A stable and dependable grid computing testbed has been highly demanded and strived to build with development and deployment of advanced middleware for several years. Since 2000, ApGrid (Asia Pacific Grid Partnership) has been coordinated by Grid Technology Research Center (GTRC), AIST, partnering with 49 universities and institutions among 15 economies in the Asia Pacific region, however, it is still below the production level computing environment for supporting neither large scale scientific applications nor real business applications. After performing a relatively long- term run of a large scale application for three months, we have analyzed error causes and have introduced new functions to fix them. This article describes our experience and a process of improving the stability of an international grid.