2

We had a Java app that showed 100% CPU usage all the time. I was trying to find out if there were a few dominating threads when I noticed some strange results with the top command.

If I run top, it showed one java process that had the 100% CPU time. Then I typed H to show threads, it started with showing several Java threads each with 100%. However, the next refresh, it showed a different batch of several Java threads again each with 100% CPU. The next refresh, another batch. This went on and on with several refresh cycles going through 100 or so threads of 100% CPU. Finally it settled down with a dozen or so Java threads each had about 10% CPU time. This "final" set of threads were the same threads that showed 100% CPU time once at the beginning but now they each had 10% only. If run top -H directly, I got to the final set directly.

My intuition is that the 100% CPU for the Java threads was somewhat bogus. But I just couldn't find an explanation for it.

I am on a Debian Wheezy box.

neurite
  • 2,798
  • 20
  • 32
  • Just learned today that the `top` command computes exponentially-weighted moving average (EWMA) on a 15-minute window. But still does not explain what I saw. – neurite Mar 22 '17 at 01:32

1 Answers1

1

With perl threads I've gotten well over 100%, so I think top is definitely not thread aware.

PID   USER    PR NI VIRT RES SHR  S %CPU  %MEM TIME+   COMMAND
19841 v807576 15 0  402m 13m 2184 S 498.7 0.1  9:14.71 mt_analyze_ets

Actually not true. With the -H option I get this:

PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+COMMAND                                                                                         
18118 v807576   18   0  653m  22m 2164 R 96.3  0.1   0:02.98 mt_analyze_et                                                                                
18124 v807576   18   0  653m  22m 2164 R 96.3  0.1   0:02.98 mt_analyze_ets                                                                                 
18117 v807576   18   0  653m  22m 2164 R 95.3  0.1   0:02.95 mt_analyze_ets                                                                                 
18121 v807576   18   0  653m  22m 2164 R 94.7  0.1   0:02.93 mt_analyze_ets                                                                                 
18127 v807576   18   0  653m  22m 2164 R 94.3  0.1   0:02.92 mt_analyze_ets                                                                                 
18133 v807576   18   0  653m  22m 2164 R 94.3  0.1   0:02.92 mt_analyze_ets                                                                                 
18136 v807576   18   0  653m  22m 2164 R 94.3  0.1   0:02.92 mt_analyze_ets                                                                                 
18145 v807576   18   0  653m  22m 2164 R 94.0  0.1   0:02.91 mt_analyze_ets                                                                                 
18142 v807576   18   0  653m  22m 2164 R 93.1  0.1   0:02.88 mt_analyze_ets                                                                                 
18130 v807576   18   0  653m  22m 2164 R 92.1  0.1   0:02.85 mt_analyze_ets                                                                                 
18148 v807576   18   0  653m  22m 2164 R 90.8  0.1   0:02.81 mt_analyze_ets                                                                                 
18116 v807576   18   0  653m  22m 2164 S  5.2  0.1   0:00.16 mt_analyze_ets
David Waddell
  • 11
  • 2