Review Request: 8009298 and 8009777

[Coleen Phillmore]

Zhengyu,
This looks good.   NIce documentation for the flag in globals.hpp. Can 
you reword slightly:

+           "Automatically shutdown native memory tracking in stress "        \
+           "situations. When it is set to false, native memory tracking"      \
+           "will try its best to stay alive, even if it means that it has to" \
+           "sacrifice JVM performance")                                      \


Coleen

On 3/11/2013 2:21 PM, Zhengyu Gu wrote:
> 8009298: NMT: Special version of class loading/unloading with runThese 
> stresses out NMT <https://jbs.oracle.com/bugs/browse/JDK-8009298>
> 8009777: NMT: add new NMT dcmd to control auto shutdown option 
> <https://jbs.oracle.com/bugs/browse/JDK-8009777>
>
> The two are related. The change is to add an addition VM diagnostic 
> command-line option "AutoShutdownNMT".
>
> By default, native memory tracking runtime shuts itself down under 
> stress situation, and release memory to allow JVM continue running. 
> But this feature can be undesirable when this is the situation we 
> really want to know what's happening to JVM memory usage, even that 
> means that we need to sacrifice JVM performance.
>
> When AutoShutdownNMT is set to false, NMT will try its best to stay 
> alive by slowing down threads calling memory allocation/de-allocation.
>
> Webrev: http://cr.openjdk.java.net/~zgu/8009298/webrev.00/
>
> Tests:
> The test case is special version of runThese, it uses 12 threads to 
> load/unloading classes without executing any tests.
>
> Following two memory usage graphs show the different behavior with 
> AutoShutdownNMT on/off.
>
> AutoShutdownNMT = true
> http://cr.openjdk.java.net/~zgu/8009298/webrev.00/autoShutdown.png
>   The steep drop in memory usage was result of shutting down of native 
> memory tracking.
>
> AutoShutdownNMT = false
> http://cr.openjdk.java.net/~zgu/8009298/webrev.00/noAutoShutdown.png
>    The steep dig was result of class unloading.
>
>
> Thanks,
>
> -Zhengyu
>