Last Modified: Sep 02, 2022
Opened: May 26, 2021
Systems with a large amount of statistics data collected in the local database (i.e., systems not working with BIG-IQ) can have high IO Wait CPU usage, peaking at 10 minutes, 1 hour, and 24 hours. This is caused by the data aggregation process that is running on the local database. Large memory footprints, particularly for avrd might be a symptom for the phenomenon.
High IO can impact various processes on BIG-IP systems. Some of them can experience timeouts and might restart.
-- The BIG-IP system is collecting statistics locally (i.e., not sending data to BIG-IQ or another external device). -- There is a large amount of statistics data. -- May occur even if AVR is not explicitly provisioned.
The most effective workaround is to lower the amount of data collected by setting the 'avr.stats.internal.maxentitiespertable' DB variable to a lower value. The recommended values are 20000 (on larger, more powerful systems with more than 16 cores) or 2148 (on smaller systems). Note: After you lower the database value, continue to monitor the BIG-IP system for long I/O wait times and high CPU usage. If symptoms persist and the system continues to experience resource issues, you may need to reset the BIG-IP AVR statistics. For information about resetting BIG-IP AVR statistics, refer to K14956: Resetting BIG-IP AVR statistics :: https://support.f5.com/csp/article/K14956.