High-Speed Sorter for Database Acceleration

Sorting is an often used and computing intensive kernel in database systems. The amount and the size of keys to be sorted can easily result in datasets of several gigabytes. The keys may represent simple numbers or even data structures with different kind of types.

Sorting is an ideal application to be accelerated by FPGAs (Field Programmable Gate Arrays) that can be adapted to different sorting tasks at runtime. Furthermore, runtime reconfiguration is useful to speed-up different phases during the sorting.

 

For example, in a first phase, sorted subsequences may be generated and merged to longer sorted sequences in one or more succeeding phases. As a sorting phase can only start after the preceding one has finished, it is advantageous to implement dedicated sorters for each phase that use as much resources as possible for executing a sort step as quick as possible. Then, runtime reconfiguration will be used to swap the hardware for each sorting step.
For this master thesis, a flexible architecture for accelerating sorting tasks on FPGAs has to be developed. This architecture will be demonstrated on a prototype board.
 

Prerequisites: INF1400, INF 2270, INF3430/4430
Apportionment: Theory (30%), Concepts (40%), Implementation (30%)
Supervisors: Dirk Koch, Jim Tørresen


 

 

Tags: Simulating Runtime Reconfigurable Systems
Published Mar. 4, 2010 5:38 PM - Last modified Apr. 4, 2011 11:03 AM