Performance and resource management overview

Performance management is necessary to keep users satisfied and to ensure that business-critical applications and transactions have the resources they need. Resource management is necessary to help companies use computing resources more efficiently and effectively, and to reduce administration costs. Many companies want to consolidate (1) their data centers onto fewer systems and (2) multiple applications onto a single server. Managing both performance and system resources requires maintaining a dynamic balance that optimizes resource utilization while also maintaining performance goals, automatically re-allocating resources in response to changing priorities and conditions. Basically, performance and resource management requires:

Monitoring the system to understand how the current combination of system resources, applications, and users affects performance and resource utilization
Controlling performance by managing the system’s resources, applications, and number of users

The methods to monitor and control performance and resource utilization can vary greatly.

Table 1-1 “Performance and resource utilization monitoring methods” explores advantages and disadvantages of several monitoring methods.

Table 1-1 Performance and resource utilization monitoring methods

Method	Advantages	Disadvantages
Wait for users’ complaints about performance	User-focused System configuration is kept simple to reduce number of complaints and speed resolution	Re-active (not pro-active) Inexact Requires one application/server model to keep number of calls low
Predict performance by monitoring a particular application’s resource usage	Detects faulty applications that are over-consuming resources Usage trends can be used to predict future loads	End-to-end performance is not always tied to resource consumption Few options if problems are detected (separate applications; buy more servers)
Directly monitor through performance metrics (for example, response time or throughput) obtained from the application	Exact performance metrics Can be used pro-actively to guarantee user’s satisfaction with performance Performance management is designed into the application	Limited number of applications are instrumented for tracking capabilities Application source code may not be available for instrumentation Instrumenting source code can be time-intensive Every different application on the system needs the tracking capability
Directly monitor utilization of CPU resources, partitions, memory, disk bandwidth, operating system instances and the demand imposed by users and specific applications	Detects applications that are over-consuming resources Detects resources that are being over- or under-consumed Can be used pro-actively to guarantee optimal resource utilization Monitoring is easier to implement	Not as accurate as monitoring metrics

After determining which methods you want to use for monitoring performance and resource utilization, decide how to control performance and resource utilization. Table 1-2 “Performance controlling methods” examines the advantages and disadvantages of various control methods.

Table 1-2 Performance controlling methods

Method	Advantages	Disadvantages
One workload per server	Workloads do not compete for resources	Server resources sit idle a large percentage of the time
Multiple workloads: No resource management	Increased resource usage Fewer servers needed	Workloads compete for resources A single workload may take over the server, preventing other workloads from getting resources
Multiple workloads: Fixed resource allocation Example: PRM with capping enabled	Workloads’ resource use cannot exceed resource allocations A workload cannot take over a server to the detriment of other workloads Consistent performance when work stays constant in a workload Excess resources can easily be tracked with PRM and GlancePlus tools	Performance may degrade when workloads cannot exceed resource allocations that are set too low Unused capacity cannot be shared optimally if resource usage is capped
Multiple workloads: Variable resource allocations based on usage Example: PRM without capping enabled	Workloads are guaranteed resource minimums and can borrow unused resources from other workloads Increased level of work is handled automatically Excess resources can easily be tracked with PRM and GlancePlus tools	Workloads get resources, but performance can still be less than desired Over-performance followed by typical performance can cause complaints
Multiple workloads: Variable resource allocations based on actual, reported performance Example: HP-UX WLM	Consistent performance levels are maintained automatically Workloads can be prioritized to ensure that high-priority workloads are guaranteed CPU resources as needed Excess resources can easily be tracked with PRM and GlancePlus tools so that high-priority workloads are guaranteed the available resources	Cannot be implemented immediately: application performance must be assessed first Getting performance metrics for workloads can be difficult; however, WLM simplifies this task with a built-in data collector
Multiple workloads: Variable resource allocations based on CPU utilization per workload Example: HP-UX WLM	Optimal resource utilization is maintained automatically Workloads can be prioritized to ensure that high-priority workloads are guaranteed CPU resources as needed Excess resources can easily be tracked with PRM and GlancePlus tools enabling reserve capacity to be deployed to save costs and guaranteesing high-priority workloads the available resources, as needed	CPU resource allocations are not made based on application-specific metrics

The remainder of this document focuses on the last two rows of Table 1-2 “Performance controlling methods”: Automatically managing multiple, prioritized workloads on a single server—possibly across partitions—based on reported performance levels or usage goals. This strategy is commonly referred to as goal-based workload management.

WLM implements this type of workload management allowing you to:

Optimize your return on investment in servers
Maintain multiple applications on a single server without increasing performance problems
Improve your ability to forecast capacity needs

Performance and resource management overview

Technical documentation

» Table of Contents

» Glossary

» Index