Choosing Disks for databases on IaaSPublished on: Author: Joanna Schrap Category: Oracle
When it comes to performance and databases, disks come in all sorts of sizes, speeds (iops, throughput, latency, etc.), and redundancy configurations. While a large, fast disk (traditionally Raid 5) is the way to go for most “generic” applications, the best practice for databases has always been to spread the load over smaller, fast disks (traditionally Raid 10). This applies to Oracle as well as SQL Server and, no big surprise, DB2 as well.
Measurement is knowledge
Monitoring is essential in right-sizing the storage. Gather data first, there is no one-size-fits all solution.
For high-performance production loads, this table is a good starting point for assessing your database disk requirements in Azure:
Balancing cost vs performance is key
Should all databases be configured like this? It depends on the performance requirements. For non-production workloads, using standard storage can be sufficient. Many modern disks are fast enough and there are many databases that perform fine on generic configurations.
For databases that are more demanding, DBA’s must work with their admins to ensure that the different file structures (datafiles, archive/transaction log files, temp files, and potentially backup files), land on the correctly configured, separate drives. For very large databases, dba’s must have the ability to spread the data load distribution to separate disks. Striping multiple Azure data disks can be employed to increase I/O bandwidth.
Refer to this checklist of storage best practices.
Learn more about sizing VMs for databases: Azure SQL VM: Get the best price-performance for your SQL Server workloads on Azure VM (Data Exposed).