In the world of massively parallel processing (MPP) architectures for analytics databases, there have been two predominant approaches: “shared nothing”, and “shared everything”. In a shared nothing environment, each server operates independently, and controls its own memory and disk resources. Data is partitioned among the servers, and the workload is distributed such that each machine operates on its own data, without sharing hardware resources with other machines in the grid. In a shared everything environment, all servers access the same shared store, and each workload has access to the store, as well as the computing resources of all servers in the grid. Shared nothing focuses on maximizing performance, and shared everything focuses on maximizing resource utilization.
The downside of shared nothing is the requirement to duplicate shared data across nodes, and to keep data evenly partitioned across nodes for consistently high performance. This maintenance overhead has been mitigated by new data sharding techniques, which automatically repartition data without user intervention. Shared everything has been criticized for reduced performance due to contention over the shared store. This has been mitigated by high performing SANs, which can serve up data to many servers across multiple high speed I/O channels.
In the interest of creating the most flexible, efficient, and easy to maintain analytics platform in the market, SAP Sybase IQ adopted a shared everything approach with its MPP architecture, called PlexQ. The PlexQ distributed query platform dynamically utilizes physical compute resources as it manages query workloads across a Multiplex grid of IQ servers. In addition, the user can define logical servers to create virtual data marts that segment analytics workloads across an enterprise community. Logical servers are subsets of physical machines in the Multiplex grid, which can be elastically configured – physical machines moved from one logical server to another- based on changing demands for compute power. IQ’s unique approach to MPP achieves maximum flexibility, as well as scalable performance.
SAP’s latest release of IQ – version 16 – takes MPP to a whole new level with its innovative soft data affinity feature. Now, the query optimizer intelligently tracks cache content across the grid, distributing each query fragment to the server that already has the pertinent data in cache. Caches stay hot, I/O is dramatically reduced, and your shared everything architecture has now taken on the high performance characteristics of shared nothing. Data affinity is managed automatically by the IQ query optimizer, and self-adjusts with changing data and ad hoc queries:
SAP Sybase IQ 16 brings you the best of both worlds – the agility of shared everything, with the performance of shared nothing. For XLDB analytics, SAP Sybase IQ stands on its own. To learn more, check out the announcement on CMS Wire: http://www.cmswire.com/cms/information-management/saps-sybase-iq-16-goes-extreme-delivering-data-driven-insights-in-record-time-019704.php
Courtney Claussen | SAP Sybase IQ Product Management | Email: firstname.lastname@example.org