Electrical and computer engineering Cache memory Multidimensional databases
We propose a novel organization for multi-dimensional data based on the conceptof macro-voxels. This organization improves computer performance by enhancingspatial and temporal locality. Caching of macro-voxels not only reduces therequired storage space but also leads to an efficient organization of the dataset resulting in faster data access. We have developed a macro-voxel caching theory that predicts the optimal macro-voxel sizes required for minimum cache size and access time. The model also identifies a region of trade-off between time and storage, which can be exploited in making an efficient choice of macro-voxel size for this scheme. Based on the macro-voxel caching model, we have implemented a macro-voxel I/O layer in C, intended to be used as an interface between applications and datasets. It is capable of both scattered access, typical in online applications, and row/column access, typical in batched applications. We integrated this I/O layer in the ALIGN program (online application) which aligns images based on 3D distance maps; this improved access time by a factor of 3 when accessing local disks and a factor of 20 for remote disks. We also applied the macro-voxel caching scheme on SPEC.s Seismic (batched application) benchmark datasets which improved the read process by a factor of 8.
Metrics
36 File views/ downloads
32 Record Views
Details
Title
Optimal caching of large multi-dimensional datasets
Creators
Dinesh Obalappa - DU
Contributors
Liz Haslam (Advisor) - Drexel University (1970-)
Awarding Institution
Drexel University
Degree Awarded
Doctor of Philosophy (Ph.D.)
Publisher
Drexel University; Philadelphia, Pennsylvania
Resource Type
Dissertation
Language
English
Academic Unit
College of Engineering (1970-2026); Electrical (and Computer) Engineering [Historical]; Drexel University