When running Java applications with -XX:+UseConcMarkSweepGC, jstat's CGCT is 5 percent higher than the pause time in -Xlog:gc.

Steps to Reproduce
Run the attached application.
java -Xms64m -Xmx64m -Xlog:gc CMSGCALot 64

Expected Result
jstat's CGCTis approximately equal to the pause time in -Xlog:gc.

Actual Result
jstat's CGCT is 5 percent higher than the pause time in -Xlog:gc.

jstat
    S0 S1 E O M CCS YGC YGCT FGC FGCT CGC CGCT GCT
    0.00 0.00 42.33 100.00 75.46 62.08 422 0.080 5934 87.779 5874 7.824 95.683

-Xlog:gc
    [0.247s][info][gc] GC(2) Pause Initial Mark 33M->33M(61M) 1.075ms
    [0.257s][info][gc] GC(2) Pause Remark 42M->42M(61M) 1.956ms
    ...

    [144.450s][info][gc] GC(9517) Pause Initial Mark 43M->43M(61M) 0.230ms
    [144.457s][info][gc] GC(9517) Pause Remark 61M->61M(61M) 2.176ms

Aggregating the gclog, the number of CMS cycle is 5874,
and the sum of CMS pause time is 7392.646 ms
jstat's CGC value is equal, but CGCT value is 5.8 percent higher than -Xlog:gc.