Content area
Full text
Abstract - (PDPTA ?4) This research presents the design and initial implementation of a database engine using an Intel Xeon Phi co-processor. The many integrated cores (MIC) of the Xeon Phi make this hardware accelerator a natural computing platform for an in-memory database engine or server. The database tables reside in the memory space of the MIC thus supporting fast in-memory database applications. This achieved by developing a coalescing parallel memory manager to allocate parallel variables in the same manner that fields are created in a table using a SQL CREATE TABLE command. The SQL interface was created using a database driver toolkit that provides an interface to the Xeon Phi server and client application. Once the basic framework was established, the algorithms for SQL select, insert, update, delete, and join were created to manipulate database information in the memory of the Xeon Phi.
Keywords: parallel databases, parallel hardware accelerators, special purpose architectures
1 Introduction
Massively data parallel computers and the SIMD model of parallel computation can be a natural model of parallel computation to consider for massively parallel database servers. Since the cores are extremely close to the parallel memory, fast parallel memory searching make it a natural platform for data parallel computing intensive applications. As described in Potter in [9] and [10], data parallel models of computation such as the SIMD, ASC, or SITDAC model conform to the concept of a parallel database server since the data can logically and physically partitioned similar to the data organization of a database table or spreadsheet [2][8][9][10][11] and [12].
This research paper discusses the initial design of an in-memory database server using a Intel Xeon Phi coprocessors. This research will discuss the design considerations and challenges for a database server and SQL engine that interfaces with the memory of a hardware accelerated data parallel computer. This system design can promote the use of massively parallel computers as database servers for use in embedded database systems, real-time database systems, and fast parallel associative search
Database management systems (DBMS) provide a structured mechanism for storing, organizing, and retrieving data in a way that is consistent with the database's format [14]. System software will allow data storage and access to a database without the user's knowledge about...