Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Download
Standard view
Full view
of .
Look up keyword
Like this
2Activity
0 of .
Results for:
No results containing your search query
P. 1
Effective Performance Management

Effective Performance Management

Ratings: (0)|Views: 73 |Likes:
Published by jmeisenb
A technical paper on performance management and capacity planning.

The paper focuses on transaction processing systems and models those systems to more effectively manage those systems that commercial capacity management tools.
A technical paper on performance management and capacity planning.

The paper focuses on transaction processing systems and models those systems to more effectively manage those systems that commercial capacity management tools.

More info:

Published by: jmeisenb on Apr 22, 2008
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as DOC, PDF, TXT or read online from Scribd
See more
See less

06/16/2009

pdf

text

original

 
Transaction Modeling for PerformanceTuning and Capacity Planning
John MeisenbacherIntroduction
Capacity planning and performance management are critical functions to properly sizelarge scale transaction processing systems. An oversized system will waste money andreduce profits. An undersized system will result in poor customer service and harm the business.This paper describes a straightforward effective method for capacity planning and performance management that does not require expensive commercial software. Thetechnique can be applied to production and test systems with minimal expense. Thisapproach has been successfully used in a production environment resulting in significantsavings during each of the past several years.
Background
Large scale global online transaction processing systems are becoming more common inrecent years. Such systems support the needs of merchants, consumers, governments,and corporations throughout the world. Many of these systems require the highest qualityand level of service 24 hours a day, 365 days a year. Such online systems are expected to process transactions in the shortest time practical. This need for high quality, reliability,and availability with high transactions rates place intense demands on these systems.Service levels suffer if these systems are undersized. However, oversized systems wastesresources that would be better used elsewhere.Some companies rely on stress testing to determine the correct size for their system. Thestress test involves sending test transactions near the maximum rate for the system andincreasing the rate until the system breaks. The analog is a racecar tested on the salt flatsat full speed until the engine blows up. This technique is useful to understand thetheoretical top speed. However, the results are not suited to forecast city and highwaydriving. Similarly, the company would typically spend significant effort and expense tocreate the correct test transaction mix and test environment. If the test transactions or theenvironment are not correctly configured then test results could be misleading.Other organizations use system monitoring software, such as Best1
, to identify the peak  periods and plan their capacity on that load. However, these products generally areunaware of the underlying transaction model that drives the system. Even when thisinformation is provided, the generic tool typically does not correlate the resource usageagainst the transactions.Best1 is a registered trademark of BMC Software, Inc
 
A simple low cost technique as an alternate to typical stress testing and commercialcapacity planning tools is available. The technique works in the test and productionenvironments. This approach can be applied of a wide range of transaction systems evenif the organization does not have access to a full scale performance test environment odedicated capacity planning personnel.The first part of this paper describes the technique in a production environment. A goalof this paper is to help organizations that are currently dealing with performance issuesfor their production systems. The paper will also describe how a test environment can beeffectively used to forecast production performance.
Transaction processing cycle – the curse and the key
Transaction processing systems typically have daily, weekly, and annual transactioncycles. The peak time of the day on the peak day of the week on the peak week of theyear is a common target for capacity planning.These transaction cycles of high and low demands require the system to be sized to meet peak demands. These systems tend to be underutilized during the lows in the cycle.Predicting these cycles and managing the systems to support the peaks is a major challenge. Fortunately, these cycles also provide the best way to understand the performance of the system.These transaction cycles provide the information necessary to create an accurate performance model of how the system acts across the full workload range. Transactions per second (TPS) is a convenient measure for transaction processing system workload. Adaily TPS cycle for a system can be seen in the graph below. The curves vary from dayto day, but generally follow more or less a bell shape. In most systems this informationcan be extracted from the transaction logs. If the production system must be used to pullthen information then this can be done during periods of low workloads to avoid impacton transaction processing.
 
 
Sample TPS patter for a typical dayTPS
   0  :   0   0   2  :   0   0  4  :   0   0  6  :   0   0   8  :   0   0   1   0  :   0   0   1   2  :   0   0   1  4  :   0   0   1  6  :   0   0   1   8  :   0   0   2   0  :   0   0   2   2  :   0   0
Local Time
   T   P   S
Figure 1 Daily transaction rates tend to follow a bell curve
The graphs presented in this paper are taken from production systems. The graphs have been generalized to remove sensitive information.This paper focuses on CPU performance. However, the approach can be used to modelnetwork bandwidth and system resources that vary with workload. The transaction dataand workload data, CPU utilization for example, are collected for the same time periodsand analyzed over one or more daily cycles.CPU costs can be obtained from capacity planning tools, such as Best1
 ®
, or from simpleoperating system utilities. This information is typically collected by the organizationanyway, so the collection process may not add any additional burden on the system.Most commercial tools tend to correlate CPU usage with time, as in the chart above. Thefollowing chart overlays the previously diagramed TPS rates on the right Y-axis withCPU utilization on the left Y-axis.

Activity (2)

You've already reviewed this. Edit your review.
1 thousand reads
1 hundred reads

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->