Analysis of parallel scan processing in Shared Disk database systems

Abstract

Shared Disk database systems offer a high flexibility for parallel transaction and query processing. This is because each node can process any transaction, query or subquery because it has access to the entire database. Compared to Shared Nothing database systems, this is particularly advantageous for scan queries for which the degree of intra-query parallelism as well as the scan processors themselves can dynamically be chosen. On the other hand, there is the danger of disk contention between subqueries, in particular for index scans. We present a detailed simulation study to analyze the effectiveness of parallel scan processing in Shared Disk database systems. In particular, we investigate the relationship between the degree of declustering and the degree of scan parallelism for relation scans, clustered index scans, and non-clustered index scans. Furthermore, we study the usefulness of disk caches and prefetching for limiting disk contention. Finally, we show that disk contention in multi-user mode can be limited for Shared Disk database systems by dynamically choosing the degree of scan parallelism.

Dynamic query scheduling in parallel data warehouses

“…For general PDBS, load balancing problems have been widely researched, for a variety of workloads and architectures [4,6,9,10,16]. Many of these approaches rely on extensive data redistribution too costly in a large data warehouse.…”

“…Conversely, load distribution on disks has largely been considered in isolation from CPU-side processing. Most of these studies have focused either on data partitioning and allocation [7,15,17] or on limiting disk contention through reduced parallelism [16]. Integrated load balancing as proposed in this paper has not been addressed.…”

“…The Shared Disk architecture has been advocated due to its superior load balancing potential especially for read-only workloads as in data warehouses [9,12,16]. It also offers great freedom in data allocation [15].…”

“…To this end, we exploit the flexibility of the Shared Disk (SD) architecture [16] in which each processing node can execute any subquery. For scan workloads, the balance of CPU load does not depend on the data allocation, permitting query scheduling with shared job queues for all nodes.…”

Analytische Bestimmung einer Datenallokation für Parallele Data Warehouses

“…Durch eine fragmentbezogene Allokations-und Verarbeitungsstrategie unterstützen wir gleichzeitig die Ausnutzung von sowohl Inter-und Intra-Queryals auch I/O-Parallelität. Wir beziehen uns insbesondere bei der Speicherung von Fragmenten auf Platten auf den Einsatz einer Shared-Disk-(SD-)/Shared-Everything-(SE-) Architektur, weil diese durch die Möglichkeit des Zugriffs von allen Verarbeitungsknoten auf alle Platten eine höhere Allokationsflexibilität bei geringeren Kommunikationskosten im Vergleich zu Shared-Nothing-(SN-)Systemen ermöglichen [23,25,26,30]. Unser Ansatz kann allerdings mit geringem Aufwand auch auf eine SN-Umgebung adaptiert werden.…”