MPP is defined as a large number of processors (separate Server / computers) to perform a set of coordinated computations in parallel. This is based on parallel programming model for developing high-performance embedded system applications.
An MPP system is also known as a "shared nothing” system. It uses an "interconnect" arrangement of data paths allows messages to be sent between processors, with each processor using its own operating system and memory and storage.
There are various Products in market like SQL server 2012 PDW, Green plum and many more. Architecture for all these products is almost same the only difference is terminology.
In MPP every CPU is having its own storage. And data in MPP is distributed (not shared) among storage. So if one have to access data every processer is responsible for data available on its own storage only.
Introduction of MPP capability takes data warehousing to new levels in Microsoft SQL Server 2008R2/ 2012 PDW. In simple terms, MPP extends the data capacity of SQL Server by distributing the data across multiple "nodes," which are distributed at distinct instances of SQL Server. However, to the end-user, the MPP appears as one database.
MPP model presents a comprehensive, fault-tolerant, less-expensive environment for VLDW applications making it an attractive augmentation to their implementations.
Recommended: What Is The Difference Between SMP and MPP