Analyzing Performance Graphs
ZCS Administrators can generate graphs showing system performance over time. Performance data is stored in CSV files under /opt/zimbra/zmstat, with a subdirectory for each day of collected data. The current day’s data is always stored in /opt/zimbra/zmstat/. Admins can use zmstat-chart to generate graphs of performance data with the following command:
zmstat-chart –d <destination_directory> -s <source_directory>
Other command line options to zmstat-chart allow admins to focus in on a specific window of time. This guide will detail each element of the generated data and help administrators understand what is being presented and how it can be used to understand whether the system is operating correctly. This document uses performance graphs taken from actual ZCS systems running in production. Most graphs are taken from an enterprise mail system running under fairly light load. Others, depicting stressed systems or systems with specific problems, are taken from systems with much heavier loads.
System CPU Graphs
The first set of graphs show total CPU usage. There is a graph for aggregate CPU, and individual graphs for User, Nice, Sys, Idle, IOWait, IRQ and SoftIRQ, broken out per processor.
Host sample.testdomain.com: Total CPU
avg(user) = 9.99 avg(sys) = 1.19 avg(idle) = 86.07 avg(iowait) = 2.69 avg(nice) = 0.00 avg(irq) = 0.00 avg(softirq) = 0.04
This graph shows a fairly typical usage pattern for a production ZCS system. Backups kick off at 1 AM, which causes a long period of high IOWait usage. The rest of the day shows an increase in User CPU, as the mailboxd process picks up more load while serving data to users.