Building reliable Ceph clusters

Lars Marowsky-Brée http://linux.conf.au/schedule/present... Ceph is the most popular Software-Defined-Storage technology today, and extremely widespread in the OpenStack world. It enables a multitude of block, object, and file storage use cases, and its flexibility allows it to be configured as required for many different scenarios. The hardware environment can be similarly tailored. This is a sizable decision matrix, but leads to an environment optimally tuned for the required balance between performance, functionality, and cost. Dependability aspects - availability and reliability in particular - are often overlooked. Based on twenty years of development and hands-on experience with designing dependable distributed systems and supporting them in production, the goal of this presentation is to make you confident in your choice of Ceph for your use case, and to build an architecture you can trust. Beginning with choosing the appropriate access method for your workload, we then continue to introduce the algorithms and technologies in Ceph as they relate to resilience and High-Availability. We will discuss the considerations involved in optimizing a distributed storage system for reliability, availability, and durability of data and fault tolerance. We will look at the performance of Ceph in degraded and recovery scenarios, and how to reduce exposure. This affects the choice of hardware, the approach to feature selection, and system architecture. We will also talk about operational procedures to reduce unplanned downtime, speed-up recovery, and improve supportability.