Authors:
Joubert de Castro Lima
1
and
Celso Massaki Hirata
2
Affiliations:
1
Federal University of Ouro Preto (UFOP), Brazil
;
2
Instituto Tecnológico de Aeronáutica (ITA), Brazil
Keyword(s):
Cube Computation, Parallel Cube Computation, Data Warehous, OLAP.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Business Analytics
;
Data Engineering
;
Data Mining
;
Data Warehouses and OLAP
;
Databases and Information Systems Integration
;
Datamining
;
Enterprise Information Systems
;
Health Information Systems
;
Sensor Networks
;
Signal Processing
;
Soft Computing
Abstract:
In this paper, we present a novel parallel full cube computation approach, named p-MDAG. The p-MDAG approach is a parallel version of MDAG sequential approach. The sequential MDAG approach outperforms the classic Star approach in dense, skewed and sparse scenarios. In general, the sequential MDAG approach is 25-35% faster than Star, consuming, on average, 50% less memory to represent the same data cube. The p-MDAG approach improves the runtime while keeping the low memory consumption; it uses an attribute-based data cube decomposition strategy which combines both task and data parallelism. The p-MDAG approach uses the dimensions attribute values to partition the data cube. It also redesigns the MDAG sequential algorithms to run in parallel. The p-MDAG approach provides both good load balance and similar sequential memory consumption. Its logical design can be implemented in shared-memory, distributed-memory and hybrid architectures with minimal adaptation.