Parallelization becomes more and more important, even for the architecture of single machines. Recent advances in processor technologies achieve only small performance improvements for single cores. Increasing the compute power of modern architectures mandates to increase the number of compute cores on a single central processing unit (CPU). Graphics Processing Units(GPUs) have a long history of scale-out through parallel processing on many compute cores. Graphics adapters nowadays offer a highly parallel execution environment that within the context of GPGPU (General purpose Processing in Graphics Processing Units) is frequently used in scientific computing. The challenge of GPGPU programming is to design applications for the SIMD architecture (Single Instruction, Multiple Data) of graphics adapters that allow only for a limited range of operators and very limited synchronization mechanisms.
In the course of the SINDPAD project, we will develop an indexing and search technology for structured data sets. We will leverage graphics adapters to support query execution. SindPad aims at achieving unprecedented performance compared to conventional systems of equal cost. We consider taking advantage of application characteristics to accelerate data processing. Especially for Business Intelligence (BI) applications, the schema enables the system to store specific data on graphics adapters. This can lead to further speed ups.
Researchers of the Database Systems and Information Management (DIMA) group at the TU Berlin will play a significant role in the conceptual planning and implementation of algorithms for hybrid GPU/CPU processing. We will analyze query processing algorithms and devise metrics to compare the performance of GPU-operators and CPU-operators .The SINDPAD project is funded by the German Federal Ministry of Economics and Technology and is carried out in cooperation with empulse GmbH.
Projektlaufzeit: 06/2010 - 05/2012