Monitoring plug-in technology
Server Resource Monitoring
Index Data Analysis
Jmeter Plugin Manager
Plug Management Pack
http://jmeter-plugins.org/downloads/all/ download package is a jar, placed apache-jmeter-4.0 / lib / ext which can
JMeterPlugins-Extras.jar jmeter_plugins-manager-1.3.jar JMeterPlugins-Standard.jar three plug-ins
Server Performance Monitoring:
Client: Jmeter end of PerfMon Metrics Collector "Add" - "" listener "-" "PerfMon Metrics Collector" performance metrics collector
Add Row to add a server needs to obtain information, IP port acquired content
All data is written to a file, you need to fill in the name of the file (txt text to change the suffix CSV file), or else may fail
Resulting chart, right click you can export a CSV file or picture
The core server monitors the number of CPU see, e.g. binuclear most 200% up to 800% eight-core; but averaged monitoring tool, a maximum of 100%
Server: Monitoring Server is running in ServerAgent, ServerAgent require a separate download and run run startAgent.sh
System performance indicators: TPS clicks per second, per second throughput processing server response time success rate of the number of concurrent transaction timeout error rate
System resource indicators: CPU IO bandwidth content
CPU usage: upper limit is not more than 80% cases 1000TPS, Latency <3S, running 12 hours, the average CPU utilization rate of less than 80%
Memory utilization: Usually at least 10% of the available memory, the acceptable upper limit is 80%
Performance analysis process:
1. Performance: response time> 10s
2. Why so high, look jmeter monitoring system metrics: CPU Memory Disk IO
3. If the CPU usage> 90%, can not be concluded that there is a problem CPU
4. The use of a process which depends on the specific rate (Top instruction)
(1) If the software is not a test of a high share - to Kill off other processes, and then test
(2) If the test itself is a software system --cup is probably the bottleneck
5. verification, it is determined whether or not a bottleneck cpu
(1) may be configured to deploy a clean & cpu system environment higher
(2) is not to reduce the number of concurrent down to see how the situation (if the code in question, CPU occupancy rate is still high)