Top 10 Myths of Performance Management and Capacity Planning
Performance management can provide return on investment for your organization, however there are many misconceptions that can hinder your efforts. In this this post, we’ll debunk several popular myths about performance management and capacity planning.
Performance management is the computer-assisted control of all aspects of the performance of a computer system. Along with other functions, such as resource management, event management and security management, it forms a key part of system management.
Performance management embraces activities such as system measurement, monitoring, analysis, resource accounting, tuning and optimization. It is a continual process in any well-run data center and it can only be avoided by having a totally static workload, or a grossly over-configured machine – both a rare circumstance.
It is important to understand that performance management is split into two branches:
- Performance analysis, which includes monitoring, tuning and optimization
- Capacity planning, which embraces capacity analysis
The level of detail required to carry out successful planning may be different from that required to solve a performance problem.
Ten myths of performance management and capacity planning
There are ten common assumptions made about performance management and capacity planning that are examined below, as well as real-world best practice to counter each.
1. I don’t need to do performance management until I have a problem
REALITY: It’s better to be proactive by predicting the problem and implementing
The basic goal of performance management is to anticipate resource requirements, identify difficulties while they are still potential problems and implement the appropriate solution before a failure occurs.
2. Hardware is so cheap that performance management and capacity planning are unnecessary
REALITY: Total IT expenditure is still rising, so there is increased cost justification leverage
Hardware price performance is improving at around 40% annually. The demand for IT resources is increasing at around 60% annually. Furthermore, although each component may have a lower unit price, IT systems are increasing in complexity and size to the point where informal and ad hoc planning methods are totally inadequate.
3. Performance management only needs to be done once a year
REALITY: It is best done as a continuous process of measuring, analyzing, predicting and tracking
Performance management and capacity planning are continual processes. The projection of future requirements is based on monitoring and analysis that take place regularly during development and production. If you don’t understand both the current situation and how it developed, you will not be able to forecast your future requirements. Without continuity, performance becomes event driven.
4. Real time monitoring, tuning and optimization is all you need
REALITY: You need a combination of detailed monitoring for analysis and aggregation for planning
The level of real time that is applicable to systems depends on their timeframe. Many commercial systems can be managed on five-minute snapshots so long as the system is seen as of five minutes ago at any time on a browser. Finer granularity for the purposes of cockpit style displays are not effective for solving real problems where patterns of behavior of large populations of users are involved.
5. I can’t do performance management until I have tuned my system
REALITY: The two processes work best together
There is no such thing as a fully tuned system. Fortunately, the Performance Management process highlights system bottlenecks and simplifies the choice of remedial actions. A good Performance Management model will allow you to identify the benefits to be obtained from tuning actions. Clearly, the Performance Management model is more reliable once the worst excesses of poor tuning are removed, but such a model will nonetheless highlight the unnecessary bottlenecks.
6. Management reporting on a regular basis takes up too much time
REALITY: With the right tools, this can be automated
The need to provide regular management reports increases with business criticality of systems. Rather than issue piles of paper, or stick colored plots on the wall, most sites now want automated dynamic reports showing the status of any node using a browser.
7. Analysis and interpretation of performance reports is too complex
REALITY: Automatic advice and exception reporting makes the data easy to understand
The growth of distributed systems with large numbers of nodes requires that management reports are exception-based and can also incorporate some intelligent interpretation automatically.
8. Money is wasted on redundant or irrelevant equipment
REALITY: Excessive spending remains undetected until measured by capacity planning
The solution lies in establishing and maintaining a well-controlled and timely procurement plan to the benefit of the enterprise. The alternative is the problem of trying to resolve performance panics and shorten procurement cycles.
9. Network capacity plans need too much time to define traffic and workloads
REALITY: Networks are readily incremented and utilization can be assessed by bandwidth
The optimal solution lies in automatic collection of a standard set of performance data (not yet defined) by the network manager for input to new tools and effective planning of networks.
Network capacity planning tools model the behavior of a network to predict the delay in response time due to the network. Network capacity planning has traditionally been a discipline that requires significant effort to collect traffic statistics and relate it to workloads.
Typical utilizations on networks were low when most communications were point-to-point and in batches. The emergence of e-commerce, electronic mail, image processing, the increased distribution of computers and the introduction of graphical user interfaces (all of which increase traffic), have all led to concerns about network saturation. The major cost of implementing network planning tools lies in the expertise required to use them and the time required to characterize the workload traffic.
So far, this has been seen only worthwhile for major networks where formal service level agreements exist. However, increasing network loadings and the close integration of networks and processing nodes — for example in client/server systems — may force a reappraisal of attitudes. A new generation of network management tools and planning tools that integrate networks and computers in a single model may be required.
Although the incremental cost of upgrading networks may be less than that for mainframes and the procurement cycle shorter, the inter-dependence of networks and open systems will lead to greater adoption of such packages.
10. Service level agreements (SLAs) need too much effort to establish and track
REALITY: SLAs are only agreed to when the service provider knows he’s safe
Service quality has several dimensions, including functionality, ease of use, performance availability and reliability. Traditionally IT managers have concentrated on functionality, usability and reliability and have assumed that performance is something that the technicians will put right once they have made the system work.
The results of this approach are clear from a quick scan of the computer press: applications have been delayed or abandoned because they couldn’t be made to perform, and computers and networks have required unscheduled upgrades to handle their work. Although this policy may have sufficed when IT systems were relatively simple, it is fraught with danger as systems have grown significantly in size, complexity and criticality. The importance of defining and managing service levels effectively is now widely recognized, especially in terms of e-commerce.
Often when these myths are believed to be true, businesses need to overcome unnecessary obstacles to then establish the need for a continuous performance management and capacity planning model. Taking the time to develop a thoughtful approach to establishing your model will ultimately give you a continuous update on the understanding of your system, reliable and automatic reporting, and help control spending.