Authors: Benjamin Lynch (University of Minnesota), Douglas Fuller (Red Hat Inc)
Abstract: Ceph is an open-source distributed object store with an associated file system widely used in cloud and distributed computing. In addition, both the object store and file system components are seeing increasing deployments as primary data storage for traditional HPC. Ceph is backed by a robust, worldwide open source community effort with broad participation from major HPC and storage vendors. This BOF session will bring together Ceph implementers to share their deployment experiences, as well as provide feedback to the developer community on needed features and enhancements specific to the HPC community.
Long Description: Most modern research is limited in some way by challenges related to collecting, storing, analyzing, and sharing large and small data sets. To address these challenges, data center managers are increasingly exploring storage platforms that can be deployed to supplement their traditional high performance storage systems. To be effective, a supplemental storage system must be inexpensive, easily scale to support multi Petabytes to Exabytes of data, support a wide range of application interfaces (i.e., object, file, and block), and be highly fault tolerant. The Ceph system satisfies many of these requirements and as a result many multi petabyte installations of Ceph can now be found in academic, government, and commercial environments. Ceph has a wide range of potential use cases (object, block, file, and other semantic storage frontends), deployment configurations, tunable parameters, security options, and other characteristics that can affect its scalability, performance, and fault tolerance.
Ceph is a well established storage technology for virtualized infrastructure. However, it is still an emerging technology in the field of HPC. Recent advances in Ceph including Bluestore, and CephFS remove performance bottlenecks and make it relevant to many HPC applications. Many centers are using Ceph or exploring a Ceph deployment in the near future. This BOF will be an opportunity for HPC staff to discuss and learn about the rapidly changing software and hardware used for Ceph deployments. Recent versions of Ceph include Bluestore, a new storage backend, as well as additional data encryption and many other new features. At the same time, new SSD developments are making tiered storage and all-SSD Ceph deployments an ideal option for some HPC use cases.
This BOF will engage a broad range of Ceph users in the HPC community. Ceph developers and system administrators will share their experience working with Ceph and discuss how it can be used to address many of the challenges that we now associate with Big Data and data intensive research. The conversation will be guided by a series of lightning talks, each be followed by discussion with the audience.
Back to Birds of a Feather Archive Listing