SSDs have blurred the line between RAM and spinning-platter drives, shrinking the gap between CPU and storage performance tremendously. Bottlenecks that were once the mainstay of the disk drive are now being distributed to other parts of the system.
Relational database vendors have been quick to use such devices to tout increased throughput records, but just as the jump from spinning disk to RAM yields only around a 10x database performance improvement, switching to SSD does not address the inherent problems in the relational model.Next »
While the storage mechanism may change, the underlying architecture of the database remains optimized for the rotating platter mechanisms that dominated the 1970s and 80s. The need for contiguous blocks of row-centric storage, for example, is still a requirement and means SSDs suffers from the sparse matrix problem, requiring de-fragmentation.
Similarly, maintaining redundant information (as foreign keys) places a great strain on SSD devices as they need to perform more writes than necessary. Factor in re-indexing efforts, and all of a sudden the lifetime of your (comparatively expensive) SSD is not as impressive, even with advanced wear-levelling algorithms.Next » « Previous
We’re not saying SSDs are bad: far from it. If budgets allow, we recommend using them over traditional hard drives. But their cost and life expectancy should be considered as part of an overall plan to utilise a more efficient data model which can place lighter demands on SSD devices. Replacing worn out drives is expensive.
With its compact list-based structure, elimination of duplicate data, no need for indexing or de-fragmenting, and true random, non-contiguous storage requirements, Ancelus can leverage every last ounce of performance and longevity from solid state devices.« Previous
Did you know...
The performance of Ancelus is not constrained by the volume of data in your database.