Introduction to NonStop Operations Management

Performance Management
Introduction to NonStop Operations Management–125507
8-8
Step 4—Optimizing System Performance
Step 4—Optimizing System Performance
Optimizing system performance involves:
•
Analyzing the results of the performance measurements
•
Identifying load imbalances and bottlenecks created by applications
•
Determining what should be done to improve system performance
•
Improving system performance by tuning the system, balancing the system,
performing online performance troubleshooting, or reconfiguring the system
The following guidelines can help the performance-analysis-and-tuning staff:
•
In a Tandem environment, performance issues that exist at a network level also exist
at a system level. In a network, a heavily used node can easily influence the
performance of nodes to which it is connected. A node can serve either as a resource
aiding the flow of work or as a bottleneck impeding the flow.
•
Performance problems are often a result of an unbalanced workload that causes
bottlenecks in the flow of work through the system. No resource should be stressed
beyond its optimal performance levels.
•
Configuration, application-development, and system-management problems are
often mistakenly viewed as performance problems; diminishing performance is
often a symptom of these other problems.
•
The key to achieving optimal performance is a workload-sharing strategy that
promotes concurrence (allows several activities to occur at the same time) and averts
contention (prevents two activities from competing for the same resource). This type
of workload-sharing strategy focuses on ensuring that the main resource users do not
contend for the same resource.
•
Load balancing is most helpful when pursued with the goal of restraint rather than
equality. To balance workloads:
1. Develop a sense of the workload and how the load is allocated among resources.
Each process uses a different mix of system resources. When you move a
process to balance one resource (for example, processor usage), you must be
aware of the effect of that process on other system resources.
2. Confine initial balancing actions to curing contention. First resolve memory
contention, and then resolve device contention.
3. After contention is eliminated, try to smooth peak use among system
components. Frequently, about 10 percent of processes and disk files account for
80 to 90 percent of processor and disk usage. You can tune and balance a system
by redistributing as few as 10 percent of the processes and disk files. Moving
processes among processors to equalize processor busy levels might not improve
system performance if the highest-usage 10 percent of the processes and disk
files are not properly distributed.