 |
| |
|
 | IMPORTANT: The information in this section is supported on only JDK 5.0.14
(or later) and JDK 6.0.02 (or later) releases. Additionally, to properly
view zero preparation Xverbosegc and allocation statistics information,
you must use HPjmeter 3.1 (or later). |
|
| |
|
For Java applications started without -Xverbosegc, you can prompt the Java Virtual Machine (JVM) to emit HP Xverbosegc
information by sending the running Java program a SIGPROF signal. For example, assume the running Java process ID is 6754.
Executing either of the two actions listed below prompts the JVM to
start emitting detailed HP Xverbosegc information:
kill -21 6754
kill -s sigprof 6754 |
Issuing a second SIGPROF signal to
the running Java program prompts the JVM to stop emitting the Xverbosegc
information. Similarly, issuing yet another SIGPROF signal to the running process prompts the JVM to start issuing Xverbosegc
information one more time.
Additionally, and only if you had started the Java application
with -XX:+PrintAllocStatistics, when zero preparation
Xverbosegc is enabled as specified in the previous paragraph, the
JVM also emits detailed allocation site statistics.
If you start your application with -XX:+PrintAllocStatistics, and if you do not specify -Xverbosegc, or send
the running process a SIGPROF signal, then
the JVM emits allocation site statistics to standard out when the
application terminates. (That is, when the Java process stops running.)
If you do not want to wait for the application to terminate, or if
the application is designed to be long running (or does not terminate),
then you can obtain allocation site statistics by sending the running
Java program a SIGPROF as described above.
Unlike Xverbosegc information emission in zero preparation mode,
which is started with a SIGPROF and stopped
with a second SIGPROF, when -XX:+PrintAllocStatistics is specified, allocation site statistics are emitted every time
the JVM receives a SIGPROF.
When -XX:+PrintAllocStatistics is specified
along with -Xverbosegc, allocation site statistics
data are directed to the same output specified for HP Xverbosegc.
If no output file is specified, allocation site statistics are directed
to standard out (to the terminal where the program was started from).
When -XX:+PrintAllocStatistics is specified alone
(without -Xverbosegc on the command line), allocation
site statistics are directed to standard out (to the terminal).
When a SIGPROF signal is sent to a
running Java process that was originally started with -XX:+PrintAllocStatistics on the command line, the SIGPROF signal
also enables Xverbosegc data collection if Xverbosegc was not specified
on the command line when the program was started. The output of both
Xverbosegc and allocation site statistics is directed to a file created
for you by the JVM. The name of the file created by the JVM has the
following format: java_<pid>.vgc. Where <pid> is the Java process ID.
The following table lists all possible permutations and corresponding
locations of emitted data for both allocation site statistics and
zero preparation Xverbosegc.
Table 1-17 Actions, JVM Options Specified, and Allocation Site Statistics
Data Locations
| Action | JVM Options Specified | Allocation Site Statistics Data Locations |
|---|
| No special
action taken | -XX:+PrintAllocStatistics | Standard out |
| -Xverbosegc |
| No special
action taken | -XX:+PrintAllocStatistics | Same location as Xverbosegc data file (mydata.vgc) |
| -Xverbosegc:file=mydata.vgc |
| No special
action taken | -XX:+PrintAllocStatistics | Standard out |
| |
| 'kill 21 <pid>' | -XX:+PrintAllocStatistics | JVM creates a file for you, file name format is: java_<pid>.vgc | |
| 'kill 21
<pid>' | -XX:+PrintAllocStatistics | Standard out | |
| -Xverbosegc | |
| 'kill 21
<pid>' | -XX:+PrintAllocStatistics | Same location as Xverbosegc data file (mydata.vgc) | |
| -Xverbosegc:file=mydata.vgc | |
| 'kill 21 <pid>' | -Xverbosegc | Standard out | |
| 'kill 21 <pid>' | -Xverbosegc:file=mydata.vgc | Same location as Xverbosegc data file (mydata.vgc) | |
Sending multiple SIGPROF signals to
a running Java process produces multiple allocation site statistics
dumps and the JVM dumps the buffered data immediately after the SIGPROF is received. Allocation site statistics counters
inside the JVM are reset after each SIGPROF induced the dump of the data. HPjmeter consolidates data from multiple
allocation site statistics dumps into one report that is presented
in a new tab in the Xverbosegc data visualizer.
Allocation sites can originate from interpreted as well as compiled
Java code when specifying -XX:+PrintAllocStatistics, and only allocations coming from compiled code are reported. The
JVM detects and compiles the application’s (and JDK's) most
active Java methods as early as possible. Though reporting allocation
sites originating from compiled code is only incomplete from a comprehensive
reporting point of view, it does always report the most active allocation
sites (the sites most likely to cause GC performance problems).
In the following example, the default sort on the table is determined
by % Allocated, meaning the percent of memory
allocated to the objects generated by the compiled portion of the
running Java application. The values in this column are displayed
in descending order, from greatest area of memory use to the least.
Making code changes first in the area where memory use is greatest
can have more significant impact on the heap.
For more details about Allocation Site Statistics in HPjmeter, see the HPjmeter User's Guide at http://docs.hp.com/en/dev.html#Java.
Printable version
