Analysis of Dynamic Load Balancing Strategies for Parallel Shared Nothing Database Systems

Paralleldatabasesystemshaveto supportboth inter-transaction as well as intra-transaction parallelism. Inter-uansaction parahelism (multi-user mode) is required to achieve high throughput, in particular for OLTP transactions,and sufficient cost-effectiveness.Irma-transactionparallelism is aprerequisitefor reducingtheresponsetimeof complexand data-intensive transactions (queries). In order to achieve bothgoalsdynamicstrategiesfor loadbalancingandscheduling arenecessarywhich takethecurrent systemstateinto accountfor allocating transactionsandsubqueriestoprocessorsandfor determining thedegreeof intra-transactionparallelism. We study the load balancing problem for parallel join processing in Shared Nothing databasesystems.In thesesystems,join processingis typically basedon a dynamic redistribution of relations to join processorsthus making dynamic load balancing strategiesfeasible.In particular, westudytheperformanceof dynamicloadbalancing strategiesfor determiningthenumberof join processo...

• Type: Conference
• Year: 1993
• Where: VLDB
• Authors: Erhard Rahm, Robert Marek
• Keywords: Database, Join Processing, Parallel Join, Parallel Join Processing, VLDB 1993