Please note: Downloads are available to CMG members only. Click here to learn more
Methodologies for Selecting and Managing IT Performance Measures
Michael Bitterman
The IT revolution has contributed to a business culture focused on productivity,measurability of systems and a decrease in defects, all aligning to enterprise strategic objectives. Today, a corporation with ill-defined direction in its strategic performance methodology is a company destined for failure. Understand the important role a structured, repeatable methodology plays in the management of your IT performance measures. This session will give your organization direction in choosing and applying performance measures with proven methodology.
Download paper (Available to CMG members only, Click here to learn more.)
Has the IT Benchmark Become Obsolete?
Michael Bitterman
Is the cost focused IT benchmark obsolete? Does the user’s concept of IT as a commodity mean IT has been devalued? The current state of IT benchmarking - What are the different types of benchmarks? - What is the true cost of benchmarking? - What kind of results can you expect? The next question is whether an IT benchmark makes sense for your IT organization. - Decide if you want to propose a benchmark for your organization - How long will it take to complete a benchmark? - What is the investment in resources and dollars? - What benefits or real value should you receive?
Download paper (Available to CMG members only, Click here to learn more.)
Objective Functions for SLA Risk Management
Dr. James Bouhana
Mike Tsykin
Effectively managing compliance with Service Level Agreements requires ongoing assessment of progress to date. We discuss how to use interim measurements for deriving bounds on SLA metrics for the full interval spanned by the SLA. We illustrate the concept of SLA objective functions, which yield the future SLA levels required for full interval compliance. Proactive risk reduction actions are also discussed for increasing the likelihood that a SLA objective will be met. System availability is used as an example, although the techniques can be applied to other SLA metrics as well.
Download paper (Available to CMG members only, Click here to learn more.)
Future Big Iron Performance and Capacity Specialists – Where will they come from?
Gregory V. Caliri
In 2007, there are still a number of computer performance evaluation specialists in the z/OS arena. However, with their aging, a crisis is brewing. Even fewer may be needed in the future; nevertheless, there may not be a sufficient number of knowledgeable individuals to continue the practice a decade from now. Current CPE professionals may have to assume a mentoring role, attempt to influence the college/university level graduate to pick up the mainframe responsibilities.
Download paper (Available to CMG members only, Click here to learn more.)
Application View Reporting
Jane G. Fiala-Curry
Barbara L.. Johnson
Application View Reporting provides a cross-platform view of an application. The on-line report correlates distributed and mainframe resource consumption metrics with JVM statistics (garbage collection events), and web usage statistics (page views) over hours, days and weeks. AVR presents clear patterns of consumption and how they correlate with usage. This is extremely helpful in stages of phased rollouts.
Download paper (Available to CMG members only, Click here to learn more.)
Who Measures the Measurers? KPIs and QoS for the Capacity Management Process.
Adam Grummitt
This session addresses issues arising from establishing an effective monitoring and reporting regime for the Capacity Management process itself. Capacity Management practitioners collect metrics and provide their analysis and interpretation to underpin all the ITSM processes. The challenge often lies in gaining an understanding of business drivers for a service workload and identifying KPIs for the quality of the service. But few in Capacity Management then address the same challenge in applying the same techniques to their own activities. This session discusses experiences of some who do.
Download paper (Available to CMG members only, Click here to learn more.)
UK CMG - Best Paper: Underneath the Spin; a Practical Look at Service Levels (Is what you see what you get?)
Malcolm Gunn
The setting measuring and reporting of service levels appears easy but getting those levels to accurately reflect the flexible requirements of the client is something that seems almost alien to Service Managers. Enev the basics can be a challenge; figures are often based around simple easy to report measures. Once the figures are available oganizations try and use the wrong figures for the wrong measure, Incident volumes to show of availability. People forget to look beyond the headline just because something is available doesn't mean it's usable. Make sure the levels agreed are flexible to meet the peaks and troughs in demand and are set at levels that allow the client to do their job. This session takes a practical look behind the headlines at service level and dashboard reporting, using real life examples it shows how things may not be all they seem at first glance.
Download paper (Available to CMG members only, Click here to learn more.)
To Instrument or Not Instrument, That is the Question
James Holtman
In reviewing many of the CMG papers, there is a lot of space devoted to analyzing data, but not much space devoted to where this data came from. Now the standard performance metrics that your favorite operating system collects will provide a high level view as to what is happening. To get a more detailed view, it will be required that the application have ''hooks'' to gather application specific data. This session will provide an overview of how to instrument an application and how to analyze the data collected. The reader can then extend it to his/her specific environment.
Download paper (Available to CMG members only, Click here to learn more.)
The Service-Oriented Organization - People, Politics & Pitfalls
Denise P. Kalm
As Garrett Hardin noted in Tragedy of the Commons, people value more what they own than property held in common. People litter in public parks, when they would never consider messing up their own yards. But in a SOA-enabled enterprise, services and servers are a common good; they are not “owned” by a line of business or a technology team. One can solve the technology challenges of SOA and still fail because the organization is not in sync with SOA design; organizational dynamics matter. Learn approaches to move to a Service-Oriented Organization (SOO) to reap SOA success.
Download paper (Available to CMG members only, Click here to learn more.)
Managing and Modernizing Legacy Applications
Gerry W. Leitao
Most IT organizations treat legacy applications as overhead rather than as valuable business assets. They spend considerable sums of money operating and maintaining these applications yet invest little in improving their returns to the business. This executive report examines how to evaluate and improve the value of legacy application portfolios, reviews the range of options available to modernize and manage legacy assets, and demonstrates the value of proactive application management.
Download paper (Available to CMG members only, Click here to learn more.)
Using a CDB to Uplevel Your Performance Management and Planning Efforts
G Jay Lipovich
Computer Performance and Planning analysts have historically used a Performance Data Dase (PDB) as the basis for performance reporting and planning. Expanding the PDB to become an ITIL-defined Capacity Management Data Base (CDB) will enable analysts to raise the maturity of their efforts to add more business relevance to performance management. This session discusses the CDB requirements and identifies traditional and non-traditional disciplines in which CDB-enabled performance data can add significant value.
Download paper (Available to CMG members only, Click here to learn more.)
Performance of the Business of a Web Portal
Mark M. Maccabee
Web sites, called portals, became to a degree the face of the business. Portal is a program and one its characteristics is performance. Thus performance supports directly the business itself. In this session we present a portal that is used by different groups of people. We explain how the portal architecture and implementation reflects the requirements of those groups. We then present the performance measurements to establish the actual performance of each user group. And finally we will list the performance results, the tuning of the different parts of the portal and the resulting performance.
Download paper (Available to CMG members only, Click here to learn more.)
Everything You Know About Monitoring is Wrong
Mazda T. Marvasti
An enterprise collects data using many monitoring tools. Some collect numerous metrics; others use metrics to diagnose an issue after IT staff has been informed of a problem. Others analyze metrics to predict problems before they occur. With a shrinking IT budget and “do more with less” directives, IT professionals face a conundrum: what metrics should be collected to ensure the availability of critical business services? A metrics-collection approach review shows that collecting 40% of available metrics eliminates 94% of the most common problems, revealing a rapid declining ROI.
Download paper (Available to CMG members only, Click here to learn more.)
Adding Business Drivers to the Capacity Planning Process
Linwood Merritt
Gil Reamy
A company may have the best capacity planning tracking and forecasting methodologies and still miss the mark due to a lack of connection with business planning. A dialog between a company’s IT and business areas should identify business drivers and combine the trending of resources with business-oriented tracking and forecasting. This session discusses issues surrounding business-oriented capacity planning and steps that companies can take to incorporate business drivers into the capacity planning process.
Download paper (Available to CMG members only, Click here to learn more.)
Generation of Test Schedules for Object Oriented Applications
Tuli Nivas
In this session, we present a new method for generating test schedules of OO programs, using the COCOMO model and OOTWorks - an object oriented testing and maintenance tool. This information is displayed in the form of PERT and Gantt charts for visual clarity. These help a testing manager effectively decide the total amount of effort required for testing each class in terms of person-month,as well as knowing the total time required to test the entire project. We have validated this method with real software projects and compared the estimated values with the actual results.
Download paper (Available to CMG members only, Click here to learn more.)
The Experience of System Responsiveness
Dr Steven C. Seow
There are a handful of industry standards that speak to responsiveness in human-computer interaction. A severe limitation of these guidelines is that they quickly become outdated when technological advancement introduces newer and better forms of interaction. It is argued that the root of this limitation is the direct association of technology to metric. This session proposes a simple user-centric as opposed to a technology-centric framework to define responsiveness. Specifically, responsiveness is defined as what users expect as instantaneous, immediate, continuous, and captive.
Download paper (Available to CMG members only, Click here to learn more.)
Building a Capacity Management Team – Lessons Learned
Mary Anne Smith
What happens when a technical systems engineer moves from a performance focus to a Capacity and Performance Management (CPM) effort? I quickly learned the transformation brought unique challenges and required detailed research and training. This session details the successful launch of a CPM team including: team development, process creation, template documentation, criteria for vendor and tool selection. It also includes information on critical issues and key learnings.
Download paper (Available to CMG members only, Click here to learn more.)
Performance Testing: A Heuristic Approach
Scott Barber
Only rarely are sufficient time, resources and skilled individuals available to teams developing commercially driven software systems to effectively apply rigorous approaches to performance test the system. In these cases, it is critical to have a risk-based, flexible approach to collecting the data required to assist the development team in identifying and mitigating risks around areas of the application that are performing sub-optimally and to assist stakeholders in making sound business decisions related to performance risks. This session briefly outlines a heuristic approach to performance testing with a track record of success when applied on projects where more rigorous approaches are unlikely to be effective in adequately mitigating business risk in time to keep pace with the commercial aspects of the project.
Download paper (Available to CMG members only, Click here to learn more.)
Automating the Analysis of Load Test Results to Assess the Scalability and Stability of a Component-
Dr. Andre B. Bondi
Thorough load testing of systems involves test runs of use cases and usage scenarios at multiple load levels. The number of use cases can be so large that timely manual data analysis is simply not possible. We present a computationally inexpensive method of automatically eliminating the warm-up and cool-down transients from measurements of stable systems. In addition, the method enables the automated detection of possible system instability. It can be used to analyse performance data obtained using commonly available system measurement tools.
Download paper (Available to CMG members only, Click here to learn more.)
Improving Simulation Accuracy through the use of Synthetic Alignment Intervals
Dr. Jeffrey P. Buzen
Monte Carlo simulations are used in many disciplines to evaluate the steady state distributions of stochastic models. This session introduces a new procedure for improving the accuracy of such simulations. The approach is based on constructing synthetic “alignment intervals” that are appended to the output of the original simulations, creating extended simulations whose output conforms to certain mathematical relationships. Satisfying these relationships is shown to be sufficient to guarantee that the underlying steady state distributions have been computed accurately.
Download paper (Available to CMG members only, Click here to learn more.)
Help! Flow Control Issues for B2B Gateway Solved by Rapid Turnaround of Designs using Simulation
Nick Chapman
BT chose Zeus Extensible Traffic Manager (ZXTM) for flow control into a B2B gateway. Additional requirements included: • Limit users to agreed volumes under overload conditions… • …but allow full capacity to be used otherwise • Reject messages according to priority scheme • “Rogue” users must not adversely affect “good” users Early testing indicated difficulties in meeting all the requirements. Simulation modelling, using Hyperformix Workbench, identified issues with the initial design but further modelling (with rapid turnaround) enabled alternative solutions to be designed and examined.
Download paper (Available to CMG members only, Click here to learn more.)
How to Select Significant Workloads in Performance Models
Paolo Cremonesi
Paolo Cremonesi
Giuliano Casale
The complexity of computer systems requires us to consider the interaction of several workloads. Only a limited number of business and technical workloads are usually required to properly model the system. In this session, we discuss regression-based estimates of service times required for model parametrization and we focus on the selection of significant workloads. We present an experimental comparison, using real performance logs of a distributed enterprise application, illustrating the benefits of constrained estimations over the traditional approach based on ordinary linear regression.
Download paper (Available to CMG members only, Click here to learn more.)
Performance Optimization on Game Consoles
Bruce Dawson
Console games are the ultimate real-time programming challenge. A delay of just a few milliseconds can mean the difference between a good game experience and a great game. Console game development is definitely not all fun and games. It requires pushing the hardware to its limit. This requires using sophisticated programming techniques and tools that can precisely measure performance and identify bottlenecks. Adding to the challenge is the customer expectation that each year’s games will be better than last year’s, even when the hardware is exactly the same. This session gives an overview of game performance tools and techniques used to push Xbox 360 CPU and graphics chip to their limits, all in the name of making games that are spectacular to watch and fun to play.
Download paper (Available to CMG members only, Click here to learn more.)
ARM: Getting Started with Application Response Measurement (ARM) and Java the Easy Way
Carl De Pasquale, PhD
To implement Application Response Measurement (ARM), several ARM API's, including ARM start with correlation and stop must be coded into the candidate Java application. Since developers are principally concerned with providing end user functionality, spending time implementing non-functional requirements, such as ARM, is secondary. To address this issue, this paper describes an approach that uses dynamic byte code instrumentation, Byte Code Engineering Library (BCEL), and the ARM SDK 4.0 to ARM Java applications without developer Intervention.
Download paper (Available to CMG members only, Click here to learn more.)
Moore’s Law vs. Gates’ Law -- Macro Capacity Management
Dr. Yiping Ding
The mission of ITIL Capacity Management is to ensure that the current and future capacity of IT infrastructure is provided to meet business application requirements at an acceptable cost. Traditional capacity management tools focus on the performance, capacity, and cost of the IT infrastructure within an enterprise. In this session, we examine hardware and software trends at the macro level and provide insight for long-term capacity planning beyond the local system and application environment. We introduce a performance model for Moore’s Law vs. Gates’ Law to quantify the study.
Download paper (Available to CMG members only, Click here to learn more.)
Comparative Architecture Performance Analysis At Design Time
Gerald S. Doyle
Elizabeth L. White
The ability to evaluate the performance of a software system's architecture early in the design lifecycle lessens the risks associated with choosing a particular architecture. Unfortunately, such early evaluation is difficult due to incomplete information. Our focus has been on techniques that provide comparative performance information regarding various architectural choices, rather than absolute information. In this session, we demonstrate our method on a small web-based example.
Download paper (Available to CMG members only, Click here to learn more.)
Modeling the Performance Impact of Internet Marketing Campaigns on E-Commerce Sites
Dr. Pierre M. Fiorini
Yiping Ding
Internet Marketing (IM) is the practice of applying advertising techniques to drive traffic to e-commerce sites. The business community has developed techniques that generates traffic to e-commerce sites, but little work has been done to develop analytic models that assess the performance of e-commerce sites resulting from Internet Marketing Campaigns (IMC). In this session, we develop analytic models to assess the performance impact of IMC’s upon web sites, which can then be used to address capacity planning issues.
Download paper (Available to CMG members only, Click here to learn more.)
The Art and Science of Measurement
Mark B. Friedman
This basic tutorial in the CMG-T foundation curriculum introduces the measurement techniques that are employed in monitoring computer performance. It focuses on the two major types of measurement procedures - interval-based sampling and event-driven measurements. It discusses the relative advantages and disadvantages of these two approaches, comparing and contrasting them in terms of their accuracy, validity, and reliability. The clock and timer facilities that interval-based sampling measurements rely upon are also described. Common measurement anomalies including missing or incomplete data capture are also discussed, along with overhead considerations. The class draws heavily on examples from the IBM mainframe hardware and software environment as well as Intel server hardware and the Windows operating system.
Download paper (Available to CMG members only, Click here to learn more.)
Capacity Planning Boot Camp
Dr. Neil J. Gunther
What is capacity planning? How does it differ from performance tuning? How do I get started? If you've been asking yourself these questions, then this CMG-T course is for you. As the name implies, capacity planning for computer systems is about predicting the future. Financial planners do that all the time so, not surprisingly, many of the tools and techniques are similar. The difference lies in the data to be analyzed and the metrics used to express computer system performance rather than financial performance. And just like today's fast-paced business climate, IT decisions are made and revised so rapidly that merely providing your management with a sense of planning direction is often more important than calculating the compass bearing. Elsewhere, I have called this kind of tactical planning, *Guerrilla Capacity Planning*. This boot camp course will get you in shape in three sessions.
Download paper (Available to CMG members only, Click here to learn more.)
A Quantitative Approach for Determining Service Demands in a Network Load Balanced Scenario
Vikas Gupta
Sachin Wagh
Meeting rigorous performance expectations, dictated by extremely competitive online business, requires that predictive performance models be built for deployed applications. However, deriving such models from complicated test-bed setups, like the clustered environments used nowadays is a vexing exercise; end-to-end performance attributes given by the load testing tools are insufficient to compute box-level service time parameters. This session presents an approach to determine the service demands for each of the server machines in a clustered environment.
Download paper (Available to CMG members only, Click here to learn more.)
Engineering Performance Using Control Theory
Dr. Joseph Hellerstein
Yixin Diao
Performance engineering can be viewed as a regulation problem in which performance targets are held within certain bounds. An example is ensuring service levels for interactive queries in a database. Other engineering disciplines, such as electrical, mechanical, and aeronautic engineering, use control theory to do such regulation. This session provides a brief introduction to control theory, and the session describes the application of control theory to managing the performance of IBM’s Lotus Notes email server and IBM’s DB2 database management system.
Download paper (Available to CMG members only, Click here to learn more.)
An Approach for Accurately Recreating Web Workloads from Production Data
Suresh K. Khemka
Nikhil Venugopal
Gaurav Caprihan
For web applications, major performance problems in production can only be corrected by recreating the production scenario in a controlled environment and arrive at the solution through performance testing and analysis. The key to this approach lies in accurately identifying parameters that can help recreate the production workload for load testing. The current practice for specifying these metrics is based on “guestimates” or derived from heuristics or rules of thumb. This session presents a methodology that removes this subjectivity by extracting these metrics from web access log files.
Download paper (Available to CMG members only, Click here to learn more.)
So I have Data on Each Transaction, How Do I Use It?
Luke E. Lofgren
Once a home-grown application logs information on each transaction, how do you make effective use of the data? Only a few data elements are needed and can be turned into a rich amount of capacity and performance management metrics and graphs. The focus of the tutorial will be on what needs to be included in summarization of the data independent of method. Tools used to illustrate the process will include no-cost, common desktop applications, or open source. Specific real-world examples of testing and usage of an internal application spanning over 300 heterogeneous systems will be provided.
Download paper (Available to CMG members only, Click here to learn more.)
An Empirical Evaluation of Communication and Processing Overhead in Service Oriented Architectures
Daniel A. Menasce
Most modern applications are networked and make heavy use of various communication protocols such as TCP/IP and SOAP over HTTP in case of Web Services. This session presents the results of an experimental evaluation of the communication and inherent processing overhead of these two protocols. Experiments were carried out to determine the impact of the size of reply message as well as the complexity of the SOAP reply message in terms of number of elements. Analysis of variance was used to determine the statistical significance of the differences obtained.
Download paper (Available to CMG members only, Click here to learn more.)
Why You Can't See Your Real Performance Problems
Cary Millsap
Reflecting across nearly 20 years of solving Oracle performance problems, I can recognize a single pattern of behavior that is the dominant reason for failure in all the projects I've witnessed. In almost every case I've seen, failures in diagnosing and repairing performance problems have been caused by unrecognized SKEW in diagnostic data. This presentation shows several examples that illustrate why skew is such a pervasive problem for performance analysts.
Download paper (Available to CMG members only, Click here to learn more.)
Using Simulation to Forecast Performance: A Case Study
Tuli Nivas
Performance is a key factor for usefulness and acceptance in business, especially when it comes to large scale enterprise systems. This session will discuss some of the issues that should be kept in mind while designing multi-tiered, multi-threaded systems using simulation. It will take into consideration models using queues to transfer requests between the tiers and also see how other application operations can be made optimized in large scale enterprise applications.
Download paper (Available to CMG members only, Click here to learn more.)
Software Performance Lifecycle at a Large National Bank
Amit Patel
Performance assurance plays a key role in today's complex applications and is an essential element of the application development life cycle. This presentation discusses a case study about integrating performance at a large national bank. Learn how custom monitoring, Six Sigma techniques, and daily production reports played an important role in identifying production issues. This presentation illustrates and examines the challenges and successes of performance planning, testing, analysis, and optimization after the release of ABC Bank’s CRM application.
Download paper (Available to CMG members only, Click here to learn more.)
Multiple Dimensions of Performance Requirements
Alexander Podelko
From the first look, the subject looks simple enough. Almost every book about performance or requirements has a few pages about performance requirements. There are many papers discussing some aspects of performance requirements. Quite often there is a performance requirements section in the requirements documents. But the more you look into performance requirements, the more issues you see. The goal of this session is not to give answers to all questions, but rather to list these questions in one place and illustrate them by different views and examples when possible.
Download paper (Available to CMG members only, Click here to learn more.)
Using SAS/Graph to Display Performance and Capacity Data
Richard S. Ralston
This presentation shows a few useful tips and information for devloping graphs using SAS/Graph. The presentationthen focuses on building and using heat charts to display performance and capacity data in an interesting and usable manner. Heat charts provide the ability to visualize 3 dimensional data in 2 dimensions. The techniques used are platform independent, although the work is done on the z/OS platform.
Download paper (Available to CMG members only, Click here to learn more.)
Modeling and Forecasting
Dr Michael A. Salsburg
Although most computing environments are heterogeneous, computer system modeling is, in most ways, platform neutral. The same techniques and tools can be used to model zSeries, Unix / Linux, and Windows. At the heart of these models is the essential queueing network. This course provides the details of the essential queueing network, including the necessary statistics that need to be collected from the system, as well as various modeling techniques that yield insights that cannot be gleaned from observing the actual computer system. Once the model is validated, it can be used to explore “what-if” scenarios where either the workload or the underlying configuration can be changed in the model so that the resulting service levels can be observed. If time permits, an additional section on the subject of time series estimation and forecasting will be presented. This course will not teach you everything you need, but it will give you a survey of the various approaches with a full bibliography for future reference
Download paper (Available to CMG members only, Click here to learn more.)
The Pros and Cons of Collecting Performance Data using Agentless technology
Dima Seliverstov
John Tavares
Tianxiang Zhang
Agentless data collection is a powerful technology which has its advantages and disadvantages. In this session we will discuss the variety of Agentless implementations for Windows and UNIX. The implementations considered are SNMP (Simple Network Management Protocol), WMI (Windows Management Interface),Windows Remote Registry, WBEM(Web Based Enterprise Management. We will discuss the system performance metrics availability, security,network issues and when the agentless monitoring component is installed as part of the operating system. We will discuss our real world experience with Windows.
Download paper (Available to CMG members only, Click here to learn more.)
Top 10 Ways to Kill an SPE Initiative
Dr. Connie U. Smith
Lloyd G. Williams
Software performance engineering (SPE) is a method for constructing software systems to satisfy performance requirements. The success of SPE in delivering systems that meet performance requirements on time and within budget is well documented. We have observed many organizations that have tried to adopt SPE. Some were successful; others were not. From our experience, we have compiled our candidates for the top ten ways to ensure an unsuccessful SPE initiative. We also discuss consequences and solutions. At the end, we’ll discuss some conclusions and how to keep from falling into these traps.
Download paper (Available to CMG members only, Click here to learn more.)
Performance Management – Top 10 Traps
Megh Thakkar
In our quest for achieving optimal system performance, we tend to follow several methods and techniques. Some of these techniques are empirical in nature and use a scientific approach while others are artistic in nature and rely on rule-of-thumbs and past experience. This presentation will discuss some common traps that are experienced while managing the performance of our systems regardless of the technique used. Tips to recognize and avoid these traps are discussed so that the desired results can be achieved.
Download paper (Available to CMG members only, Click here to learn more.)
System Management by Exception, Part Final
Dr. Igor A. Trubin
Ray White
Statistical Exception Detection System (SEDS) has been successfully used for more than seven years to automatically produce web-based exception reports and smart alerts against the performance data warehouse for a large, multi-platform environment. This session starts with an overview of how SEDS uses SPC and MASF techniques and how SEDS could be used as a part of Lean/Six Sigma. Then it focuses on the memory usage exceptions that SEDS captures to proactively identify server and application performance issues.
Download paper (Available to CMG members only, Click here to learn more.)
Performance Monitoring Process for Out of Standard Applications
John S. Van Wagenen
At Caterpillar we call this the out of standard process. By identifying our applications and tracking monthly cpu utilization, we calculate the expected performance baseline. Each month all applications are measured for current activity and compared to the baseline. Those applications that exceed a specified threshold of variance are designated as “out of standard” or OOS for short. The OOS applications are scrutinized for a named cause of the variance. This process assigns one of three dispositions to the OOS application :anomaly, growth or tuning.
Download paper (Available to CMG members only, Click here to learn more.)
Achieving near Linear Scalability Using Solaris on Numa Architectures
Rickey C. Weisner
Many system architectures have NUMA characteristics. Some applications have difficulty scaling on such architectures when CPUs are added. In this session I discuss how we dramatically increased performance on a Sun E2900 by reducing the effect of the NUMAness of the machine. I will also discuss how similar results were achieved with a G4600 for a different application. In the session I will discuss what has been characterized as the scientific method to perform performance analysis, the importance of the business metric, and give an overview of several NUMA tools from OpenSolaris.org.
Download paper (Available to CMG members only, Click here to learn more.)
Statistics for Performance Analysis & Capacity Planning
Ray Wicks
This session reviews some of the statistical techniques which can be useful in performance analysis and capacity planning. The introduction of basic statistical concepts will emphasize the relationship between what you see in a graph and the statistical formulae. Topics will include the following: • Descriptive Statistics: • The Basics: average, variance, coefficient of variation, standard deviation, etc. • Graphic Techniques: linear plots, distributions, histogramme, box plot, etc. • T-test : comparing averages. • Predictive Statistics: • Linear regression, multivariate regression as an approximation for Time Series Analysis. The selection of a single number to represent the behavior of a variable, given in a set of values, requires some statement about the representative-ness of the single number. This examination looks at the statistical value of your input and output as seen in any modeling process. Excel implementations will be discussed.
Download paper (Available to CMG members only, Click here to learn more.)
Software Performance Engineering: A Tutorial Introduction
Dr. Lloyd G. Williams
Connie U. Smith
It should come as no surprise that, when it comes to performance, the software industry is in a pretty sorry state. Many software systems must go through an expensive and time-consuming tuning process before they can be used. Others must simply be abandoned. This tutorial presents a systematic, quantitative approach for cost-effectively building performance into software systems. It provides an overview of Software Performance Engineering (SPE) and illustrates the steps in the SPE process. A case study illustrates the SPE models and their solution.
Download paper (Available to CMG members only, Click here to learn more.)
Improving Packing Algorithms for Server Consolidation
Yasuhiro Ajiro
Atsuhiro Tanaka
Minimizing the number of servers and preparing sufficient resources are contradictory requirements in server consolidation. These requirements can be formalized as a variant of the bin packing problem, which is known to be NP-hard. The least-loaded, a load-balancing algorithm, is applied to the problem and compared to the classical First-Fit Decreasing algorithm designed to address the bin packing problem. Our technique of improving the algorithms is then presented.
Download paper (Available to CMG members only, Click here to learn more.)
Late Breaking: Leveraging the RMF Distributed Dataserver for Real-Time Performance Intelligence
Scott Chapman
The RMF Distributed Dataserver (DDS) emits XML in response to HTTP requests. The author built a custom browser-based real time dashboard that combines that RMF-provided data with historical information to provide a holistic view of current mainframe performance. This new tool has greatly improved the operations staff's ability to detect and respond to performance anomalies. Best of all, this was all done cheaply--without purchasing any additional software! This paper will discuss the design and implementation details of this dashboard with the hope of inspiring others to explore their own potential uses for the DDS. Sample code will be provided to illustrate retrieving data from the DDS.
Download paper (Available to CMG members only, Click here to learn more.)
Millicomputing: The Coolest Computers and the Flashiest Storage
Adrian Cockcroft
Power consumption is a hot topic, but using CPU designs from the world of battery powered devices and flash memory based storage we can make cool systems. A millicomputer uses less than one watt, so its power is specified in milliwatts. Enterprise millicomputer arrays provide large numbers of small computing units at a total cost, performance and power level that redefines the limits of what is possible. These systems are being constructed as open source hardware by their end users. The architecture and performance characteristics are described in this session.
Download paper (Available to CMG members only, Click here to learn more.)
Directing the Stream: Using Web 2.0 Technologies for Marketing and Performance
Dr. Bernard Domanski
Rob Domanski
Objectives: to sort out various Web 2.0 technologies, to explore assorted tools available to the average Web user, and to show how to create a blog and an RSS feed. Ultimately we’ll show how to harness the power of Web 2.0 for effective Internet marketing / publicizing strategies. To the performance reader, we explore several performance blogs via Web 2.0 sites. Through these and other sites, we hope that the performance community takes even greater advantage of the “community knowledge sharing” that Web 2.0 technologies provide.
Download paper (Available to CMG members only, Click here to learn more.)
The Right Mix of Utility, Consolidation and Virtualization for Optimum Cost-Capacity-Performance.
Adam Grummitt
Utility computing and consolidation have been around for some time. They have been accelerated by the use of virtualization to ease the combination of applications on lightly used servers. The question is how to find the right mix of these interwoven approaches at a given site. This session describes popular hypervisor and partitioning options and other aspects of abstraction and virtualization. It reviews the factors involved, and outlines criteria for the capacity manager, in determining the appropriate mix for selected services to achieve the optimum cost-capacity-performance solution.
Download paper (Available to CMG members only, Click here to learn more.)
Parallel File System Technologies in a Linux Cluster and GRID Environment
Dr. Dominique A. Heger
Local file systems support a persistent name space and view devices as being locally attached. Contemporary interconnects allow multiple cluster nodes to share storage devices. IBM’s GPFS or Red Hat’s GFS parallel file systems take a shared, network-attached storage approach. These file systems are built on the premise that a shared-disk file system has to exist within the context of a cluster or GRID. This session elaborates on the file system designs, focusing on scalability, availability, and performance. Analytical models are used to discuss centralized verses distributed locking scenarios.
Download paper (Available to CMG members only, Click here to learn more.)
Seeing It All at Once with Barry
Dr. Mario Jauvin
Neil J. Gunther
Improving visualization of performance data is an orphaned area of performance tool development. Tool vendors avoid investing in development if they see no demand, while planners and analysts do not demand what they do not know. We attempt to cut this Gordian knot with 'Barry' a 3-d visualization tool based on barycentric coordinates. Potentially hundreds of computing nodes can be viewed as an animated cloud of points whose shape is correlated with the workload dynamics. 'Barry' provides an optimal impedance match between the measured computer and the cognitive computer (the analystÕs brain).
Download paper (Available to CMG members only, Click here to learn more.)
Death to Dashboards: Alarming, Performance Management Based on Variance, System Prioritization and Other Thoughts on Data Visualization
Peg McMahon
Justin Martin
When does the light on the executive dashboard turn from yellow to red? When do you order new hardware? Traditionally, these decisions are handled by setting thresholds — picking some number to use as an upper or lower limit. Thresholds might have worked well in the days of a handful of beloved systems. But for today’s complex environments, thresholding is not only painful to manage but conceptually bankrupt. Let’s talk about the problems with thresholds and dashboards and work to identify some practical alternatives. Vendors, put on your iron underwear and attend this session.
Download paper (Available to CMG members only, Click here to learn more.)
The ""Powerful"" Capacity Manager
Chris Molloy
Recent studies have indicated that servers in US data centers consume over one percent of the electrical power generated in the US. With increased demand for IT applications, many data centers are not prepared for the growth. This session analyzes the characteristics associated with IT electrical power, and how the capacity planner is uniquely positioned to apply their process, procedures, and products previously used on IT resources to data center resources. It also discusses technologies that the capacity manager may already be advocating to mitigate the projected data center constraints.
Download paper (Available to CMG members only, Click here to learn more.)
Late Breaking: Mainframe Virtualization: The New Alternative
Chris Molloy
On August 1, 2007, IBM announced that it was moving the workload from 3900 distributed servers to 30 mainframe servers, an overall ratio of 130 virtual servers to one physical server using its Linux on z implementation. This paper analyzes several case studies of companies that have implemented mainframe virtualization and provides additional specifics on the approach, business case, and lessons learned from the IBM conversion.
Download paper (Available to CMG members only, Click here to learn more.)
The Myth of Precision Planning: Understanding Capacity in an Age of Virtual Parallelism
Dr. Tim R. Norton
Virtualization is the current hot solution for a variety of computing problems. However, the complexities it introduces create additional problems when trying to precisely plan capacity. Virtualization is really the application of techniques to increase parallelism, either actual or perceived, and has been used in many different ways for a very long time. This session explores many of the techniques used to virtualize and parallelize resources, the impact of those techniques on capacity and the resulting changes in the perception of precision for both planning needs and measuring usage.
Download paper (Available to CMG members only, Click here to learn more.)
VMWare ESX Server Workload Analysis: How to Determine Good Candidates for Virtualization
John Paul
This presentation will describe several approaches for categorizing Microsoft Windows server workloads based on the server resources consumed or infrastructure capacity available. The session will cover the steps necessary to measure and quantify those workloads and specific industry best practices for the target infrastructure.
Download paper (Available to CMG members only, Click here to learn more.)
Adjusting the Fit: VMware ESX Server Workload Analysis Lessons Learned
John Paul
The maturing of some of the performance analysis tools and techniques, along with the multi-core technology platform introduction have broadened the range of target candidates for virtual solutions. This presentation will provide detailed performance analysis steps, utilizing generally available tools, that allow quick review of workload performance. It will explain how to distinguish between virtualization-related and native performance issues, which are often the real cause of performance problems.
Download paper (Available to CMG members only, Click here to learn more.)
Beyond the Hypervisor Hype
Dr Michael A. Salsburg
This session provides a simple model of today’s hypervisors. It then discusses a set of benchmarks that have been published by VMware and XenSource. The performance implications of the benchmark results are discussed. Newly emerging virtualization enhancements are appearing on processor vendor roadmaps. The motivation for these, plus some insights into the announced enhancements are discussed. Finally, the impact of server virtualization on the role of capacity planning is discussed.
Download paper (Available to CMG members only, Click here to learn more.)
Enterprise Data Management Optimization
Dr. Boris Zibitsker
In this presentation we will cover how performance prediction models can be used to evaluate Enterprise Data Management alternatives, justify decisions, set performance expectations and verify results. Several examples that illustrate best practice of proactive performance management during application and information life cycle.
Download paper (Available to CMG members only, Click here to learn more.)
Late Breaking: Installing DB2 V9 on a Linux Server
Robert Andresen
This presentation will cover installing DB2 V9 on a Linux server to allow the IT staff a place to learn how to use the new features of the product before using it to build SOA applications. The goal is to have a non-critical server on which to test out different application designs and make the inevitable mistakes before attempting to design, develop and implement production applications. IBM's new DB2 V9 supports several ways of storing XML data. Historically, companies have a long lead time adopting new releases of DB2. Part of this is due to the learning curve of new releases, part from incompatibilities with applications developed on previous releases. Because of all the new features it will take some time working with this new release to understand how to use these new features in your SOA applications and the impact they will have on performance and older applications. This learning must take place before new applications can be designed that use this new technology. A good way to start this process is to install DB2 V9 on a sandbox system so developers and analysts can start learning the features and pitfalls of the product, building test applications and measuring the performance impact and resource requirements of the new features. Existing applications may be transported to this sandbox to test compatibility. Building such a database sandbox may not be in the current budget for many organizations. The operating system will require a license, the server will need a machine to run on and DB2 is licensed software. Most companies budget on an annual basis, which can delay getting approvals for a database sandbox, and there are no guarantees that the expenditures will be approved in the next cycle. This can delay the start of this project months or more. There are solutions to these issues other than waiting for the next budget cycle. Choosing Linux as the operating system can remove that license expense, and Linux will perform acceptably on less resources that some of the other operating system choices. This allows it to run on older hardware that has been decommissioned from production service or under virtualization. A sandbox learning system will not drive system requirements much beyond the minimum hardware requirements for DB2 itself. IBM also offers ways to obtain DB2 for free or at low cost. They have options for trial software as well as free personal use versions of DB2. Not only is DB2 Express-C free to download, but it is also free for development, deployment and distribution. It may also be beneficial to look into joining Partner World or Developer Works, if that is applicable. We will look at the different Linux distributions available that can run DB2 V9 and discuss relative merits of each. We will cover obtaining the Linux install media and preparing the server to run DB2 V9. We will look at tools to discover and install or upgrade Linux packages. Next, we will follow the installation of DB2 V9 step by step on the Linux server. After briefly looking at the new features of DB2 V9, we will review the hardware requirements for the install, then identify and install the software prerequisites. We will run the basic install script and then launch the DB2 first steps dialog in order to do post-install setup and customization. We will also install the IBM sample applications and development tools and look at some optional 3rd part tools. The goal is to quickly have the new DB2 V9 available for learning, testing and development using existing hardware without needing to purchase any additional software licenses. The sooner you begin this process the sooner you will be able to exploit the new XML capabilities of DB2 V9 in your SOA applications.
Download paper (Available to CMG members only, Click here to learn more.)
An Open Source ARM 4 Implementation
David Carter
One of the key factors in adopting new technologies is the availability of inexpensive tools for developers and experimenters. The ARM4.org open source project intends to produce an open source implementation of the ARM 4, and now ARM 4.1, standards suitable for a developer or an enterprise. Although still in it's infancy, the project has produced a working collector. This session describes the current state of the project and its development direction.
Download paper (Available to CMG members only, Click here to learn more.)
Unix/Linux CMG Quick Start Course
Adrian Cockcroft
This course focuses on the measurement sources and tuning parameters available in Unix and Linux including TCP/IP measurement and tuning complex storage subsystems, and a deep dive on advanced Solaris metrics such as microstates and extended system accounting. The meaning and behavior of metrics is covered in detail. Common fallacies, misleading indicators, sources of measurement error and other traps for the unwary will be exposed. Free tools for Unix/Linux are mentioned briefly in this class.
Download paper (Available to CMG members only, Click here to learn more.)
Monitoring and Tuning Open Source Databases
Peter Johnson
Has this happened to you? A user reports performance issues with a certain online application, and when you look into it the application in question is using MySQL or PostgreSQL. If it were Oracle, DB2 or SQL Server, you would know how to properly configure and tune it for the best performance. But what do you do if the database is open source? In this session I present some of the ways that you can monitor the performance of MySQL and PostgreSQL, and give some tips on configuring them for optimal performance.
Download paper (Available to CMG members only, Click here to learn more.)
Java Performance Analysis 401
Peter Johnson
An earlier session in this series looked at tools that can be used to monitor Java Platform, Enterprise Edition (Java EE) applications. But what happens if the data you are really interest in is not available, how can you gather that data? One possibility is to use a commercial profiling tool, but there is another, possibly simpler, mechanism: use the power of Aspect Oriented Programming (AOP) to generate you own statistics. This session introduces AOP term and concepts and shows how they can be applied to performance monitoring.
Download paper (Available to CMG members only, Click here to learn more.)
Consolidated Capacity and Performance Reporting
Craig A. Leikis
There are dozens of tools available for monitoring application and system performance, each with their own particular flavor or specialty. Over time, tools may be added to meet specific needs and “tool sprawl” can occur. This session explores the effort to deploy a consolidated Capacity and Performance Reporting solution. The team pursued a solution using Cacti, an open-source front-end to RRDTool, and existing tools. The end result was a cost-effective, multipurpose solution to the “tool sprawl” problem.
Download paper (Available to CMG members only, Click here to learn more.)
A Queue Simulation Tool for a High Performance Scientific Computing Center
James W. McGalliard
Carrie Spear
The NASA Center for Computational Sciences (NCCS) at the Goddard Space Flight Center provides high performance highly parallel systems to a community of computational earth and space scientists. Long running and highly parallel jobs are common in the workload. NCCS management structures batch queues and allocates resources to optimize system use and prioritize workloads. NCCS technical staff use a locally developed discrete event simulation tool to model the impacts of evolving workloads, potential system upgrades, alternative queue structures and resource allocation policies.
Download paper (Available to CMG members only, Click here to learn more.)
Configuring Linux on z/VM for Performance
Barton C. Robinson
Many Linux proof of concepts failed due to performance reasons, many of which would have been easily resolved by following guidelines in this presentation. Configuring Linux on z/VM is often counter-intuitive when new to the platform, and many system defaults do not default to good performance. This presentation will guide you through the performance etiquette and ensure you start a Linux project with proper guidelines.
Download paper (Available to CMG members only, Click here to learn more.)
The Promise of Network High Availability Systems—How to Properly Measure Convergence Times
Ankur Chadda
Convergence time is one of the main tests for resiliency while deploying a network. It’s the time taken to completely restore traffic once an event triggering a convergence occurs. The most common way of measuring is packet loss-derived convergence time. In this, a test tool sends traffic at a known rate, measuring the number of packets lost during convergence. If the measurements are done on a complex system, there are some caveats we need to be aware of. This session will discuss typical case and possible scenarios which could allow one to report a lower convergence time than the actual value.
Download paper (Available to CMG members only, Click here to learn more.)
Introduction to TCP/IP Performance Management
Nalini Elkins
Today's internet and corporate intranets are based on the TCP/IP protocols suite. Do you know if your network is tuned? Are your TCP Profile parameters defined correctly? What errors are your TCP/IP stack encountering? What is the impact on CPU time of TCP/IP transmission, connection and errors? This session will describe how you may be able to do a step-by-step performance check of the TCP/IP network. We will do the following: 1. Describe the workload and response time 2. Find tuning opportunities 3. Review profile parameters 4. Find trouble spots in stack / socket performance 5. Detail areas where further investigation is needed In a recent Network Health Check, we were able to eliminate communications errors affecting the important production DB2 application, eliminate unnecessary traffic and lower the CPU usage of TCP/IP. The session will address z/OS, Windows, and Linux platforms.
Download paper (Available to CMG members only, Click here to learn more.)
Your OSPF Network is BAD!
Major Lai
OSPF is designed to converge fast and scale in order to meet the challenges of the ever growing IP networks of today. Do you know how your OSPF network performs? This session provides you methodology to evaluate the performance of an OSPF network in terms of convergence time and scalability.
Download paper (Available to CMG members only, Click here to learn more.)
Fighting Against Spam: The ETIS Anti-spam Pilot Project
Dr. Pieter Meulenhoff
Unsolicited e-mail, better known as spam, is considered to be one of the largest problems of todays Internet. Some sources claim that up to 90 percent of all e-mail is spam which is a large resource claim on e-mail infrastructures. In this session we report the results of an anti-spam pilot held between four European ISPs during the first six months of 2007. The combined effort on several areas from technical to procedural is tested. One of these is passive network monitoring technology (such as Lobster) used in cooperation with spam filtering appliances.
Download paper (Available to CMG members only, Click here to learn more.)
Collecting Actionable Information to Effectively Manage Web Application Performance
Hon Wong
The adoption of Web applications is accelerating.While performance of client/server apps can be inferred by monitoring the server/network,same cannot be said about Web apps.Using client-side agents to measure performance is impractical for Web applications given privacy concerns and ubiquitous nature of the Web. Performance data must be actionable to quickly restore service levels and understand business impact. We will examine techniques used to measure Web application performance and their tradeoffs and offer a method to relate raw performance data to effective IT remedial action and business reporting.
Download paper (Available to CMG members only, Click here to learn more.)
The What If’s of Storage Capacity Planning
John Baker
We obviously want the best hardware for our customers but financial constraints often force us to make do with less. Wouldn’t it be nice to be able to model YOUR workloads against various hardware configurations to predict the impact on response time? Compare: drive sizes and speeds; cache sizes; port configurations including remote copy. Predict: workload growth impact; the impact of merging workloads.
Download paper (Available to CMG members only, Click here to learn more.)
New Developments in Storage Network Protocols
Tom Clark
This session will review recent innovations in SAN transport protocols, including iSCSI/iSER, Fibre Channel, Fibre Channel over IP, and Fibre Channel over Ethernet (FCOE). Also known as Data Center Ethernet, FCOE is of particular interest because it enables convergence of storage and LAN traffic on a common infrastructure.
Download paper (Available to CMG members only, Click here to learn more.)
Performance Tuning of Storage System using Design of Experiments
Prabu Dorairaj
Devi Prasad Bhukya
Lakshminarayana Prasad Kantam
This work discusses the suitability of Design of Experiments (DOE) methodology for storage performance tuning. The DOE estimates the main and interaction effects of three major storage factors, Cache Partition Size,LUN Strip Size and Cache Segment Size settings of an enterprise level storage system on the overall storage performance. Other factors influencing the performance are kept constant throughout the experiment to minimize the experiment duration and complexity. DOE has provided key idea, for effectively analyzing the storage factors relationship in storage performance tuning.
Download paper (Available to CMG members only, Click here to learn more.)
A Queueing Analysis of Selected Performance Issues with Virtual Tape and Robotics
William T. Gray
This presentation examines three important issues with virtual and robotic tape systems. Each of these topics is analyzed using queueing models with feedback. The approach is simplistic but reasonably accurate; more importantly, it allows us to examine each system from its roots to its impact. Topics include the following: 1. Robotic Pass Throughs and Drive Busy, 2. Concurrent Virtual Drives and job elongation, 3. The performance characteristics of synchronous vs. asynchronous replication with respect to concurrent virtual drives and elapsed time.
Download paper (Available to CMG members only, Click here to learn more.)
To CUP, or not to CUP, that is the FICON question!
Stephen R. Guendert
SAC Storage
CUP, or Control Unit Port is a holdover from ESCON directors. In a FICON environment, CUP allows for in-band management, and opens the door to FICON director performance metrics via the RMF 74-7 record, more commonly known as the FICON Director Activity Report. In an effort to reduce acquisition costs and be more competitive on price, many vendors will try and make the case why you do not need CUP on FICON directors. This session will present the reasons why, from a performance management perspective, ''not to CUP'' is the wrong answer to the question posed by the session title.
Download paper (Available to CMG members only, Click here to learn more.)
Understanding the Performance Implications of Buffer to Buffer Credit Starvation In a FICON Environment
Stephen R. Guendert
This session will give a brief review on BB credits including current schema for allocating/assigning them. It will then discuss the one way available to detect BB credit starvation on FICON directors, and focus on the concept of frame pacing delay and its impact on performance. Finally, the author will outline two new concepts: a mechanism to count BB credits actually being used and dynamic allocation of BB credits.
Download paper (Available to CMG members only, Click here to learn more.)
Performance Considerations for Archiving Data
Randy Kerns
Archiving data has been perceived as simply moving data off primary storage when it is unlikely to be needed again. Performance considerations for archiving involve the characteristics of the archive system as well as the nature of what constitutes archiving of data. Understanding the performance characteristics of the archive system is important to implementing a strategy and meeting requirements. This session will highlight what are the performance considerations for archiving data and present some real-world examples to aid in understanding.
Download paper (Available to CMG members only, Click here to learn more.)
Storage Performance Analysis – Drill-Down Case Study
Charles T. McGavin Jr
Data centers are consolidating more and more storage into large disk arrays. In hyper-consolidated environments, where perhaps hundreds of servers are all accessing the same disk array, determining the causes of performance issues is becoming more and more difficult. This session describes a drill-down methodology that was employed to quickly sift through large amounts of data to determine the ultimate causes of the client’s performance issues and propose effective remedies. This methodology could be applied to any application running under any operating system on any vendor’s disk arrays.
Download paper (Available to CMG members only, Click here to learn more.)
Late Breaking: Storage Performance Council Benchmarks for Storage Management and Information Lifecycle Management
Bruce McNutt
The third major Storage Performance Council benchmark effort, SPC-3, is aimed at performance thests in the area of Storage Management and Information Lifecycle Management. At its core, SPC-3 is built upon the ability to construct a realistic content repository, such as a file system. Performance tests (such as backup, restore, or tests of access times) may then be conducted against the content repository. This paper introduces the types of performance testing envisioned for SPC-3, with particular emphasis on the importance of performance in assessing the effectiveness of an ILM solution.
Download paper (Available to CMG members only, Click here to learn more.)
Disk Storage Performance Warranties: Why and Why Not?
Rich Milner
John Gatch
This presentation discusses why and how to create a disk performance warranty. Issues addressed include: the motivation for warranties, the method through which performance expectations can be determined, the vendor perspective, the customer perspective, and reasonable terms and conditions for a warranty. Case studies will be used including the use of a disk performance modeling system.
Download paper (Available to CMG members only, Click here to learn more.)
Mainframe Disk Performance Modeling and Open Systems Applicability
Lee Reiersgord
John Gatch
This session presents a mainframe disk modeling case study using LeadTime, a disk performance modeling system, for configuration analysis and the setting of performance expectations. Can the same process be used for open systems?
Download paper (Available to CMG members only, Click here to learn more.)
A New Way to Expose Performance Counters on Windows Vista
Wedson Almeida Filho
Ricky Buch
Insung Park
The Windows platform exposes performance counters as a means to measure the performance of system components and applications. Creating performance counters for general applications has not been straightforward and presented challenges. Windows Vista introduces a new performance counter infrastructure that greatly reduces the burden in adding counter providers. This session will describe how the new infrastructure enables easy creation of general counters and give examples of how to implement new counter data providers using the new managed and unmanaged APIs.
Download paper (Available to CMG members only, Click here to learn more.)
Understand the Performance Measures of MS Virtual Server
Dr. Jie Lu
Tianxiang Zhang
Microsoft Virtual Server is a multithread application that runs as a system service on top of Windows Server 2003. Performance analysis in such environment is always a challenge. The performance concerns are also quite different comparing to traditional Windows environment. With understanding of the architecture, this session discusses critical metrics to track for various scenarios, such as CPU or I/O intensive computing. Based on experiments and benchmark test, the paper presents a practical guide on how to understand and correlate the performance measures obtained from different sources.
Download paper (Available to CMG members only, Click here to learn more.)
Intelligent Aggregation of System Health Data for Managing Large IT Environments
Vishal Radhakrishnan
Aaron Spinks
Insung Park
Recent Windows platforms have incorporated instrumentation technologies for analyzing system problems. However, even when IT organizations effectively utilize these technologies, the cost of remediating issues on individual machines is still substantial in large business environments. We explore a technique to identify and solve frequent problems by aggregating reliability and performance data across many systems, enabling IT organizations to take pre-emptive measures to address top user issues. We present our experiences from the deployment of this technique in a large IT environment.
Download paper (Available to CMG members only, Click here to learn more.)
Techniques for Identifying and Optimizing Resource-Intensive Sql Server Queries
Jeffry Schwartz
Analysts are often provided with vague descriptions of database system performance like “sluggish” or “this report takes forever” instead of quantitative measurements. This session discusses methodologies, metrics, and techniques that can be utilized to ascertain the cause(s) of poor database query performance and determine appropriate solution(s).
Download paper (Available to CMG members only, Click here to learn more.)
Windows System Performance Measurement and Analysis
Jeffry Schwartz
This basic tutorial in the CMG-T foundation curriculum introduces the metrics that are available from the Windows operating system and most prevalent applications. The sheer number of available metrics makes it difficult for anyone, even those analysts who are well versed in performance analysis measurements on other platforms, to discern the most important performance counters. This course will provide the necessary information to enable the Windows performance analyst to ascertain what the most important metrics are, how to interpret them, and the most appropriate collection mechanisms. It will also highlight those measurements that are not easily obtainable. Performance data collection and analysis issues using standard tools will be discussed.
Download paper (Available to CMG members only, Click here to learn more.)
Optimize .NET & J2EE Application Performance Using Browser-to-Database Monitoring and Diagnostics
Hon Wong
While .NET and J2EE frameworks simplify the challenge of developing distributed Web apps, it increases complexity of deploying and managing those apps. Adoption of Web 2.0 techniques and SOA Web Services only amplifies this complexity, making performance assurance a serious challenge as apps move through deployment, ramp-up and optimization. The session offers a simple holistic browser-to-database approach to measure performance and triage incidents. Using practical examples, we examine key monitoring points & data that must be collected and analyzed in real time to allow IT to restore optimal performance.
Download paper (Available to CMG members only, Click here to learn more.)
Modeling Virtualized LPARs using Coadunation
Clea Zolotow
Elaine Zizzi
Migration from non-virtualized systems to virtualized LPARs has many sizing challenges. One of these challenges, the optimization of virtualized LPAR configurations, is addressed in this presentation. We use the term ''coadunation,'' which means to closely join together. In conjunction with IBM Research, we can model migrations of both currently virtualized and non-virtualized hosts to ensure better hardware utilization. We present both the theory behind this process and examples of its implementation within the ODCS.
Download paper (Available to CMG members only, Click here to learn more.)
A z/OS WLM Update for DB2 Environments
Glenn R. Anderson
Begin this fast-paced session with a review of how DB2 exploits WLM on z/OS, including advanced WLM functions such as enclaves and application environments. The relationship between the different types of DB2 work and WLM will be explained. Then move into a look at the most recent updates to WLM functionality and their relationship to DB2, including zAAP and zIIP support, contention management, sysplex routing enhancements, stateful session placement, and the latest z/OS 1.9 WLM enhancements. If you are at all interested in the powerful combination of DB2 and WLM, this session is for you. It does assume a basic knowledge of the z/OS Workload Manager.
Download paper (Available to CMG members only, Click here to learn more.)
A z/OS WLM Guy Discovers Enterprise Workload Manager (EWLM)
Glenn R. Anderson
Enterprise Workload Manager (EWLM) provides a way to monitor and respond to workload processing across multiple systems in a distributed heterogeneous environment. EWLM is mainframe technology (z/OS WLM) migrated out to distributed platforms. In this session Glenn Anderson, long-time WLM instructor, will share his impressions and experiences with EWLM. What is this EWLM thing anyway? How does EWLM fit with z/OS and WLM? Do we need both products? Why should System z people care about EWLM in the first place? The session will also review the latest EWLM status, announcements and updates.
Download paper (Available to CMG members only, Click here to learn more.)
Introduction to z/OS Monitoring, Tuning, and the Workload Manager
Glenn R. Anderson
Build your z/OS measurement and tuning skills through this fast-paced seminar, jam-packed with information you can use today. First up we take a look at monitoring z/OS performance using SMF and RMF. Next, dive into the world of System z shared LPARs, including the new specialty engines like zAAPs and zIIPs. What do the terms mean, and how does your configuration manifest itself in the RMF reports? Next, jump into the world of z/OS Workload Manager, as we delve into assigning goals and importances for your workload, proper use of WLM classification, and an introduction to advanced WLM services like enclaves and application environments. Of course any discussion of WLM would not be complete without a look at the current RMF Workload Activity Report. Finally, a short review of the basics of DASD performance, z/OS paging, and GRS. What are the important metrics to watch, and what do they mean? A steady stream of important information awaits you, but trust that your instructor Glenn Anderson will keep the proceedings lively, interesting and highly useful.
Download paper (Available to CMG members only, Click here to learn more.)
Evolving DB2 CPU and Response Metrics
Ned A. Diehl
DB2 is a growing subsystem in many large system environments. A major part of this growth is as an enterprise server. Accompanying this growth is evolution of DB2, operating system, and processor capabilities. Installations exploiting this evolution must adjust data analysis and reporting to allow for changes in available data. Recent releases of z/OS and DB2 have included significant updates to RMF, DB2, and SMF 30 records. This session will discuss the sources and analysis of DB2 CPU metrics and corresponding response times. Primary focus items include DDF and zIIP.
Download paper (Available to CMG members only, Click here to learn more.)
zIIPS and zAAPs - Understanding Transaction Flows and CPU Measurements
Peter Enrico
Today’s transactions on z/OS can run on zIIP and zAAP processors, as well as traditional general purpose processors. This session will discuss some typical transaction flows involving zIIP and zAAP processors and how the CPU time consumed is accumulated to the address space SMF 30, the processor SMF 70, and the WLM service class period SMF 72.3 records. Additional topics discussed will include dependent and independent enclaves, client SRBs, unmanaged treads, and other key concepts necessary to understand the interpretation of zIIP and zAAP CPU times.
Download paper (Available to CMG members only, Click here to learn more.)
CE CMG - Best Paper: J2EE & Mainframes - When Quality Matters
Wolfram Greis
The main objective of this sessionis to help you select the best deployment platform for your e-business applications. J2EE is THE platform for vendor-independent application development. When your applications are ready for production, you have to select the deployment platform which is best suited to host them. E-Business applications are only valuable if its accessibility, currency, consistency, security and integrity are guaranteed. The topics covered are: Mainframes and e-Business Java 2 Enterprise Edition (J2EE) The Mainframe as a Deployment Platform WebSphere on zLinux or WebSphere on zSeries? Practical Examples and Experiences
Download paper (Available to CMG members only, Click here to learn more.)
E2E Performance Monitoring to the Mth Tier (Mainframe Integrated) – The New Industry Standard
Thomas A. Halinski
Eric Repec
How many reports do you use to determine if your E2E (including the Mainframe) needs tuning? Are you sure you are looking at E2E from the same perspective as your End-User (i.e. user satisfaction with enterprise applications)? Would it help if you could convert many measurements into one number? Would this benefit your SOA applications? Taking it to the next level, how is the customer satisfaction of you application performance affecting your business goals? Here’s what the Apdex Alliance and Industry Leaders are now suggesting and what it takes to accomplish it.
Download paper (Available to CMG members only, Click here to learn more.)
The Myth Of MSU or How Big is the Bucket?
Jim Horne
Knowing how big your mainframe is may sound simple, but is it? If you are only looking at it today, and not looking at yesterday or tomorrow, it is, but how do you compare it to the way it used to look or the way it may look in the future? IBM’s Intelligent Resource Director and new specialty engines may be a good thing but they also mean new challenges for capacity planning and performance reporting. This session is intended to offer a review of the RMF TYPE70 and TYPE72 records, and point out why what you actually get may not be what you think it is.
Download paper (Available to CMG members only, Click here to learn more.)
Planning for Newer Mainframe CPU Technologies (zAAPs, zIIPs and OOCoD)
Linwood Merritt
Newer mainframe CPU technologies show great promise for reducing hardware and software costs. Mainframe customers purchase books of CPUs that are enabled (as needed) as general purpose or lower-cost specialty engines. Another nuance of this capacity model is the temporary enabling of general purpose engines (Capacity on Demand). This session will discuss capacity planning techniques to size and plan for these new technologies.
Download paper (Available to CMG members only, Click here to learn more.)
Reclaim your Performance with DB2 v8
William R. Miller
DB2 v8 offers greater capabilities than v7. The theme of the release is “Breaking the Limits”. Many of the existing limitations were greatly relaxed. IBM changed DB2’s internal architecture to take advantage of 64-bit processing. For your application and data needs, the improvements in the existing functionality and the advancements of the new capabilities are very welcome. These advances come at a cost. As with any new version of DB2, IBM reckons with an increased CPU usage. Estimates vary between 3 and 10%. This session describes some actions you can take to reclaim some MIPS back.
Download paper (Available to CMG members only, Click here to learn more.)
Taming the Wild Service Definition
Rich Olcott
The most effective way to learn about an old but complex piece of equipment is to take it apart and rebuild it. The author was given such a learning opportunity when presented with an antique Service Definition dating back to the dawn of WLM. We learned several valuable lessons for the future while gaining control of its 89 Service Class Periods, 20 goals at Importance 1, and SRM coefficients smelling of old technology.
Download paper (Available to CMG members only, Click here to learn more.)
The Hidden Cost of TCP/IP on z/OS
Bill Yeager
Mainframe systems have traditionally used channel protocols (CTC) to exchange data, thereby realizing performance and efficiency benefits. Most enterprises use communications networks (TCP/IP) to exchange data between homogenous and heterogeneous systems. This session will contrast the performance and processor impacts on z/OS with the use of TCP/IP for high speed data transfer, provide empirical data and a measurement methodology, as well as describe alternative mechanisms to dramatically reduce processing while increasing performance for high volume data movement.
Download paper (Available to CMG members only, Click here to learn more.)