Skip to end of metadata
Go to start of metadata

Reports

In Q3 and Q4 of 2010, we have been investigating the suitability of the Graphics Processing Units (GPUs) technology for the acceleration of image characterization algorithms used to find similarities between documents with embedded images. In Q3 we have ported image characterization algorithm used in doc2learn to GPUs using both CUDA C targeting NVIDIA GPUs and OpenCL? targeting NVIDIA and AMD architectures and conducted an extensive study of the impact of GPU acceleration for documents with varying number of images and image sizes. 

  • Q3 report: V. Kindratenko, G. Shi, Evaluation and Exploration of Next Generation Systems for Applicability and Performance, Technical report 1, October 2010 (report, presentation, software)

In Q4, we have re-designed in C image extraction framework used in doc2learn and conducted an extensive study of the performance of various components of this framework, such as file I/O, image extraction from PDF files, and image analysis with the goal to provide feedback to the doc2learn team about optimization techniques that can be applied in doc2learn to improve its performance. We also have conducted a study to characterize the impact of using GPUs on power consumption. 

  • Q4 report: V. Kindratenko, G. Shi, Evaluation and Exploration of Next Generation Systems for Applicability and Performance, Technical report 2, January 2011
    (report, presentation)

In Q1 of 2011, we investigated the suitability of GPU technology for two important classes of data-intensive applications: computation of checksums used for data integrity verification, encryption, and data comparison, and lossless data compression. Our findings indicate that while these algorithms can benefit from GPU acceleration to some degree, their practical use on GPUs is limited due to the PCIe bandwidth bottleneck between the host and the GPU. We also integrated GPU implementation of Scale-Invariant Feature Transformation (SIFT) algorithm with doc2learn and Versus software and demonstrated its performance for image comparison within these two frameworks.

  • Q1 report: V. Kindratenko, G. Shi, Evaluation and Exploration of Next Generation Systems for Applicability and Performance, Technical report 3, April 2011 (report, presentation)

In Q2 of 2011, we conducted XtremeData dbx data analytics appliance evaluation. We performed XtremeData dbx data analytics appliance evaluation. We used a database supplied by NARA consisting of a collection of approximately 79 million records. We also used a database with randomly generated records, ranging from 1 million to 1 billion records. We measured database deployment time as well as the time to run complex queries involving joining tables. We compared our measurements with the results supplied by NARA.

  • Q2 report: V. Kindratenko, G. Shi, Evaluation and Exploration of Next Generation Systems for Applicability and Performance, Technical report 4, July 2011 (report, presentation)

In Q3 of 2001, ISL team worked on two projects: development of the software infrastructure for power monitoring on GPU HPC clusters and a reference design and implementation of pattern matching algorithm.

  • Q3 report: V. Kindratenko, G. Shi, Evaluation and Exploration of Next Generation Systems for Applicability and Performance, Technical report 5, October 2011 (report, presentation, software)

Software

  • No labels