In-memory databases

Recently I came across an announcement of the first enterprise-class in-memory database - SQL Server 2017. The article declares some very impressive performance benchmarks and I strongly recommend you have a look at them.

The in-memory database technology can be a game changer, especially when the technology will become a commodity. At present the IO is the biggest bottleneck of the databases because most relational database systems require transactional consistency of writes. This means that the transaction logs have to be written on permanent storage. Despite the tremendous progress that was achieved thanks to SSD technology, the write-ahead paradigm was throttling the development of relational (and not only!) databases.

With in-memory database technology, you can change the transaction logging approach without compromising the safety of the data and the recoverability of the system. The persistence of data stored in NVDIMM memory means you no longer need to log to the SSD - you can log to a virtual disk in the NV (non-volatile) memory, and this change on its own brings the promise of speed gains never imaginable before!