Monitoring EBS Metrics for Performance and More

​

CloudWatch Metrics for EBS Volumes

• Volume Read Operations:
  Indicates the total number of read operations performed on a volume. It helps gauge the volume’s read activity.
• Volume Write Operations:
  Tracks all the write operations executed on the volume, offering insight into the volume’s write workload.
• Volume Bytes:
  Divided into read bytes and write bytes, this metric shows the total amount of data read from or written to the volume during a given interval.
• Volume Queue Length (or Queue Depth):
  Represents the number of I/O requests waiting in the queue. For example, if your system handles 500 read operations per second but receives 1,000, a backlog of 500 requests will occur. A high queue length generally signals that the disk is under heavy load.
• Provisioned IOPS Throughput Percentage:
  For provisioned IOPS volumes, this metric displays the percentage of I/O operations processed. Lower percentages can indicate performance limitations.
• Burst Credits (for GP2 Volumes):
  GP2 volumes, which are older than GP3, use burstable I/O credits. Monitoring burst balance is similar to tracking CPU burst on T2 or T3 systems.
• Volume Idle Time:
  Measures the duration when no operations are requested. The impact of idle time varies; for example, 60 seconds of inactivity might be concerning in a short monitoring window, whereas 10 minutes could be normal during low-usage periods.
• Volume Total Read Time and Total Write Time:
  These metrics show the cumulative time spent on read and write operations, which helps determine latency (the delay between requesting and completing an I/O operation).
• Volume Consumed Read/Write Operations:
  For provisioned IOPS volumes, this metric tracks the total I/O consumed by read and write activities.

​

Interpreting Key Performance Indicators (KPIs)

• IOPS (I/O Operations Per Second):
  Measures the number of read or write operations processed per second.
• Latency:
  Represents the delay of each I/O operation, calculated from total read or write time. Higher latency translates to slower performance.
• Queue Length:
  Also known as queue depth, an increasing number of pending requests indicates that the volume cannot handle the operations quickly enough.
• Burst Balance:
  Especially important for GP2 volumes, burst balance provides insight into the available burst capacity. In some GP3 scenarios, this metric can also be monitored.

​

Configuring CloudWatch Alarms

​

Summary

• Read and write operations (including the number of bytes transferred)
• Total operational times for read and write requests
• Queue length (queue depth)
• Burst balance (notably for GP2 volumes)
• IOPS and latency