Quantile-selection based data partition for parallel joins in hypercube database computers = 병렬 결합 연산에서 변량표본값 선택을 이용한 데이타 분할기법

Cited 0 time in webofscience Cited 0 time in scopus
  • Hit : 276
  • Download : 0
As performance becomes a crutical issue in large database applications, parallel database machines are receiving more attention in both theory and practice. In a ralational database management system running on such a machinem database operations are divided into a set of relational operations which are executed on a collection of processors in parallel. Among such operations join operations are the most complex and time consuming. These operations will limit the performance of such a database system. The effectiveness of parallel execution under join operations will largely depend on data distributions among processors in such a database machine. Several parallel join algorithms are proposed for parallel relational databases systems, and the hash-partitioned parallel join algorithm (HPJA) is the most commonly used. It should be clear that for good parallelism the number of tuples mapped to each processor should be approximately equal. Most of the conventional parallel join algorithms have assumed uniform distribution of data. Therefore all algorithms including HPJA may limit the degree of parallelism in execution as degrees of replication in the join attribute values become large. For efficient parallel joins a hash function must map tuples with equal join attribute values to the same processor. However, there is no such a hash function that can avoid the skewness of data distribution resulted from these replicated values. The problem is that such skewness is inevitable even if an dieal hash function is used. Therefore if the data distribution is heavily skewed, load imbalances are caused by a skewed node. This will lose any of the advantages due to parallelism, andperformance will be degraded. This thesis proposes a new partitioning method to resolve the problem of data skewness for parallel join operations in a hypercube database system. Based on quantile selection, the algorithm makes data distribution to be approximately even among node processors in the h...
Kim, Tag-Gonresearcher김탁곤researcher
한국과학기술원 : 전기및전자공학과,
Issue Date
101711/325007 / 000835533

학위논문(박사) - 한국과학기술원 : 전기및전자공학과, 1995.2, [ ix, 109 p. ]


Partition; Selection; Hypercube; Parallel join; Quantile elements

Appears in Collection
Files in This Item
There are no files associated with this item.


  • mendeley


rss_1.0 rss_2.0 atom_1.0