External sorting: I/O analysis and parallel processing techniques
This thesis deals with sorting of data that are much too large to fit in main memory or external sorting. The author focuses on two aspects of external sorting: I/O analysis and parallel processing techniques. Storage device models are defined and applied to analyze the I/O complexities of multi-way merge sort and tag sort (or key sort). It is shown that using higher-merge order, through, reduces the number of merge passes, causes excessive random I/O accesses and degrades the overall I/O performance of multi-way merge sort. Techniques are developed for producing long runs in merge sort and for rearranging the records in tag sort after their ranks are determined. A lower bound for the I/O access time or rearranging the records in tag sort is derived. Two methods are explored for implementing distribution sort on parallel computers. The first method, multi-pass distribution sort, determines the bucket ranges with one read pass over the input file, and uses subsequent passes to distribute the data into buckets and sort them. The distribution and sorting of the buckets are processed in parallel using a two-stage pipeline. The second method, one-pass distribution sort, coalesces the bucket partition, bucket distribution, and sort-bucket phases all together so that the input file needs to be processed only once.
- Research Organization:
- Washington Univ., Seattle (USA)
- OSTI ID:
- 7203989
- Resource Relation:
- Other Information: Thesis (Ph. D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
External parallel sorting with multiprocessor computers
Sorting large files on a backend multiprocessor