performance of embedded systems. In order to reduce processing overhead due to frequently requested communication interrupts at Asynchronous Serial Communication, this paper introduces the method of Expanded Asynchronous Serial Communication with the Interrupt Coalescence(IC) that accumulates a fixed number of interrupts and processes them in one time. we implement the existing Asynchronous Serial Communication that requests communication interrupts by one byte at an LN2440SBC embedded board with a uC/OS-II and compare interrupt processing time for the performance evaluation about proposed method. As a result, the communication interrupt processing time of proposed method appears in case of low speed(9,600 bps), the decline of an average 25.18% at transmission, the decline of an average 41.47% at reception. and in case of hight speed(115,200 bps), the decline of an average 16.67% at transmission, the decline of an average 25.61% at reception.

• Proceedings of the Korean Information Science Society Conference
• /
• 2008.06d
• /
• pp.105-109
• /
• 2008

클러스터 파일시스템은 근거리 또는 원거리에 있는 클러스터 시스템 간에 연구데이터 공유 뿐 아니라, 실시간 계산을 위한 데이터 저장 공간으로 사용되는 네트워크 기반의 파일시스템이다. 고도의 과학계산을 처리할 때 계산노드들은 네트워크를 통해 연결된 클러스터 파일시스템으로부터 대용량의 데이터를 송수신하는 과정에서 CPU의 부하가 생기게 되고 이러한 문제는 계산노드로 하여금 과학계산의 속도를 저하시키는 요인이 된다. 본 논문에서는 패킷 송수신으로 인한 CPU 부하를 줄이고 이를 통하여 계산 성능을 향상시킬 목적으로 계산노드에서 수신하는 패킷들에 대해 인터럽트를 통합할 때 CPU 사용률에 미치는 영향을 분석하였다.

• The KIPS Transactions:PartC
• /
• v.13C no.5 s.108
• /
• pp.575-582
• /
• 2006

Guaranteeing network bandwidth and system performance becomes important technical matters to make seamless delivery of bulky data at high speed. Tuning system and network parameters including jumbo frame, kernel buffer size, and interrupt coalescence determines end-to-end transmission throughput. Additionally, fine-tuning of the parameters alleviates workload on high-end systems. Until now, many studies have concentrated on how to increase transmission throughput but rarely discussed how to mitigate system workload. In this paper, we have investigated various tuning parameters, which positively affect networking and processing performance of uncompressed HDTV system.