HPC clusters are often equipped with node-local non-volatile memories (NVMs), which provide an accumulated peak bandwidth greater than the bandwidth of the backend parallel file systems. Latencies of small accesses to NVM are furthermore much smaller than accesses to backend magnetic disks. Nevertheless, the distributed nature of the NVM devices makes them more difficult to manage than backend file systems or centralized burst buffer instances. Ada-FS tackles this challenge by deploying short-lived file systems on the cluster, which are coupled to individual applications or workflows. The coupling between Ada-FS and a single application enables Ada-FS to simplify the file system design and to overcome several POSIX limitations. This talk will present the design concepts of Ada-FS, its implementation based on Margo and Mercury, as well as performance results.
Prof. Dr.-Ing. André Brinkmann is a full professor at the computer science department of JGU and head of the ZDV (since 2011). He received his Ph.D. in electrical engineering in 2004 from the University of Paderborn and has been an assistant professor in the computer science department of the University of Paderborn from 2008 to 2011. Furthermore, he has been the managing director of the Paderborn Centre for Parallel Computing PC2 during this time frame. His research interests focus on the application of algorithm engineering techniques in the area of data centre management, cloud computing, and storage systems. He has published more than 100 papers in renowned conferences and journals and is an associated editor of the ACM Transactions on Storage as well as a steering committee member of the IEEE Symposium on Massive Storage Systems and Technologies (MSST) and IEEE International Conference on Networking, Architecture, and Storage (NAS). Prof. Brinkmann is a member of the steering committee of the French Grid'5000.