[Mullins02] Section 9.2. Monitoring vs. Management

来源：百度文库编辑：神马文学网时间：2024/07/03 09:14:22

Monitoring vs. Management

Unfortunately,the DBA usually attacks performance in a reactive manner. A user callswith a response time problem. A database runs out of space. The batchwindow extends into the day. The problem has happened and now it needsto be remedied. Such activity is purely reactive.

Evenmany of the supposedly proactive steps taken against completedproduction applications might be considered reactive. A change to acompleted application that requires code to be rewritten cannotreasonably be considered proactive. A proactive approach would haveinvolved correcting the problem before completing the application.

Someevent-driven tools can be used to make performance tuning easier byautomatically taking predefined actions when prespecified alerts aretriggered. This is the first step toward performance management.Managing performance differs from monitoring performance because itcombines monitoring with a detailed plan for resolving problems whenthey arise.

Performancemanagement consists of three specific components that need to beperformed in conjunction with each other: monitoring, analysis, andcorrection, as shown in Figure 9-1. Monitoringis the first component of performance management. It consists ofscanning the environment, reviewing the output of instrumentationfacilities, and generally watching the system as it runs. Monitoring isthe process of identifying problems.

Figure 9-1. The components of performance management

Performance management consists of three steps: monitoring, analysis, and correction.

Analysis is the second component of performance management.A monitoring task can generate hundreds or thousands of messages, orreams and reams of paper reports. A monitor collects the pertinentinformation for making performance tuning and optimization decisions,but it is essentially dumb. A monitor cannot independently makedecisions based on the information it has collected. This requiresanalysis—and analysis typically is performed by a skilled technicianlike a DBA.

Optimization—the corrective action—is the third component of performance management.Some performance tools allow the technician to automate certain aspectsof performancemanagement by automatically kicking off corrective actions when thecertain prespecified monitoring agent identifies certain conditions.However, most of these tools are limited in scope.Furthermore, a skilled technician must set up the automation to makesure that the appropriate optimization step is taken at the correcttime. Eventually performance management tools and solutions will becomeintelligent—with built-in knowledge of how to optimize a DBMS and withthe ability to learn what works best from tuning exercises.

Performancemanagement can be achieved only by using a proactive performance plan.Many problems can be identified and solutions mapped out in advance,before an emergency occurs. With a proper plan, correction ofperformance problems becomes easier, and indeed, some performanceproblems can be avoided altogether because a potential problem causingsituation can be corrected before a problem occurs.

Performance management can be achieved only by using a proactive performance plan.

Fortrue proactive performance management to be achieved, the DBA must planthe performance of an application before it is completed. This requiresthe DBA to be involved in the application development life cycle and toensure that performance is designed into the application. Achievingsuch a high degree of DBA involvement in the application developmentprocess can be problematic. DBAs are stressed for time, and it seemslike there is always time to do it over later instead of simply doingit right the first time.

Reactive vs. Proactive

Reactiveperformance managementwill always be required because unplanned performance problems willalways occur. It is impossible to foresee every type of performanceproblem; after all, systems and applications change over time. Reactiveperformance management is not, in itself, a bad thing, but it is amanual, time consuming process. Proactive performance managementcombines forethought, planning, and automation to minimize reactivemonitoring and tuning. In other words, proactive performance managementreduces the amount time, effort, and human error involved inimplementing and maintaining efficient database systems.

Proactive performance management combines forethought, planning, and automation to minimize reactive monitoring and tuning.

Preproduction Performance Estimation

Ideally,DBAs and developers should engineer high performance into theirapplications during design and construction by implementing amethodology. Such a methodology must address the ADLC by incorporatingtactics to achieve high performance into the creation of the databaseand application code. A rigorous process focusing on verifiable resultscan build performance into applications and databases, therebyeliminating costly redesign and recoding efforts—at least with respectto most performance problems.

Problemsidentified earlier in the ADLC are easier to fix and cost less to fixthan problems identified later in the application's life, as shown in Figure 9-2.Proactive performance management can reduce the cost of applicationdevelopment because it occurs before the application becomesoperational in a production environment. Correcting problems after anapplication is operational is the costliest method of performancemanagement, because users are relying on the operational application toperform their jobs. Performance problems in a production applicationcan increase the amount of time it takes to perform mission-criticalwork, such as servicing customers. Moreover, severe performanceproblems can cause outages, as discussed in the last chapter.

Figure 9-2. Cost of performance problems across the application development life cycle

Estimatingthe performance of applications differs from analyzing and optimizingsingle database queries. Performance should be modeled for the entireapplication, because individual queries may optimize at the expense ofother queries. A model will show the overall effect of all the queriesand how they affect each other's performance. Such a model enables theDBA to optimize overall performance.

Performance should be modeled for the entire application.

Creatingan accurate performance model is an iterative process. Each change mustbe reviewed and updated, and its impact gauged for effectiveness. DBAs,SAs, application developers, and capacity planners must cooperate toshare information and address any business requirement issues that mayaffect the performance criteria.

Historical Trending

Capturingand analyzing resource usage trends and performance statistics overtime is another valuable performance task. Historical performance andresource trending allows DBAs to predict the need for hardware upgradesweeks, and perhaps months, in advance. Administrators can track keyperformance statistics (such as buffer hit ratios, file I/O, and logswitches) and store that information in tracker tablesin the database. This provides valuable historical information that canbe reported and analyzed. DBAs can track performance and resourceconsumption and predict when hardware resources will be consumed byincreasing usage. Furthermore, historical trends can illuminate periodswhen database performance is slower than usual due to increased useractivity. For example, database applications tend to run slower thefirst three days of the month due to month-end processing requirements.Maintaining key historical performance indicators can provide a hugebenefit to DBAs as they attempt to comprehend the performancecharacteristics of their applications, databases, and systems.

Maintaining key historical performance indicators can provide a huge benefit to DBAs.

Create Bookmark (Key: b)Create Bookmark
Create Note or Tag (Key: t)Create Note or Tag
Download (Key: d)Download
Email This Page (Key: e)Email This Page
PrintPrint

Html View (Key: h)Html View
Zoom Out (Key: -)Zoom Out
Zoom In (Key: +)Zoom In
Toggle to Full Screen (Key: f)
Previous (Key: p)Previous
Next (Key: n)Next