Java Performance Tuning – A top down approach

The Performance Zone is brought to you in partnership with AppDynamics. See Gartner’s latest research on the application performance monitoring landscape and how APM suites are becoming more and more critical to the business.

Symptoms

  • Memory
  • CPU
  • I/O
    • Devices
    • Network
Indicators – Windowsundefined

Indicators – Linux

  • Sar, top, dstat, etc.
undefined

Jumping Board – JConsole

  • Study the JVM Args
  • Heap Usage should follow a saw tooth pattern denoting there is no leak
  • There should not be any anonymous threads like ‘Thread 1’
  • Look for creation of dynamic classes in large numbers
  • CPU Usage

Heap ?

  • Heap is utilized throughout the Object Life-cycle starting from new object creation, variable references, dereferences by variables going out of scope and finally till Garbage Collection
  • Object references can be represented in a tree fashion as we do deep references and the cumulative size retained at each level can be determined using offline utilities
  • There can multiple top level objects in the current program flow context that can retain individual object tree
  • A parent node should be shallow in size such that it does not retain memory compared to its child nodes
  • When there is a large difference in retained size between parent and child node, i.e drops, there is a possibility of memory leak

Deep Dive – Heap Analysis

Online

  • VisualVM, JRockit, IBM Health Center
  • Profilers

Offline

  • IBM Heap Analyzer
  • Eclipse MAT
Deep Dive – Memory Leaks
undefined

CPU?

  • Thread dumps provides thread status, monitors, number of threads waiting for the monitor and call stack of methods
  • Comparing thread dumps taken at frequent intervals can reveal as to which tasks (call stack) are being performed across thread dumps
  • Online profiling tools can provide hotspot methods
  • Snapshots can provide thread wise call tree, along with response time

Deep Dive – Threads & Call Stack

Online
  • VisualVM, JRockit, IBM Health Center
  • Profilers
Offline
  • IBM Thread and Monitor Dump Analyzer

Deep Dive – Threads & Call Stack

undefined

undefined

I/O?

  • Study the read, writes and response time of the Java process and identify the device that is heavily utilized or not responding
  • Study the network traffic in terms of number of connections, bytes sent / received and transfer rate

Deep Dive – I/O

  • Based on the heavy reads / writes look for input and output streams without buffering, multiple reads of same resources, unnecessary serialization, etc
  • Re-use of connection, file transfer in uncompressed form, releasing network resources (such as result set, statement and connection), etc

Performance Case Study – Heap

undefined

Performance Case Study – CPU

undefined

Best Practices

  • Understand your application environment such as firewall, layers (web, app & db), third party data providers if any
  • Use offline tools and less intrusive tools in production environments
  • Look for tools provided by the JVM vendor or the Application Server vendor
  • Use JMeter for recreating production scenario in your environment
The Performance Zone is brought to you in partnership with AppDynamics. See Gartner’s latest research on the application performance monitoring landscape and how APM suites are becoming more and more critical to the business.

Leave a Reply