Time: 15:00 - 15:50
Room: Alfa 1
In my experience of PostgreSQL consultant I could meet several customers asking for an objective evaluation about basing PostgreSQL on NFS-based storage: many articles have been written on this, but when you start to work on this "in practice", you can discover a lot of things that are not yet documented.
NFS could provide really high performance, but it is generally strongly impacted when you require to have a minimum of reliability.
A possible solution is to base the database on two distinct NFS partitions: one dedicated to the data and the other to the xlog, in order to ensure enough reliability during WALs persistence and enough performance during data access. But silent corruption of data pages are still possible. Starting from PostgreSQL 9.3, it is possible to require the checksum of data pages, and full page images recorded in WAL records are protected by a CRC-32 check: this has enhanced the reliability of PostgreSQL against many kinds of data corruption due to storage failures, but it has to be proved on each specific infrastructure, and the impact on the performances has to be evaluated.
In this presentation I would like to talk about my personal experience on working with PostgreSQL based on NFS-storages, trying to bare "myths and truths" about this union basing on benchmark results and "simulated" crash tests, discussing about the (real or not) possibility of implementation of this kind of infrastructure, both in terms of reliability and achievable performance.