Java + JDBC的CPU使用率

Question

While testing my webapp under load I got following top excerpt under Linux:

 PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
3964 nobody    20   0 4965m 622m 6048 S  8.5 11.0   6:02.49 java 
1985 mysql     20   0  294m 125m 3804 S  2.1  2.2   0:05.39 /usr/sbin/mysqld

I need an explanation on java's %CPU column. As I understand it, during web request one of java 's thread is performing some pure Java logic, consuming some of CPU time (let it be 5ms). Then it connects to database via JDBC, sends some sql query and waits, say 10ms , for response. These 10ms are obviously counted as /usr/sbin/mysqld CPU usage. Then java thread resumes it's operation and finishes, consuming another 20ms and amounting to 5+10+20=30ms total execution time.

And the newbie question is: don't you think that these db-related 10ms are counted twice: one time as java thread waiting for db to process the query and second time as database CPU usage itself? What don't I understand here?

Answer 1

When a process is simply waiting on a socket it will not normally be using significant amounts of CPU. Basically, the kernel knows that it is in a waiting state and will not schedule it to run - except for little things like signal handling, which can be used eg to run the garbage collector regularly.

(OK, if there is a lot of garbage to collect, that "little thing" might not be so little.)