Selecting a Volume Layout

VERITAS Volume Manager offers a variety of layouts that allow you to configure your database to meet performance and availability requirements. The proper selection of volume layouts provides optimal performance for the database workload.

An important factor in database performance is the tablespace placement on the disks. Disk I/O is one of the most important determining factors of your database's performance. Having a balanced I/O load usually means optimal performance. Designing a disk layout for the database objects to achieve balanced I/O is a crucial step in configuring a database.

When deciding where to place tablespaces, it is often difficult to anticipate future usage patterns. VxVM provides flexibility in configuring storage for the initial database set up and for continual database performance improvement as needs change. VxVM can split volumes across multiple drives to provide a finer level of granularity in data placement. By using striped volumes, I/O can be balanced across multiple disk drives. For most databases, ensuring that different database files and tablespaces are distributed across the available disks may be sufficient.

Striping also helps sequential table scan performance. When a table is striped across multiple devices, a high transfer bandwidth can be achieved by setting the Oracle parameter DB_FILE_MULTIBLOCK_READ_COUNT to a multiple of full stripe size divided by DB_BLOCK_SIZE. See Tuning for Performance for more information.

Choosing Appropriate Stripe Unit Sizes

When creating a striped volume, you need to decide the number of columns to form a striped volume and the stripe unit size. You also need to decide how to stripe the volume. You may stripe a volume across multiple disk drives on the same controller or across multiple disks on multiple controllers. By striping across multiple controllers, disk I/O can be balanced across multiple I/O channels. The decision is based on the disk and controller bandwidth and the database workload. In general, for most OLTP databases, use the default stripe unit size of 64 K or smaller for striped volumes and 16 K for RAID-5 volumes.

Choosing Between Mirroring and RAID-5

VxVM provides two volume configuration strategies for data redundancy: mirroring and RAID-5. Both strategies allow continuous access to data in the event of disk failure. For most database configurations, we recommend using mirrored, striped volumes. If hardware cost is a significant concern, but having higher data availability is still important, use RAID-5 volumes.

RAID-5 configurations have certain performance implications you must consider. Writes to RAID-5 volumes require parity-bit recalculation, which adds significant I/O and CPU overhead. This overhead can cause considerable performance penalties in online transaction processing (OLTP) workloads. If the database has a high read ratio, however, RAID-5 performance is similar to that of a striped volume.

Volume Configuration Guidelines

Follow these guidelines when selecting volume layouts:

• redo logs - Put the redo logs on a file system created on a striped and mirrored (RAID-0+1) volume separate from the user tablespaces or data files. Stripe multiple devices to create larger volumes if needed. Use mirroring to improve reliability. Do not use VxVM RAID-5 for redo logs.
• When normal system availability is acceptable, put the tablespaces on file systems created on striped volumes for most OLTP workloads.
• For most workloads, use the default 64 K stripe-unit size for striped volumes and 16 K for RAID-5 volumes.
• When system availability is critical, use mirroring for most write-intensive OLTP workloads. Turn on Dirty Region Logging (DRL) to allow fast volume resynchronization in the event of a system crash.
• When system availability is critical, use RAID-5 for read-intensive OLTP workloads to improve database performance and availability. Use RAID-5 logs to allow fast volume resynchronization in the event of a system crash.
• For most decision support system (DSS) workloads, where sequential scans are common, experiment with different striping strategies and stripe-unit sizes. Put the most frequently accessed tables or tables that are accessed together on separate striped volumes to improve the bandwidth of data transfer.