Noxa DWH is provided as a set of the software components that are deployed on shared nothing cluster of commodity 4 socket servers. Logically, the servers grouped in nodes by 3 hardware servers. Each group is called Logical Node and hosts at least 3 replicas of processing applications, cache and relational data sets. Minimal installation consists of 1 logical node, but may be easily scaled up to any number of nodes.
Since Noxa DWH is designed for storage purposes, the vertical expansion strategy is focused on extending disc space of RDBMS table spaces by adding network partition storage nodes. Logical nodes are united by Data Centers and may be geographically distributed.
Each hardware server runs the multi-threaded BOS application and two instances of SQL RDBMS on dedicated CPUs and serves incoming queries. The application implements the business logic and external bindings via ODATA protocol as well as recovery and cache synchronization mechanisms. Optionally, the product may support custom direct load mechanisms for performance optimization. Normally the logical node exposes only two BOS end points for querying, one is always under maintenance. Each RDBMS instance of physical node is configured to work with table spaces stored on external devices and may handle significant amount of tables and partitions.
BOS application is provided as a standard Java servlet and runs under servlet container. It consists of:
Optionally it may be bundled with:
Noxa provides a comprehensive set of tools for monitoring, data mining and problem detection. These tools are installed separately and may be integrated in cloud infrastructure of the customer.
Unlike the traditional relational data management systems such as RDBMS, Noxa distinguishes Master and Transaction data in the data model by design. One of the key values for business is the long-term development and expansion of the model keeping it homogeneous. Since master data management is implemented in the system by itself there is no need to spend money on 3d party MDM solutions. Master data may be modified completely atomically and consistently, and it can be isolated along the whole data landscape. Transaction data must be available for high load and high performance processing, and it can be distributed and organized according to the following requirements: