(If the figure is small, you can Ctrl + mouse wheel to zoom)
Based on the last "Legend distributed server cluster concurrent landing on-line game test results" after the logical processing stress testing, still only configured a single gateway and a single logic service in the cluster, the results are as follows:
Test Case:
The first step: on-line 10,000 players, and then every 10 seconds for each player to send a logical calculation request to the gateway server, the gateway server forward the request to the logical server processing, the logical server 1 million floating point operations, and then return the results to the gateway server, The final result is forwarded by the gateway server to the player client,
First 10 robots, each on-line 1000,
It can be seen from the first online a batch account to the last online J batch account completed, a total of 10,000 people took 30 seconds, an average of 333 players/second, next to each robot to make floating point operation request
The second step: the single player robot forwards every 65 milliseconds to the gateway server to make a request for the logical server to make 1 million floating-point operations, and finally to calculate the cost of returning to the robot,
Finally look at the CPU and configuration status of the gateway server and logical server:
Test results:
10,000 people in the 10-second interval to do millions of floating-point operations, a single player to do the same operation request when the response time is about 30ms, if the gateway server CPU configuration is higher, the effect will be more obvious, if configured as a multi-gateway server, multi-logical server, the effect will be further improved.
Legend distributed server cluster logic computation processing stress test