Identifying trends of I/O response time on cluster components

You can use Unified Manager to view the performance trends for all monitored cluster components for a volume workload. You can see, over time, which components have the highest usage, whether the highest usage is from read or write requests, and how the usage has impacted the workload response time.

Before you begin

About this task

Identifying performance trends for the cluster components helps the administrator decide whether the cluster is being overused or underused.

If you are viewing the Event details page, you can click the name for a volume to go directly to the Performance/Volume Details page.

Steps

  1. In the Search bar, type the name of the volume.
    The name of the volume is displayed in the search results.
  2. Click the name of the volume.
    The volume is displayed on the Performance/Volume Details page.
  3. On the Historic data chart, click 30d to display the last 30 days of historical data.
  4. Click Break down data by.
  5. Under Latency, select Cluster Components and Reads/writes latency.
  6. Click Submit.
    Both charts are displayed below the Latency chart.
  7. Review the Cluster Components chart.
    The chart breaks down the total response time by cluster component. The response time at the aggregate is the highest.
  8. Compare the Cluster Components chart to the Latency chart.
    The Latency chart shows spikes in the total response time that are aligned with the spikes in response time for the aggregate. There are a few at the end of the 30-day time frame, where the performance threshold was crossed.
  9. Review the Reads/writes latency chart.
    The chart shows a higher response time for write requests than read requests, indicating that the client applications are waiting longer than usual to have their write requests fulfilled.
  10. Compare the Reads/writes latency chart to the Latency chart.
    The spikes in total response time that align with the aggregate in the Cluster Components chart also align with the writes in the Reads/writes latency chart. The administrator must decide whether the client applications using the workload must be addressed or whether the aggregate is being overused.