Outline
Locally made a simple stress test for the login interface project. 200 concurrent continuous 120s, throughput was observed
After the end of the run, the throughput is this
As shown, the great fluctuation in throughput, completely normal. Now we need to look at the server
mpstat -P ALL 1 * look at the operation of the cpu
It can be found cpu utilization present a stepwise increasing trend, but the load is not high, indicating that the problem is not running cpu
jstat -gcutil 1 1000 to look at the situation gc memory
Old's memory space is insufficient, and therefore lead to a new generation of objects do not come into frequent fullgc, fullgc time will very long, so do not increase the throughput has been
checked jvm memory space configuration
A total of only 1g of heap memory, almost the whole part to a new generation, resulting in poor years old space of only 5M
Modify memory configuration
now to amend the memory parameters, then add a parallel mechanism for recovery
Run the script, TPS and gc observed frequency again
This run, fullgc the frequency becomes very low, but the throughput is relatively stable, there is no large fluctuations. But to run a half minutes of time, throughput declined collapse of the way, while there has been an exception.
Observed abnormal log and found that more than the maximum number of connections a tomcat
** tomcat modify the number of connections configured , run the script again
Unlike the larger area being given just that, but there are still some exceptions. Part of overtime, some of it is Software caused connection abort: recv failed
Adjust the connection request mode using java, and maintain a long connection, and then observe the operation results
This time a error are gone!