Database Reference
In-Depth Information
different database models are being swapped and shared. The proliferation of nonre-
lational database technology was a reaction to the limitations of the relational model
for dealing with Web-scale data. Over time, key design concepts from nonrelational
databases have started to appear in traditional relational databases and vice versa. Some
relational databases are becoming easier to run in a distributed setting. Scalable data-
bases are starting to recognize the benefits of providing an SQL interface. An example
is the open-source database RethinkDB. RethinkDB advertises itself as a having “an
intuitive query language, automatically parallelized queries, and simple administra-
tion.” Firebase, a commercial database product optimized for application backends,
claims that it is a “cloud database designed to power real-time, collaborative applica-
tions.” Another example of a product born from the concept of feature mashup is
CitusDB's
SQL on Hadoop
product. CitusDB runs PostgreSQL instances on individual
Hadoop data nodes to enable users to run queries without kicking off costly MapRe-
duce jobs.
Finally, another example comes from the company that published the first paper
about the MapReduce paradigm as well as many early ideas about nonrelational data-
base models. Google's new F1 database, built on top of Spanner and described in a
publicly available paper,
6
is another example of a trend to combine the best of many
worlds. Simply put, F1 is a mostly relational database that attempts to achieve a strong
consistency model using the Spanner storage layer that is distributed geographically
across multiple data centers. This planetary-scale consistency comes at the expense of
the speed that it takes to fully commit data, but techniques in the application layer
of programs that use Spanner help to minimize delays. The query language used in
Spanner is a variant of SQL, which the research paper claims was due to strong user
demand.
The future will result in popular database-technology tools converging toward a
bare minimum of features for continued utility and growth. The end result of this
evolutionary period will hopefully be an era in which the most “good enough” data
solutions become commonplace, reliable, and invisible.
Although data technologies are in the midst of a Cambrian-like period of evolution-
ary feature swap, an interesting corollary is also happening on the human front. In
this topic, a common theme is the tension between the traditional enterprise worlds
of the data warehouse and business analyst versus the world of developer-driven,
MapReduce-based tools such as Hadoop. This clash is often played out between open-
source and proprietary database products. These cultural silos mirror the physical data
silos that thwart gaining the maximum value from organizational data.
Search WWH ::
Custom Search