On Sunday, 22 November 2015 at 01:53:01 UTC, Chris Wright wrote:
There's nothing inherent in having a rigid schema or using SQL
as a query language that prevents scaling.
That's right, except that joins and desirable consistency
requirements creates a ceiling when it comes to scaling. It does
so on NOSQL databases too, but you tend to avoid those there by
making joins "manual labour". A database like BigTable does
encourage two fundamental concepts:
1. eventual consistency, meaning:
- indexes are out of date
- reads based on identity are up to date
2. transactions and consistency on shared ancestors, meaning:
- transactions acts on a hierarchical database structure
(hierarchical data bases have always been more performant than
relational, but the latter is more convenient).
3. massively frequent updates and infrequent queries are handled
by representing a single object as many objects (sharding) (like
voting in an election)
The relational model is flat and flexible, but is resistant to
scaling.
Oracle, in contrast, is nearly forty. A system designed
originally to scale to PDP-11s requires extensive effort and
redesign to scale to scale as well as modern applications
require.
The main challenge is to break relationships with consistency
requirements across compute-nodes (servers).
designed) when we already had algorithms like RAFT and a lot of
experience with distributed systems.
Looks interesting, but I don't see how it helps with the
fundamental model used in RDBMSes?
