Lessons learned from development and operation of the K computer

We report operational experiences of the K computer which is one of the most powerful supercomputers in the world. The K computer achieved excellent results for system availability, job-filling rate and failure rate. On the other hand, approximately 70% of the unscheduled system stop time was caused by file system failures. We analyzed the reasons for the failures and found that a massive and complex system configuration of the file system is one of the crucial factors for the failures. It revealed many potential bugs in the file system software, and such bugs caused many failures which gave severe impacts to the operation.