Introduction to Performance Management

Performance management of a specific process or of the entire system largely depends on the manager's common sense. The resource (i.e., the computer system) is finite, and performance improvements cannot be squeezed out of it indefinitely. Before embarking on any system improvements, a plan must be carefully developed. Key elements of that plan include the following:

• Measure and characterize the average system performance as it exists before any changes.

• Based on step 1, identify a single bottleneck. It is important to attack only one problem at a time to eliminate unintended interactions between multiple changes. Quantify the improvement you expect to see (e.g., a goal might be stated as "reduce I/Os per second by 10 percent").

• Investigate that single bottleneck to determine how best to attain the improvement goal.

• Implement that change. Record everything that is done so the change can be backed out if necessary.

• Repeat exactly the measurements taken in step 1 and examine the bottleneck identified in step 2 again. Determine if the goal was attained. Also, make sure the change did not adversely affect any other measurement.

• If the bottleneck improvement goal was not reached, back out the change and start again at step 3 to find another approach.

• If the bottleneck goal was reached, you may want to repeat the entire process, concentrating on another bottleneck, and create a new goal.

Generally, tuning OpenVMS is unnecessary and should be considered only as a last resort, because running AUTOGEN regularly (discussed in the next section) should identify and correct normal retuning activities. Assuming the manager uses AUTOGEN and considers its suggested changes, several other options should be considered before changing OpenVMS parameters. Some possibilities are:

• Restrict the number of interactive logins.

• Reduce the priority of resource-intensive interactive programs.

• Move resource-intensive programs to the batch stream or run them at night or on weekends.

• INSTALL programs that are used often. This will reduce disk I/O.

• Move certain system files off the system disk to reduce system disk contention.

• Likewise, scatter often-accessed user disk files over several disks.

• Stripe heavily used disks to improve I/O throughput.

An OpenVMS manager-novice should be warned that without a thorough understanding of the OpenVMS parameters and without a complete understanding of the OpenVMS algorithms depending on those parameters, changes may do more harm than good. For instance, the OpenVMS paging algorithm is complex and involves several parameters. If a process is thrashing (thus causing excessive page faults), modifying the process, not OpenVMS, is the recommended approach. For instance, the process might be recoded or its quotas could be increased; however, the Compaq/HP and third-party tuning manuals describe the paging algorithm in great detail, explaining the purpose of each of the parameters. So the manager can knowledgeably tune the algorithm if necessary.

The resources that can be monitored and tuned in OpenVMS are:

• CPU

• Memory

• Disk I/O

• Several network protocol I/Os

The first step to effect performance management is to establish average peak workload baseline measurements on the system as it is currently configured. The baseline data is collected at the busiest time of day, or perhaps the busiest time of the week or month depending on your business's computer needs.

There are many possible approaches to resolving performance issues. It is important for the manager to understand the difference between workload and system capacity. Depending on the characteristics of the jobs running on the computer, there may be several possible solutions. The OpenVMS Performance Management Manual includes detailed algorithms used to identify and correct performance issues.

Several programs can be used to investigate system performance that may lead to either a hardware or software modification. This section introduces only two of the most commonly used performance measurement programs. Compaq/HP recommends that even if you are not having performance problems, you should record key performance measures to establish a baseline for future investigations.

For instance, MONITOR (described later) has an option (/RECORD) to save results in a file.

The other programs that are available in OpenVMS are:

• Accounting utility (ACCOUNTING)

• Audit analysis utility (ANALYZE/AUDIT)

• Authorize utility (RUN AUTHORIZE)

• Various SHOW commands

• Compaq/HP availability manager (also called DECamds)

• DECevent utility (on Alpha platforms)

• Error log utility (ANALYZE/ERROR_LOG) on VAX platforms