The following hot list contains recommendations that have improved performance or scalability, or both, for many applications.
Review the hardware and software requirements on the IBM WebSphere Application Server supported hardware, software, and APIs Web site to get started.
The list of recommended updates is maintained on the Support site.
Sometimes transient errors can cause Ethernet
adapters to shift down to a lower speed. Verify that the system has
adequate memory and that the number and position of memory dual inline
memory module (DIMM) are optimal. With some systems, there are some
memory DIMM configurations that permit higher performance than other
DIMM configurations. Verify that the hardware that is used is the
hardware that is supposed to be used.
You can track many performance problems back to the application design. Review the design to determine if it causes performance problems.
Operating system configuration plays a key role in performance. In many cases, adjustments to some TCP/IP parameters might be necessary for your application.
Many applications need a larger heap size for best performance.
The type 4 JDBC drivers generally perform faster than type 2 JDBC drivers. Use the immediately preceding link to view a list of database vendor-specific requirements, which can tell you if a type 4 JDBC driver is supported for your database.
The JDBC data source configuration might have a significant performance impact. For example, the connection pool size and prepared statement cache need to be sized based on the number of concurrent requests being processed and the design of the application.
See the Connection pooling topic for more information.
Use applications that can take advantage of the pass by reference option to avoid the cost of copying parameters to the stack.
Some applications generate a high rate of writes to the WebSphere Application Server transaction log. Locating the transaction log on a fast disk or disk array can improve response time.
In many cases, some other component, for example, a database, needs adjustments to achieve higher throughput for your entire configuration.
For more information, see the
Queuing network topic.
One of the goals of the WebSphere Application Server for z/OS programming model and runtime is to significantly simplify the work required for application developers to write and deploy applications. Sometimes we say that WebSphere Application Server for z/OS relieves the application programmer of many of the plumbing tasks involved in developing applications. For example, application code in WebSphere Application Server for z/OS does not concern itself directly with remote communication--it locates objects which may be local or remote and drives methods. Therefore, you won't see any direct use of socket calls or TCP/IP programming in a WebSphere Application Server for z/OS application.
This separation of what you want to do from where you do it is one aspect of removing the application programmers from plumbing tasks. Other considerations are not having to deal with data calls for some types of beans, potentially user authentication, and threading. There are generally no calls from the application code to touch sockets, RACF calls, or management of threading. Removing this from the application programmer doesn't mean this work won't get done. Rather, it means that there may be more work for the DBA, the network administrator, the security administrator, and the performance analyst.
We deal with the first three in separate sections under this article and briefly touch on the fourth. For more information on tuning applications, refer to Using application clients.
Steps involved in tuning the z/OS subsystems to optimize WebSphere performance include:
The first thing to do is review the WebSphere for z/OS configuration. One simple way to do this is to look in your application control and server regions in SDSF. When each server starts, the runtime prints out the current configuration data in the joblog.
WebSphere traces can be extremely helpful in detecting and diagnosing problems. By properly setting trace options, you can capture the information needed to detect problems without significant performance overhead.
You should check your WebSphere for z/OS tracing options to ensure that ras_trace_defaultTracingLevel=0 or 1, and that ras_trace_basic and ras_trace_detail are not set.
If you are tracing to sysprint with ras_trace_defaultTracingLevel=3, you may experience an almost 100% throughput degradation. If you are tracing to CTRACE, however, you may only experience a 15% degradation in throughput.
This will get 512KB of storage for the trace buffers (the minimum allowed) and reduce memory requirements.
com.ibm.ejs.*=all=disable com.ibm.ws390.orb=all=disableEnsure that both lines are set to =disable or delete the two lines altogether.
The next thing to review in the configuration is where your program code is located. IBM recommends that you install as much of the WebSphere for z/OS code in LPA as is reasonable, and the remainder in the linklist. This ensures that you have eliminated any unnecessary steplibs which can adversely affect performance. If you must use STEPLIBs, verify that any STEPLIB DDs in the controller and servant procs do not point to any unnecessary libraries. Refer to UNIX System Services (USS) tuning tips for z/OS for USS shared file system tuning considerations.
If you choose to not put most of the runtime in LPA, you may find that your processor storage gets a bigger workout as the load increases. At a minimum, WebSphere for z/OS will start three address spaces, so that any code that is not shared will load three copies rather than one. As the load increases, many more servants may start and will contribute additional load on processor storage.
Review the PATH statement to ensure that only required programs are in the PATH and that the order of the PATH places frequently-referenced programs in the front.