Identifying the number of AIO servers

You can set the value for the maximum number of servers in a number of ways.

• Limit the maximum number of servers to a number equal to ten times the number of disks that are to be used concurrently.
• Set the maximum number of servers to a high number (for example, 100), depending on the system and the number of processors.
  • Change the server_inactivity tunable value to 600 (the default value is 300).
  • Monitor the number of servers every 10 minutes (600 seconds) throughout the course of a normal workload. If the number of servers is constantly at the aggregate value for the maxservers tunable (the maxservers tunable value times the number of processors) and there is unused processor and IO bandwidth, consider increasing the maxservers tunable value. If you find that at times the system is not performing as expected, consider lowering the maxservers tunable value.

  The goal is to find the performance balance between processor usage and IO bandwidth.

• Take statistics using vmstat -s before any high I/O activity begins, and again at the end. Check the field iodone. From this you can determine how many physical I/Os are being handled in a given wall clock period. Then increase the maximum number of servers and see if you can get more activity or event completions (iodones) in the same time period.

  In general, consider changing the minservers tunable value only when an application will be issuing a high number of I/Os that is beyond the server_inactivity timeout value (thus causing a high number of AIOS kprocs to be created in bursts). To help smooth this condition, either increase the minservers value to keep the required number of AIOS kprocs active, or increase the server_inactivity tunable value so that the system will naturally keep them active if the application keeps issuing requests within the timeout window.