Types of Capacity Planning

3 4

Capacity planning comes in two forms: pre-capacity planning and post-capacity planning. Pre-capacity planning, or sizing, involves anticipating the hardware requirements necessary to process your workload within a specified time, as spelled out in the Service Level Agreements (SLA). SLAs are set up to ensure that the response times (the time it takes for an activity or transaction to complete) of certain functions are maintained.

NOTE


An SLA is a condition of operation agreed upon by all organizations involved with the system in question, developed to ensure high performance and smooth operation of the system. For example, an SLA might be developed to ensure that the system meets a certain response time for a query. This response time is agreed upon by the users, the operations group, the applications group, and the performance group.

Also, a certain amount of reserve capacity (space allocated for CPU processing power, space available on a disk drive, or available memory) is set aside to maintain the response times of these activities under steady state operation and peak load conditions. In pre-capacity planning, there is no real performance data to work with because the system is not yet functioning. You must use whatever other information is available. The results will vary depending on the accuracy of this information. For example, the database group that is designing the system can provide details about the database layout and the initial size. The applications group that is designing the application and the various queries that are associated with the application can provide information about how these queries will use system resources. The management group will have information about the number of concurrent users and the number of queries they will put through the system. All this information will give details as to the workload (so you can guess at the number of CPUs), the database size (so you will know how many disk drives you will need), and so on.

Post-capacity planning, or predictive analysis, is a complex and ongoing study of hardware and software resource consumption on a system that is already set up and running. Post-capacity planning ensures adequate preparation for workload growth in relation to system resources. These studies are primarily established to provide data to the database administrator (DBA). The DBA uses this data to justify system alterations designed to maintain the level of system performance defined in the SLA. In this chapter, we will look at the two kinds of capacity planning—post-capacity planning and pre-capacity planning—and examine their similarities and differences.

In a typical post-capacity planning scenario, you perform the analysis using historical performance data stored in a database. Through this analysis, you can project trends in the normal growth of CPU utilization (the amount of time a CPU is busy during an observation period), disk usage, memory usage, and network usage. You will also be able to project sudden rises in CPU, disk, and memory utilization caused by the addition of new users to the system. These studies can be extremely detailed and can involve profiling the activities of specific users, enabling you to project rises in system utilization due to the addition of users.

Post-capacity planning studies offer other highly useful features in addition to predictive analysis, including the ability to project "what if" scenarios on workloads. Armed with data regarding how resources are used by the various types of users, you can also add specific types of users to the system workload scenario (such as accounts payable personnel) to predict exactly what kind of resource consumption would take place. This predictive analysis gives the system manager ample time to obtain the necessary hardware before the new users are added to the system, thus averting any degradation of system performance or response time.

Tuning information can also be obtained through post-capacity planning studies. Tuning information, such as information about disk I/Os going to the drive arrays that process queries, is derived from historic performance data and can be used to determine changes in the system configuration that are needed to increase performance. This information can show bottlenecks in performance such as too much activity to one drive array over another. For example, adding users will result in more database table accesses. The number of tables users access and how often they access them can be monitored and tracked. This information can be useful in determining whether the relocation of some of these tables will prevent a potential bottleneck in the disk subsystem.



Microsoft SQL Server 2000 Administrator's Companion
Microsoft SQL Server 2000 Administrators Companion
ISBN: B001HC0RPI
EAN: N/A
Year: 2005
Pages: 264

flylib.com © 2008-2017.
If you may any questions please contact us: flylib@qtcs.net