Performance management with ONTAP QoS on ASA r2 systems
Suggest changes
PDFs
Safely and efficiently managing multiple Oracle databases on ASA r2 requires an effective QoS strategy. This is especially important because ASA r2 systems are all-flash SAN platforms designed for extremely high performance and workload consolidation.
A relatively small number of SSDs can saturate even the most powerful controllers, so QoS controls are essential to ensure predictable performance across multiple workloads.
As a reference, ASA r2 systems such as the ASA A1K or A90 can deliver hundreds of thousands to over a million IOPS with sub-millisecond latency. Very few single workloads would consume this level of performance, so full utilization typically involves hosting multiple databases or applications. Doing this safely requires QoS policies to prevent resource contention.
ONTAP QoS on ASA r2 works the same way as on AFF/FAS systems, with two primary types of controls: IOPS and bandwidth. QoS controls can be applied to SVMs and LUNs.
IOPS QoS
IOPS-based QoS limits the total IOPS for a given resource. In ASA r2, QoS policies can be applied at the SVM level and to individual storage objects such as LUNs. When a workload reaches its IOPS limit, additional I/O requests queue for tokens, which introduces latency. This is expected behavior and prevents any single workload from monopolizing system resources.
|
Be cautious when applying QoS controls to database transaction/redo log data. These workloads are bursty, and a QoS limit that seems reasonable for average activity may be too low for peak bursts, causing severe performance issues. In general, redo and archive logging should not be limited by QoS. |
Bandwidth QoS
Bandwidth-based QoS limits throughput in Mbps. This is useful when workloads perform large block reads or writes, such as full table scans or backup operations, which consume significant bandwidth but relatively few IOPS. Combining IOPS and bandwidth limits can provide more granular control.
Minimum/guaranteed QoS
Minimum QoS policies reserve performance for critical workloads. For example, in a mixed environment with production and development databases, apply maximum QoS to development workloads and minimum QoS to production workloads to ensure predictable performance.
Adaptive QoS
Adaptive QoS adjusts limits based on the size of the storage object. While rarely used for databases (because size does not correlate to performance needs), it can be useful for virtualization workloads where performance requirements scale with capacity.