Rethinking Insertions to B<sup>+</sup>-Trees on Coupled CPU-GPU Architectures

Huang, Han; Luan, Hua

doi:10.1109/ispa-bdcloud-socialcom-sustaincom52081.2021.00139

Cited by 1 publication

(1 citation statement)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Finally, the statement is executed, either by the proposed SQL plan (hard parse) or by getting an already existing plan from the Shared pool (soft parse). As a result of the overall processing, the SQL plan is obtained (producing the set of operations covering the data retrieval process via index or direct data block loading [21,22]), followed by the execution process itself. There are two basic Oracle architectures -single and multitenant architecture [23].…”

Section: Database Infrastructurementioning

confidence: 99%

The Complexity of the Data Retrieval Process Using the Proposed Index Extension

Kvet

Papán²

2022

IEEE Access

View full text Add to dashboard Cite

Data retrieval, access, and tuple identification are inevitable in database processing, ensuring performance. The entities and relationships form a relational database. Data themselves are not specifically formatted, requiring sequential data block scanning if the particular index is unavailable. This paper summarizes existing indexing principles focusing on the B+tree, which forms the default structure for data access based on the index key. Such design is reliable and prone to an increase in the amount of data. However, it cannot manage undefined values properly, whereas they cannot be mathematically compared. Secondly, migrated rows can be present due to the size demand extension after the update operation. Thirdly, the index is always balanced, resulting in additional demands of the transaction. All these factors are covered by the proposed paper, discussing the limitations, opportunities, and own solutions to improve the performance. Several architectures are discussed, maintained, and computationally studied, focusing on the size demands, processing time, and costs. By using proposed techniques, significant improvements can be reached. A pointer list is introduced for migrated row management to reference the index set from the data block and store reference path or using a data reflector. When dealing with transaction support, index management and rebalancing are shifted to the separate autonomous transaction. Thanks to that, the main transaction can be approved sooner with no reliability issues. Finally, the proposed paper introduces the NULL value management structure in the instance memory for the index node reference.

show abstract

Section: Database Infrastructurementioning

confidence: 99%