wip/jdk16 mysteriously slow on x86_64

Sinine Murakas sinine.murakas at gmail.com
Mon Aug 29 10:45:25 PDT 2011


Hi

I compiled wip/jdk16 on my new machine which has 2 cores, uname -a gives

DragonFly [name]  2.10-RELEASE DragonFly v2.10.1.1.gf7ba0-RELEASE #2:
Mon Apr 25 12:53:34 PDT 2011
root at pkgbox64.dragonflybsd.org:/usr/obj/usr/src/sys/X86_64_GENERIC_SMP
 x86_64

but it runs 2x to 4x times slower than lang/jdk16 on my old machine
which has a single core, slower processor and an i386 UP kernel. In
addition, lang/jdk16 on old box runs on linux emulation.

The slow performance of wip/jdk16 on the new machine is very puzzling
since the new box is faster on every count!

What could be wrong? I noticed is that when running java on the new 2
core machine, top shows that java is permanently in state umtxsl:


--- top ---
89626 john      0  1201M   352K  umtxsl  2   0:00    0:00  0.00% java
--- --- ---


In addition I noticed that when I run my java program in a terminal,
there appeared this output (lines starting with "thr_umtx_wait"
below):


--- java console output START ---

.  . . [program output; process runs ca 2 min] . . .

thr_umtx_wait: FAULT VALUE CHANGE 25 -> 26 oncond 0x80067071c

.  . . [program output; process runs ca 2 min] . . .

thr_umtx_wait: FAULT VALUE CHANGE 32 -> 33 oncond 0x80067071c

.  . .[more program output] . . .

thr_umtx_wait: FAULT VALUE CHANGE 36 -> 37 oncond 0x80067071c

.  . .

thr_umtx_wait: FAULT VALUE CHANGE 49 -> 50 oncond 0x80067071c

.  . .

--- java console output END ---


I wonder if this relates to system rather than java? Many thanks in
advance for help!

Sini

---






More information about the Users mailing list