An Introduction to Oracle Fusion – a Solid Foundation for Service Inventory - Applied SOA Patterns on the Oracle Platform - page 121

Oracle DB, Framework 6, and accessed via a unified inventory's endpoint. It is worth
mentioning that the rule engine SDK is available for the DB, Java, and PL/SQL.
We have mentioned RDBMS several times as you may have noticed, and we did it on pur-
pose. Yes, XDB with XML-relational model mapping is crucial to handle well-structured,
query-intensive data such as Order, Invoice, Client, and so on. It is rather hard to justify
why intensively queried data elements should be stored in CLOB or XMLType fields. But
what if all we need is to pertain the object-oriented data and pass it further on for client-
side processing? Yes, that's an AJAX with the JSON payload type of processing. Never-
theless, the payload data could also be the Lists, Sets/Ordered Sets, Hash maps, primit-
ives, and so on. All types of message-object-related mappings are quite expensive and
could be complex; moreover, we don't always need them on the Message Oriented Mid-
dleware ( MOM ) side.
Oracle offers good additions (we wouldn't describe it as a pure alternative) to RDBMS in
the form of Oracle Big Data Appliance (optimized for Oracle Exadata Database Machine,
but it can run on any hardware). This includes an open source distribution of Apache Ha-
doop ( http://hortonworks.com/hadoop/ ) , the Oracle NoSQL Database (frontend for well-
known Berkley DB), Oracle Data Integrator Application Adapter for Hadoop, and Oracle
Loader for Hadoop.
The last two components implement a MapReduce-distributed computing pattern ( ht-
tp://highlyscalable.wordpress.com/2012/02/01/mapreduce-patterns/ ) , linking Hadoop Big
Data realized as Hadoop Data File System ( HDFS ) to Oracle's external tables, which are
transparent for regular SQL queries. The pattern's name MapReduce could be misleading
as the object-relational mapping is not the primary goal of this solution. In this pattern, we
try to achieve the highest level of processing parallelism for clustered multitenant data in
HDFS. We are literally mapping chunks of data with unique numbers (key-value pairs),
where the key is associated with processing nodes and responsible for the reducing of the
overall workload by the parallel processing of related data chunks. Processing could be
online or offline and multiconsumer AQs can be used quite extensively by job controllers
in some pattern's realizations.
So, we could potentially have the classic RDBMS and NoSQL key-value distributed
stores at the same time. All these types of data access, loading, and distribution put seri-
ous requirements on the data access layer, technically residing in the App Server layer (the
next layer in the technical infrastructures hierarchy). Not only will components of the
mentioned MapReduce pattern be implemented there, but practically the whole data lay-
er must be abstracted in a way that data manipulation routines do not affect the applica-