Okeanos: Fast and Reliable Stream Storage Through Differential Data Journaling
Andromachi Hatzieleftheriou, Stergios V. Anastasiadis · 2008
Real-time storage of massive stream data is emerging as a critical component in modern computing infrastructures used for continuous monitoring purposes. Traditional file and database systems are not designed for such operation en-vironments and incur excessive resource requirements when handling high-volume streaming traffic. In the present pa-per, we examine the possibility of employing data journaling in order to combine sequential throughput with low latency during synchronous writes. Experimentally we demonstrate that low-rate streams incur remarkably high data journaling traffic in a commonly-used production file system. In order to alleviate the problem, we designed and implemented dif-ferential data journaling in the kernel of the above system. Through extensive experimentation we show substantial re-duction in the required disk throughput combined with very low write latency.