Giant Oracle RAC Databases Efficiency on MPP Machines

Not like a cluster the place just a few nodes are interconnected, massively parallel processing (MPP) techniques enable a lot of nodes to be interconnected. There are MPP techniques with multiple thousand nodes. Every these nodes is a separate system having its personal CPU, disks, controllers, reminiscence, and inside system buses to type a “loosely coupled”, share-nothing structure. All of those nodes are linked by way of a high-speed, high-bandwidth interconnect. Every node has separate copy of the working system. In MPP structure, Oracle is put in in RAC mode.

Every Oracle occasion on every node is answerable for all sources it holds, has a view of the whole database, and may discover out which node holds a lock on any a part of the database. If it wants one thing that’s locked by one other node, it is going to require an inter-instance ping, the place the opposite occasion must write to disk all of the adjustments it has made. A typical database question would go in opposition to one node; information can be picked up from the reminiscence and or disks of any applicable node and journey by way of interconnect to return the information to the requesting node. Every node controls its personal set of disks and may take over management of one other set if a node fails. Thus, all of the nodes could also be configured to be main or a mix of main and secondary. The database information are positioned on the first nodes, whereas the secondary nodes present the required redundancy to take over and substitute the first nodes, in case of the latter fail.

The most important drawback in MPP architectures is just not the structure itself, however software design flaws that exist in a typical implementation. Most software designers are effectively versed in implementing functions on SMP machines. Nonetheless, MPP architectures require a complete paradigm shift when it comes to analyzing which information must be positioned on which node to cut back information sharing and inter-instance ping throughout nodes. Since every node is extremely impartial in a share-nothing structure, any scenario that causes nodes to commerce large-scale sources attributable to information sharing ends in excessive visitors throughout the interconnect. This example is inflicting heavy efficiency issues in Oracle RAC databases. Efficiency of Oracle RAC database is usually relying on unsuitable {hardware} structure configuration. It is best to want to contemplate {hardware} structure and deployment of working system earlier than processing for Oracle Actual Software Cluster configuration. In any other case, you’d get numerous issues after deployment and altering {hardware} configuration is requiring downtime of your excessive out there databases.

Leave a Comment