An optimal hardware configuration enables applications to get the
greatest benefit from performance tuning. The hardware speed impacts all
types of applications and is critical to overall performance.
About this task
For proper system sizing for WebSphere Application
Server workloads, use the IBM Systems Workload Estimator.
The following
parameters include considerations for selecting and configuring the hardware
on which the application servers run.
Procedure
- Optimize disk speed
- Description: Disk speed and the number of disk arms have a significant
effect on application server performance in the following cases:
- Your application is heavily dependent on database support .
- Your application uses messaging extensively.
- Recommendation: Use disk I/O subsystems that are optimized for
performance, for example, Redundant Array of Independent Disks (RAID). Distribute
the disk processing across as many disks as possible to avoid contention issues
that occur with 1 or 2 disk systems. For more information about disk arms
and how they can affect performance, see the iSeries Disk Arm Requirements documentation.
- Increase processor speed and processor cache
Description: In
the absence of other bottlenecks, increasing the processor speed often helps
throughput and response times. A processor with a larger L2 or L3 cache yields
higher throughput, even if the processor speed is the same as a CPU with a
smaller L2 or L3 cache.
Description: In
the absence of other bottlenecks, increasing the processing power can improve
throughput, response times, or both. On WebSphere Application Server for i5/OS,
processing power can be related to the Commercial Processing Workload (CPW)
value of the system. For more information about CPW values, see the http://www.ibm.com/servers/eserver/iseries/perfmgmt/resource.html Web
site.
- Increase system memory
- Description: The amount of storage required for z/OS is mostly
dependent on the number of servers and the size of the Java Virtual Machine
(JVM) heap for each server.
- Recommendation: For a single server with 1 GB JVM heap, allocate
a minimum of 1GB of memory.
- Description: Increase memory to prevent the system from paging
memory to the disk to improve performance. Allow a minimum of 256 MB of memory
for each processor and 512 MB per application server. Adjust the available
memory when the system pages and the processor utilization is low because
of the paging. The memory access speed might depend on the number and placement
of the memory modules. Check the hardware manual to make sure that your configuration
is optimal.
- Recommendation: Use 256 MB of memory for each processor and 512
MB per application server. Some applications might require more memory.
- Description: If a large number of page faults occur, performing
the following tasks to improve performance:
- Increase the memory available to WebSphere Application Server.
- Move WebSphere Application Server to another memory pool.
- Remove jobs from the WebSphere Application Server memory pool
- Recommendation: To determine the current page fault level, run
the Work with System Status (WRKSYSSTS) command from an i5/OS command line.
For information about the minimum memory requirements, see the IBM Support Web site.
- Run network cards and network switches at full duplex
- Description: Run network cards and network switches at full duplex
and use the highest supported speed. Full duplex is much faster than half
duplex. Verify that the network speed of adapters, cables, switches, and other
devices can accommodate the required throughput. Some Web sites might require
multiple gigabit links.
- Recommendation Make sure that the highest speed is in use on 10/100/1000
Ethernet networks.
Verify that the activity levels for storage pools are sufficient
- Description: Verify that the activity levels for storage pools
are sufficient. Increasing these values can prevent threads from transitioning
into the ineligible condition.
- Recommendation
An IBM
S/390 or zSeries Model that supports the software requirement of z/OS
V1R2.
Storage
- Storage requirements are higher than for traditional workloads
- Recommendation
- Virtual storage default should be about 370 MB per servant, which includes
a 256 MB default heap size and a default initial LE heap size of 80 MB.
- Real storage minimum is 512 MB per LPAR for a light load such as the IVP.
For most real-world applications, we recommend 2 GB or higher. We have seen
applications that require as much as 8 GB of real to operate at peak load.
DASD
- Recommendation
- To maximize your performance, we recommend a fast DASD subsystem (for
example, IBM Shark), running with a high cache read/write hit rate.
Networking
- Recommendation
- For high bandwidth applications, we recommend at least a 1 Gb Ethernet
connection. If your applications have extremely high bandwidth requirements,
you may need additional Ethernet connections.