Objects are created in the memory to service incoming requests.
Once requests are serviced, newly created objects will become useless
(i.e. garbage). This garbage must be evicted from the memory so that there is
enough room created in the memory to service the new incoming requests. If
there isn’t sufficient memory, the application can experience poor response
times, OutOfMemoryError, and fatal crashes.
In Java,
Android, C#…, garbage collection is automatic, whereas in the several
predecessor programming languages (C, C++) – programmer must write code
explicitly to release the objects after they are used. So, it’s a major
convenience for Java, Android, and C# application developers. But this
automatic garbage collection is not free, it comes with a price. Automatic
Garbage Collection can have a profound impact on:
1.
Application Response Time
2.
CPU
3.
Memory
Application Response Time
To
garbage collect objects automatically, entire application has to be paused
intermittently to mark the objects that are in use and sweep away the objects
that are not used. During this pause period, all customer transactions which
are in motion in the application will be stalled (i.e. frozen). Depending on
the type of GC algorithm and memory settings that you configure, pause times
can run from few milliseconds to few seconds to few minutes. Thus, Garbage
Collection can affect your application SLA (Service Level Agreement)
significantly.
CPU
Garbage
collection consumes a lot of CPU cycles. Each application will have
thousands/millions of objects sitting in memory. Each object in memory should
be investigated periodically to see whether they are in use? If it’s in use,
who is referencing it? Whether those references are still active? If they are
not in use, they should be evicted from memory. All these investigations and
computation requires a considerable amount of CPU power.
Memory
Of
course, poor GC configuration can lead to high memory consumption and vice
versa. Most applications saturate memory first before saturating other
resources (CPU, network bandwidth, storage). Most applications upgrade their
EC2 instance size to get additional memory rather to get additional CPU or
network bandwidth.
Thus to have top notch SLAs and
reduce the bill from your cloud hosting provider, your applications Garbage
collection has to be function effectively.
In order to study and optimize Garbage Collections impact on the
application’s performance, one has to enable Garbage Collection Logging.
Besides that, Garbage Collections logs can be used to troubleshoot
memory-related problems in the application.
Enabling GC
logs
GC
Logging can be enabled by passing below-mentioned system properties during
application startup
Until Java 8:
Below is
the system property that is supported by all version of Java until JDK 8.
From
Java 9:
Below is
the system property that is supported by all version of Java starting from JDK
9.
How to analyze
GC logs?
Here is
a sample GC log generated when above system properties were passed:
GC log
has rich information, however, understanding GC log is not easy. There isn’t
sufficient documentation to explain GC log format. On top of it, GC log format
is not standardized. It varies by JVM vendor (Oracle, IBM, HP, Azul, …), Java
version (1.4, 5, 6, 7, 8, 9), GC algorithm (Serial, Parallel, CMS, G1,
Shenandoah), GC system properties that you pass (-XX:+PrintGC,
-XX:+PrintGCDetails, -XX:+PrintGCDateStamps, -XX:+PrintHeapAtGC …). Based on
this permutation and combination, there are easily 60+ different GC log
formats.
Thus, to analyze GC logs, it’s highly recommended to use GC log
analysis tools such as GCeasy, HPJmeter. These tools parse GC logs and
generate great graphical visualizations of data, reports Key Performance
Indicators and several other useful metrics.
Here is a sample GC log analysis report generated by
the GCeasy tool.
About the Author
Every single day, millions & millions of people in North
America—bank, travel, and commerce—use the applications that Ram
Lakshmanan has architected. Ram is an acclaimed speaker in major
conferences on scalability, availability, and performance topics. Recently, he
has founded a startup, which specializes in troubleshootingperformance problems.
