This panel is intended to explore best practices around capacity planning on Linux and Unix platforms. We will also discuss ways in which our panelists hone their skills and update the models that they utilize for capacity planning on their platform.
8226 - Computer Performance Evaluation
A short discussion of different software performance engineering aspects (including requirements, modeling, profiling, load/performance testing). Experts will be available to answer questions in a Q&A session.
Placeholder for optional panel as time/space permits. Papers would take priority.
The Storage Performance Council (SPC) is a cross-vendor team of storage performance experts that has built the industry’s first benchmarks for storage. These benchmarks are the standard for decision making in many organizations. The SPC has used real-world workloads as the basis of the benchmarks that are vendor-neutral and platform independent. Many SPC-1 and SPC-2 results have been published to date. This panel session will discuss the status of the SPC and the Storage Industry standard performance benchmarks available and under development including Component Benchmarks and SPC-3.
If you have a DB2 question regarding performance tips, functionality, insights, or DB2 special features such as stored procedures, zIIPs,version 8 conversion and version 9 performance, then this is THE panel to ask. The members will be able to answer your questions and enlighten you with their valuable expertise. Come prepared with questions that affect your work life. Email them as soon as you can to db2_panel@cmg.org, or drop a written question into the Q&A box you will find at various "System z" track sessions, or hand your written questions to any "System z" session monitor.
If you have a zSeries performance question, this is the panel to ask. Some of the many performance related questions the panel of experts can answer include: System z processors, processor configurations, general Sysplex, z/OS system performance, WLM anything, variable Workload License Charges, WebSphere, etc... Come prepared with questions, email them as soon as you can to zos_panel@cmg.org, or drop a written question into the Q&A box you will find at various "System z" track sessions, or hand your written questions to any "System z" session monitor.
Capacity planning is undergoing a transformation as virtualization technologies gain adoption in IT environments. This paper describes the nuances of capacity supply in virtual environments, how to model the demands that workloads place on this supply, and how to effectively align supply and demand in a way that maximizes efficiency and minimizes operational risk.
In the many-core era, the modest increases in the processing speeds of host computers that are forecast are not keeping pace with improvements in network bandwidth. To scale effectively, the processing of TCP/IP packets must be distributed across multiple CPUs. This paper motivates the necessity of taking this approach to high-speed networking. It also discusses the implications of parallelizing TCP/IP on the performance of the applications that reside one layer above the TCP protocol on machines with NUMA characteristics.
In this session, we will discuss energy features and the concepts around energy management. We will cover capacity planning with energy management in mind on the UNIX platform. IBM / POWER 6 energy features will be used as an example and a reallife example will be presented.
AIX: The New Era – Are Defaults Enough? This session will describe the new AIX tuning parameters along. The new algorithms for memory management will be discussed as well as the defaults that AIX V6 now deploys. We will have a discussion on workload tuning vs. system tuning and have an open forum/panel on tuning and Power 6 experiences at the end of the session.
This presentation only session will talk about some of the basic system tuneables in AIX for an Oracle environment and will then go into some of the resources you can request from your DBA to help with tuning. We will also look at an AWR report to figure out what other areas you should look at for tuning the system. This is written from the perspective of an AIX admin, not a DBA
8216 - Computer Performance Evaluation
The growing suite of Open Source Performance Testing tools seem to work well when testing "simple" static webpages. But how easy (or difficult) does it become to trap and submit dynamic data when using tools such as OpenSTA, Jmeter, and Webload? A quick comparison and tutorial of this trap/submit technique across OpenSource tools will save performance engineers alot of time when scripting more complex web applications. Is it any easier to build dynamic webpage load tests with commercial products such as LoadRunner? We’ll take a look at the usability and outputs of these tools collectively.
8215 - Computer Performance Evaluation
Analytic modeling and sizing techniques for performance analysis and capacity planning have existed for over three decades. How do the latest UNIX and Windows virtualization strategies get represented? This presentation demonstrates a selection of modeling and sizing techniques specific to the architecture of the virtualization’s measurement data as well as to the architecture of the virtualization itself. Examples include IBM AIX partitions, HP nPar and vPars, Solaris containers, and VMware virtual machines. Similarities and dissimilarities of sizing techniques are cataloged.
This article focuses on configuring LUNS on Solaris systems with mixed HBAs. For brevity, implementation of SAN on a NetApp storage appliance using FCP protocol is illustrated. The content herein provides valuable information about NetApp, Solaris and VERITAS utilities for a SAN administrator to detect, configure and manage LUNs.
8213 - Computer Performance Evaluation
Most every performance analyst and capacity planner has a Performance Data Base (PDB), in which performance data is archived. Typically, the key metric – the star of the show – is CPU % utilization. However, if your PDB has additional metrics, and you have some creativity, imagination and curiosity, you can find gold nuggets of information to add to the meat and potatoes of your reports. This paper presents a few experiences in mining the PDB.
8212 - Computer Performance Evaluation
Application Performance Index (Apdex) is a single number used to capture user satisfaction with application performance. In this paper we use an analytical model to study the relationship between Apdex and the capacity of the system that supports the applications. The analytical model provides guidelines on how to provision the system capacity to achieve the desired Apdex rating. It also helps us to better understand the correlation among application response time thresholds set by the users, the user satisfaction level, the Apdex value, and the mean response time delivered by the system.
8211 - Computer Performance Evaluation
We were too busy running many 7x24 applications to buy or build a sophisticated capacity analysis system. We would examine graphs and take action in a reactive manner. As our data center grew, we were drowning in graphs -- top daily, top weekly, top monthly. There were too many graphs to review. We turned to an old standby. Using Excel’s Descriptive Statistics features and few macros, we’re able to operate proactively -- we’re now catching and solving capacity problems before alarms go off.
Today’s culture of rapid technology refreshes benefit the technology vendor provider far more often than it does the customer. In a financial climate where technology must pass ever increasing justification tollgates it is imperative architects/engineers/support managers understand when and how to avoid unnecessary refreshes and how to structure necessary refreshes to their benefit. This paper will identify the significant driving forces behind technology refreshes and provide a frame of reference for end user’s to identify when and how to leverage technology for cost savings.
This session suggests a methodology for diagnosing network traces using visual metrics, animation and sonification. We expect that VDL will allow the diagnostician to find the problem in a trace in far less time than the normal methods used today. We have applied for a patent for this technology.
8208 - Computer Performance Evaluation
One of the challenges in building analytic queuing network (QN) models is obtaining service demands for the various workloads and devices. While some of these parameters can be easily measured, some may be hard to obtain due to measurement complexities or because it may not be feasible to stop the operation of a production system. This paper discusses an approach for computing unknown service demand parameters for QN models, i.e., it solves the problem of finding a subset of the service demand values given the known values and given the values of the response times for all workloads.
8207 - Computer Performance Evaluation
Profiling Oracle is an excellent way to identify the cause of performance problems and ways to correct them. In the early design stages of new software systems, however, code is not yet written so profiling has limited applicability. This paper explores the applicability of SPE models for predicting the performance of Oracle applications. It examines a case study and compares performance results from measurements and models. We present some conclusions on the applicability of profiling, system performance monitoring, and SPE models for managing the performance of Oracle applications.
8206 - Computer Performance Evaluation
Managing the performance of software that is outsourced can be a minefield. As the client you might assume that your contractor is using good software performance engineering (SPE) practices to develop your software. All too often, however, the delivered product does not meet its SLAs. This paper presents some best practices that were used by the Canadian government to manage performance on a contract for outsourced software development. Some are revised to reflect more recent SPE developments.
8205 - Computer Performance Evaluation
This presentation details a proven method for defining the workload model and identifying system bottlenecks for post production enterprise applications developed from numerous consulting engagements at Fortune 1000 corporations. Although this presentation focuses on post production, the outlined approach to understand business drivers, model system transactions, and conduct performance tests that can be used to address bottlenecks and conduct planning at any phase of the software development life cycle.
This presentation details the necessity for I/O parallelism given the increasing I/O appetite of today’s processors and the prevailing breed of diverse applications. It scrutinizes the complimentary technologies such as high speed FICON, Extended Address Volumes (Large Volume support), multi-level caching and priority structures. It also looks at storage consolidation and the issues faced in the provision of differentiation of the workload and its unique attributes. Finally the impact of all of this upon the ever evolving role of the Capacity Planners and Performance Analysts.
A Computer Network does not work in ideal condition. It is important to know the characteristics of such a network for its management. Network characterization is the study of properties like latency, loss, and available bandwidth of a network. In this paper we discuss our method to measure latency, loss and available bandwidth of a link across a wide area network. This method neither introduce large traffic into the network, nor require both ends of a link during measurement. Ours is a simple to use method which runs at one end of the link and needs only the IP address of the other end.
This session will show how to use SQL PL , from beginner through advanced topics, including stored procedures, UDFs and triggers. Advanced topics include temp tables, nesting, dynamic results sets and performance. The presenter will offer examples of each, as well as discuss debug and performance profiling.
8201 - Computer Performance Evaluation
Addressing performance testing formally, using the "waterfall" approach, often leads to prolonged agony of performance troubleshooting or even missing performance problems altogether. With small extra efforts, making the process more agile, efficiency of performance testing increases significantly – and these efforts usually pay off multi-fold even before the end of performance testing.
8200 - Computer Performance Evaluation
Probabilistic models have been used to analyze the performance of computer systems and communication networks for more than four decades. They have been used in other branches of science and engineering even longer. In many cases, such models provide accurate results despite the fact that their underlying mathematical assumptions are unlikely to be met. The key to this puzzle lies in understanding the distinction between distributional and trans-distributional properties. This distinction leads to new insights into the surprising accuracy of stochastic models in many real world settings.
8199 - Computer Performance Evaluation
System Virtualization allows multiple O/S images to execute on a single physical host computer. Measuring the host resource usage is straightforward and the necessary tools are included with most virtualization environments. Complexities introduced by the different virtualization techniques create problems with measurements within the guests that impact the ability to precisely plan capacity for applications. This paper proposes a more holistic approach using the Simalytic Modeling technique to planning capacity needs by understanding the effect of resource usage on application performance.
8198 - Computer Performance Evaluation
The presentation will demonstrate how business processes, from frontend to backend can be profiled on high level, by a new developed method called “GAPP”. The method makes it possible to get a high level profiling of a business process without actual change anything in the business process code, or start any tracing in complex architectures, like SOA environments. The method is based on data mining and uses this in a very spectacular way. The data mining is done on already available data like CPU and I/O metrics, but also other metrics can be used. The power of the method is astonishing.
Storage virtualization has been around for years, and server virualization is maturing. Now, network virtualization is emerging. Enterprises need to ensure separation of data and controlled access to network resources for many reasons. This can result in parallel networks with attendant cost, management, and usability issues. New solutions to this problem virtualize a common infratructure into communities of interest that ensure the security of the data and resources associated with them. This paper examines some of the techniques employed to accomplish this network virtualization.
8196 - Computer Performance Evaluation
Customer today invest huge bucks in acquiring network resources. The decision however needs to be backed with some sensible and structured approach towards sizing. We need pointers and statistics that can help make this decision less vulnerable. The last thing that one needs is a saturated network, playing stimulus to poor end user response times. The paper presents a formal approach adopted in a real life application to automate the capture of network statistics during performance tests. It further proposes a model for estimating the network bandwidth using the captured network demands
8195 - Computer Performance Evaluation
Customer today invest huge bucks in acquiring network resources. The decision however needs to be backed with some sensible and structured approach towards sizing. We need pointers and statistics that can help make this decision less vulnerable. The last thing that one needs is a saturated network, playing stimulus to poor end user response times. The paper presents a formal approach adopted in a real life application to automate the capture of network statistics during performance tests. It further proposes a model for estimating the network bandwidth using the captured network demands
Measuring and forecasting Data Center electrical power usage can be an elusive goal. This paper shares tips and lessons learned in trying to measure and forecast UPS usage profiles as equipment is removed and added. Overall UPS usage profiles are examined to dispel some common myths about equipment power usage. Power usage profiles for common equipment types are examined, especially configurable assets. Power usage metrics are provided for some common equipment types. Measuring techniques are explored and some suggested reporting formats are provided
Green Capacity Planning is the holistic approach to effectively plan for today’s data centers. The approach builds on traditional computer and network capacity planning to include environmental considerations. This paper describes the terminology, theory, and practical application of the methodology using a case study. The case study evaluates the capacity, storage, and energy footprint of an existing e-Business system. It then quantifies the potential energy improvements to be gained through virtualization as the system is scaled up to meet demands of a growing business.
Our company has been using application network traffic analysis as part of our application performance problem determination. TCP/IP network activity often identifies the portion(s) of the application that result in poor overall performance. Training in this technique has the ability to identify poor performing application modules/objects, CPU constrained components, implementation configuration issues, and specific poor performing application SQL. This paper presents the basic techniques of this type of performance issue determination, along with several real life examples.
The identification of business driver information is an early step in its use for capacity planning. The information may be in the form of business metrics and affected IT resources or a compilation of growth rates. The ideal capacity forecasting methodology creates predictions of future resource demand by combining business forecasts with resource trending. This paper discusses forecast techniques and issues around the identification and use of business growth information, and outlines approaches to integrate business forecasting with capacity forecasting.
DB2 SQL executes using three main DB2 address spaces, SSAS, DBAS and IRLM. To measure and see how it performs, we need to capture its activity from its "originating" address space. This varies depending on the source of the SQL: Mainframe, Distributed or Stored Procedures, making a trilogy. Let’s see what each is and how they work. With a good methodology we can "tune" any of them.
Aligning IT with Business, including End User Experience, is accomplished in a business climate with the following three items: 1) Business value - determined by the End User 2) Performance = the End User Experience. 3) Business and IT Views - graphically displayable from each perspective. With these items, IT can identify and monitor its business services quality and able, on a real time basis, to adjust their quality level. Conversely, business is able to manage their end user services by alleviating the IT issues which negatively impact the bottom line. Let’s see how this is done.
IBM’s PowerVM on pSeries provides a powerful virtualization layer to the Power platform. This paper covers information learned while spelunking through the documentation along with experimentation in AIX environments running in various configurations of PowerVM partitioning. Specifics include how to interpret data values provided by performance tools like: iostat, lparstat and topas.
8187 - Computer Performance Evaluation
ITIL has huge momentum and is well summarized as being “documented ITSM common sense.” Its main objective is to help newcomers find a sensible path towards Good Practice. This paper describes what Bad Practice can achieve. ITIL is a bit like motherhood and apple pie – you can’t knock it. But you can observe some of the misconceptions and spurious detail that has emerged around it over the years. This paper starts with comments about ITIL and its history and then submits piercing reviews from two perspectives, that of application development and that of the ITIL process practitioner.
8186 - Computer Performance Evaluation
Do you know how many zIIPs and zAAPs your shop is using? Do you know how many it could be using? Does your management know? This paper will show you ways to let management know what your specialty engines are doing in simple yet effective terms – namely pictures. Several graphs will be discussed and examples will show why there is no one right graph. For the programmers, the SAS code necessary to build these graphs will then be covered and will include publishing them on the web in a z/OS environment.
8185 - Computer Performance Evaluation
Successful enterprise capacity planning depends on swiftly (and ideally automatically) detecting and removing process pathologies (loops, ramps, shrieks, etc.) prior to analyzing real consumption needs. While detection algorithms are similar across all firms, your firm might also have specific needs that you are unlikely to be able to buy "canned" from a vendor. You need a way to apply these rules to all machines and processes. You need a rule engine. Join your author as he shares important techniques, shortcuts, problems encountered and labor saving ideas that he’s learned building his.
8184 - Computer Performance Evaluation
Virtualization introduces a new capacity management paradigm in your datacenter. Anytime there is shared resources, you have to plan, coordinate and reserve capacity in advance in order to avoid contentions, inefficiencies or shortfalls. In this session, we will provide some guidelines, recommendations and best practices to better manage and plan capacity with your virualized infrastructure. We will discuss some classic concepts of traditional capacity management, and how they can be adapted for your virtualized environment
A number of initiatives are being adopted by companies to make their datacenters as "green" as possible in order to not only save cost and improve efficiency but also do their part in optimizing resource consumptions. This presentation takes a look at strategies that can be used for Oracle database servers as part of the "Green Datacenter" initiatives.
Consolidation can deliver significant savings by improving efficiency as well as driving down TCO and IT infrastructure complexity. Oracle features such as automatic storage management, resource manager and real application clusters (RAC) provide numerous opportunities for database consolidation. This presentation will discuss strategies that can be used for database consolidation using Oracle. A real-life database consolidation exercise will be used as a case-study.
8181 - Computer Performance Evaluation
Capacity planning for Oracle systems is more than sticking a wet finger in the air. It requires some special considerations and tools. The database server should not be treated just like another server. In addition to a robust and reliable infrastructure, you should gather several key metrics using Oracle provided tools that will help you determine when and where you can use more capacity: on a node level (advisable) or on a per machine level (CPU, Memory etc). In this presentation, we will look at capacity planning for servers that use Oracle databases.
8180 - Computer Performance Evaluation
In the process of engaging a custome to deploy a cloud solution we encountered a custoemr that had extreme goals for provisioning. They desired to provision 100 AIX systems in one hour. This brigns up the question then of how many NIM srevers woudl this require. To understand the scaling behavoir of a NIM server we took a 16 way P5 and ran a series of tests where we provisioned clients. In our paper we will show the factors tat you must consider when trying to scale a NIM server beyond the typical 10 concurrent clients that are most common.
It’s no secret that performance metrics are key to performance management. Having immediate access to those metrics will simplify your life as a performance analyst. There are numerous advantages to automating the collection of performance metrics and building a Performance Database. We’ll show you how to reduce hours of research time into seconds with easy preparation and planning. Real world examples will be used to demonstrate the impact you can have on your company’s bottom line with automated charts and reports generated from your PDB.
8178 - Computer Performance Evaluation
Alchemy is the science of turning base metals into gold. Capacity Management is about optimising performance to save money, in other words to create gold. But today, in the IT world, green is becoming the new gold. A survey amongst our customers shows that green issues of one form or another will be one of main issues of 2008. This presentation will show how traditional Capacity Management techniques can be adapted to the new green world. It will also discuss what new techniques need to be developed so that we can start turning gold into green and save both the environment and money
8177 - Computer Performance Evaluation
Macro 4 has been helping users to manage and optimize their systems for over 40 years and we have picked up many tips and tricks that we would like to share with you. The challenges range from the mainframe to complex, multi-platform, composite application issues. We’ll share real examples of how users have solved seemingly difficult problems with the help of the right approach and (sometimes) the right tools for the job. We hope that you will take away 10 tips, many of them “quick wins”, that you will be able to apply in your own environments.
Data centers are becoming more constrained and it’s time for the capacity manager to step up to the forefront and help out. This case study analyzes one such data center. It describes a five step approach to mitigating data center constraints, delaying a costly data center upgrade and data center relocation. Best practices will be described, including data center level metrics in the industry, and recent governmental work in the data center metrics.
This paper describes an OS-level virtualization technology on the Windows platform called feather-weight virtual machine (FVM) and several of its applications. Compared with HAL-based virtualization such as VMware and Xen, FVM is more flexible and scalable, incurs less start-up and run-time performance overhead, and requires no infrastructural changes. The key idea in FVM is name space virtualization, which isolates virtual machines by renaming resources at the system call interface and thus guaranteeing that the resource name spaces visible to different virtual machines be disjoint.
Traditional fault-tolerant systems assume a system can detect a fault immediately when it occurs. When a fault is caused by a human error or malicious attack, this assumption no longer holds because the interval between when a fault occurs and when it is detected, could be relatively long. Some of the side effects during a fault detection window are considered corrupted while others benign. A repairable information system is one that after an error/attack, can quickly and automatically erase corrupted side effects while preserving benign ones, all without human intervention.
Existing storage virtualization systems focus only on one storage attribute, capacity. This paper describes a novel multi-dimensional storage virtualization system, Stonehenge, that virtualizes a cluster-based physical storage system along multiple dimensions, including bandwidth, capacity, and latency. Stonehenge can multiplex multiple virtual disks, each with a distinct bandwidth, capacity, and latency attribute, on a physical storage system, and allows creation of virtual disks that are as tangible as physical disks, and yet much more flexible and easier to manage as software entities.
The focus of this research is to provide a targeted case study on a specific configuration of SQL server that exists in a corporate environment. This installation was designed without performance architecture considerations and will be compared to a benchmark test system that should be very inadequate compared to the production system. The results provide valuable information on assumptions that are commonly made about what is the "best" way to do storage. The research is also intended to show the value that computer performance evaluation brings to the architecture process.
8171 - Computer Performance Evaluation
Recently Claire Cates of SAS has published “Threads – The Good, the Bad, and the Ugly” within this venue. In that article, Claire introduces you to the various threading considerations (the Good, the Bad, and the Ugly). In this article, I will address various tools on the market, to help with the threading issues that Claire introduced to you. In Addition to exposing you to the tools – I have been working with Claire Cates on using Intel® Thread Checker at SAS. Within this article I will expose you to some of the real life experiences we have shared in dealing with the bad.
Dateline: December 2007- Bali, Indonesia The United Nations Framework Convention on Climate Change required the design, implementation and operation of the largest temporary network erected to-date in Indonesia. Managing both expected and un-expected challenges is par for course in International Project Management. Ranging from the usual limited budget and extremely short timeframe to scope changes and cultural issues driven by a diverse constituency from over 100 countries. Tune in to find out more about how to apply lessons from UNFCCC Bali to your fast-moving domestic/intl. projects.
This paper focuses on practical field techniques for measuring and baselining wireless network performance. Topics: Comparing LAN performance to WLAN performance. What, Why, How and When to capture 802.11a/b/g/n wireless network performance metrics including signal strength, noise levels, latency, network speed, wireless retransmissions and (most useful of all) throughput. Measurement Interpretation Establishing performance baselines (site surveys) for comparison pre- and post-implementation. Previous knowledge of wireless networks is not required.
8168 - Computer Performance Evaluation
Data center power usage has both economic and environmental ramifications. This paper presents research done to determine the power performance of a typical data center blade server with various operating systems installed and the processes idle. CentOS is found to be more efficient than Windows 2003 Server. Multiple VmWare ESX virtual machines are found to have no power penalties. If an unused blade is left in a running chassis, we determined that an operating system should be running on it so that the software will manage the hardware power efficiency.
8167 - Computer Performance Evaluation
I will explore ways to analyze performance data for tens of thousands of servers using common OLAP and Visual Analytics tools to work with the data; leveraging the multi dimensional nature of this many servers in a large complex environment to understand where capacity issues and optimization opportunities exist.
With all the tools for capacity and performance management available from SAP which ones do you use and how? This session discusses Intel’s Capacity and Performance Management Procedural Practices. It details Intel’s use of QuickSizer, Early Watch Reports, Service Level Reports, Wily Introscope, Central Performance History dashboard and reporting, and the development of SAP capacity plans in the day-to-day workings of SAP capacity and performance management.
8165 - Computer Performance Evaluation
Planning changes in system configurations using performance prediction tools is a challenge task. Prediction models can be used to forecast the system utilization in most situations. However, different conclusions can be derived for the analysis of a scenario in different time scales. In this paper we investigate the mispredictions that can happen while migrating workloads in multi-core systems. In addition to system utilization, we may need to consider other metrics to get a comprehensive picture of the system performance and to make accurate forecasts in the time scale of interest.
This article focuses on configuring LUNS on Solaris systems with mixed HBAs, for brevity implementation of SAN on a NetApp storage appliance using FCP protocol is illustrated. The content herein provides valuable information about NetApp, Solaris and Veritas utilities for a SAN administrator to discover and manage LUNs On Solaris 10 update 2 systems emlxdrv utility needs to be run to choose which driver controls the Emulex HBAs in the system. This article delves into making persistent binding changes, discovering targets and LUNs without rebooting the host & multipathing.
8163 - Computer Performance Evaluation
Problem Statement: Our call centers were complaining of “slow performance” for their home grown front-end to our CRM system. Approach: After having a brainstorming session with the developers, we decided to take an ARM approach, without ARMing the code. Results: We had a custom agent developed that reports on the end-user experience for our entire call center staff. Conclusion: We now have the ability to alarm, report, track, and trend on end-user experience across our entire foot print.
World renowned storage experts to present their lastest insights and field questions about many of the problem faced by datacenters today. Invited speakers are: Gilbert Houtekamer of IntelliMagic, Bruce McNutt of IBM, and Tony Mungal of EMC. Confirmed panel members TBA.
8161 - Computer Performance Evaluation
In this paper we present a method for identifying bottlenecks using a new technique called kneepont analysis. This technique can predict the current operating state of a computing system as captured in a queuing model and can predict how much headroom is left. From the utilization data of each element, we define an operating curve that can be used to identify a kneepoint. The kneepoint is used to identify when the system enters a region of non-linear performance response. In this paper we describe the formulation of a kneepoint, and examples of how we use kneepoint analysis in practice.
8160 - Computer Performance Evaluation
Performance Visualization technology can dramatically improve performance analyst productivity. This is nowhere more true in the field of storage performance analysis. This paper will explore innovative performance visualization approaches for storage performance analysis.
Enterprise Flash Drives were recently introduced into the high-end storage array market. These drives have unique characteristics that influence their deployment. This paper will present some preliminary performance numbers on EFDs and discuss best practices discovered so far.
What is the real cost of usage for IT investments across organizations? How should infrastructure costs be allocated to departments, locations, applications or external entities based on usage? This presentation examines the current state of IT resource metering and cost allocation with attention to trends and available tools. The presenter will share a proven methodology for planning, executing and sustaining a usage and cost allocation project. Attendees will receive a checklist of project action items, the tools and information necessary for a successful project and tips to manage IT costs.
*** This needs to be a solicited talk, as the information won’t be publicly available until after WinHEC (fall) and thus can’t be submitted as a paper. I have not heard yet from CMG whether or not this talk can be classified as solicited! *** Previous versions of Windows Server have enabled increasing levels of power optimizations and controls, and Windows Server 2008 R2 will continue that trend. This talk will discuss the various power management optimizations in past and future Windows releases, as well as provide power measurement data from multiple workloads and servers.
8156 - Computer Performance Evaluation
Dynamic performance stubs provide a framework for the simulation of the performance behaviour of software modules and functions. They can be used to realise a cost-benefit analysis of the gain from performance optimisation and therefore, for a gain oriented improvement. It is also possible to identify “hidden” bottlenecks and the most relevant optimisation candidates. This paper classifies several types of stubbing possibilities and evaluates the CPU Stubs more in detail which can be used to optimise CPU bound modules or functions.
In an era of enterprise virtualization expansion the ability to provision just the right amount of resource and anticipate necessary changes, not only saves money but also ensures that valuable resources are utilized effectively. This presentation will explore how SAS Institute’s Remote Access Computing Environment uses IT intelligence best practices to manage its own VMware infrastructure resulting in the optimization of resources and costs.
8154 - Computer Performance Evaluation
Application throughput can be dependent on a number of computing subsystems. When a performance issue arises, it is difficult to quickly identify the root cause of the problem. A methodology is needed to identify bottlenecks in a computer system. If we can construct a queuing network model to capture the dynamics of each of these hardware resources, and then accurately represent the associated application workload, we can begin to identify bottlenecks during application execution. In this paper we present a method for identifying bottlenecks using a new technique called kneepont analysis.
8153 - Computer Performance Evaluation
Project to reduce IT Costs with focus on optimizing applications running on Mainframe DB2 and a better analysis automation. Reducing over than 20% CPU on BATCH and ON-LINE with only 3 work days. Immediately financial results delaying upgrades and reducing future upgrades, getting more availability and quality from DB2 Applications.
8152 - Computer Performance Evaluation
When running multiple applications concurrently in a shared environment, it becomes difficult to identify which one is causing a specific performance issue. I’s helpful to separate out individual contributions of each workload on the system. If we do the identification either in the time domain using raw workloads fitting (reconstruction), or in the frequency domain using Fourier methods, we won’t accurately separate each workload present in the concurrent environment. We present a method to identify an application based on performance characteristics using least-squares spectral analysis.
8151 - Computer Performance Evaluation
Queueing network (QN) models, commonly used for prediction in complex systems, typically provide only the mean for predicted response time. Service level agreements (SLAs), however, normally include criteria such as "90% of the response times for Transaction X should be less than five seconds." Percentiles are not determined by the mean, so predicting SLA conformance for response time based on QN models involves significant guesswork. In this paper we propose and justify heuristic methods for making the guesswork more scientific, thus reducing the risk of inaccurate predictions.
8150 - Computer Performance Evaluation
We explore performance challenges IT professionals face scaling dynamic web applications; strategies to achieve scale and performance ensuring the strategy includes measurement for closed-loop performance management. Should the strategy be specialization, distribution, both? These techniques let you isolate specific elements for independent scaling spreading the workload across multiple servers. Combining both offers the largest scale, but more complexity to manage and diagnose. Critical is measuring web app performance, knowing what to measure and how to use the available tools.
8149 - Computer Performance Evaluation
Business organizations face a daunting task of maintaining the peak performance of their applications, on failing which they have to incur heavy revenue loss in terms of increased IT investments and abated reputation. To alleviate this problem, performance engineers often turn to various performance testing solutions, which, on the other side, introduce their own hefty expenditure in organization’s overall IT investment. This paper proposes an approach that allows organizations to leverage open source technologies for economical and effective performance testing of applications.
Electricity to run servers is becoming more expensive than the server hardware. Data center power use is growing in amounts that exceed planned growth in generating capacity. Sustainability and Green IT initiatives are becoming corporate imperatives. Measuring power consumption and carbon footprints of business areas, applications and the supporting computing infrastructure is becoming critical. Metrics and physical measurement are in their infancy. This paper describes approaches to combining resource usage measurements with benchmark and estimated data for power use and carbon footprints.
8147 - Computer Performance Evaluation
This paper will cover the capacity planning process and considerations in a multi generation node Teradata DataWarehouse environment. It will include workload characterization based on data loads, transforms, and reporting. Also covered will be customer interactions with application owners, job scheduling, and the use of advanced analytics in planning and forecasting capacity requirements.
8146 - Computer Performance Evaluation
In order to obtain more throughput from the CPUs on a system without adding faster or more CPUs, vendors have developed CPUs with multiple threads of execution. In this architecture, the threads of a particular CPU share certain resources, such as low-level cache. In this paper, we will give a basic comparison between single-thread vs. multi-thread CPUs. Then, we will review operational performance of a number of different types of multi-threaded CPUs on various operating systems. Finally, we will demonstrate considerations and methodologies for modeling multi-threaded CPUs.
Performance issues in complex environments are a fact of life. When they happen, its important that the end user have their own tools to effectivly diagnose the problems. If the end user does not have their own tools, they must rely on vendors, and trust the vendors to educate, not filter out relevant details, not play politics, and give them the right solution, even if its painful. Finally, problem solving the root cause and not biting off on things that are symptoms and not the root cause. This session is a case study that serves as an example to show what not to do.
PerfSheet is a freeware data visualization tool which is based on Excel and VBA code. This tool is useful for performance analysts who often have ad hoc need to flexibly visualize performance data, without being constrained by static dashboard mentality. PerfSheet is able to fetch data from multiple data sources such relational databases, files, FTP/HTTP locations and interpret resultsets of SQL queries, CSV, XML files and HTML tables. The performance analyst then can use Excel’s capabilities for viewing, correlating, slicing & dicing the data and storing the favorite “views” for future use.
Most generic performance tools display only system-level performance data using 2-dimensional plot or diagram, and this limits the informational detail that can be displayed. A modern relational database system like Oracle, however, can concurrently serve thousands of client processes with different workload characteristics and generic data displays inevitably hide important information. This paper demonstrates the application of Barry007 multidimensional visualization for analyzing Oracle end-user session-level performance showing both collective trends and individual performance anomalies.
8142 - Computer Performance Evaluation
When an analytic model or sizing tool predicts CPU Performance, it uses (1) queuing theory calculations and (2) a representation of relative performance for each CPU configuration. This paper focuses on (2) using a sample CPU rating system from www.spec.org. Our research results include an algorithm for finding functions to describe relative performance, as well as specific properties of general and vendor-specific functions. These results can be directly applied to any case where the CPU vendor supplies limited benchmark data compared to the number of CPU configurations actually available.
8141 - Computer Performance Evaluation
On machines that support some variant of HyperThreading, ie. Thread Level Parallelism, the meaning of utilization is unclear. In particular, reported utilization is likely a non-linear function of OS dispatching, the arrival distribution, the processor architecture and how the average is calculated. That is, there is little reason to believe it obeys the "utilization law". This issue is looked at for Xeon/Pentium processors available from Intel since 2002. A new approximation for actual utilization is presented.
This paper is targeted at IT Leaders, Service Managers and ITSM process owners who intend to improve the quality of IT services in their organization and lower operational costs. The paper introduces the meaning of quality in ITSM context and highlights the key quality goals for an ITSM Organization. Statistical inputs from various studies and experimental models have been included to provide quantitative indicators to the impact of contributing factors. A quality improvement approach has been suggested based on our findings, experience and experiments.
8139 - Computer Performance Evaluation
It is difficult to create a consolidated report of performance information due to wide differences in the scale and variability of performance data. One technique to avoid this problem is to put different metrics on a similar scale by normalizing, or calculating the z-statistic of each metric. This paper discusses how to scale the data using normalization or z-statistic calculations and how to create a performance dashboard using spreadsheet software.
8138 - Computer Performance Evaluation
Software development organizations are increasingly becoming more focused on performance evaluation and improvement. An initial process for performance evaluation is proposed and has been studied on a software application under various configurations and workloads. This method, using a combination of threshold analysis and correlation cluster analysis on collected performance measurements, identifies key relationships among different computer performance measurements, as well as existing bottlenecks without detailed knowledge of internal system information.
8137 - Computer Performance Evaluation
A critical parameter for ETL system’s evaluation is the time taken to process the data. As the data volume varies, challenge is to predict the time taken to process that data and the capacity required to meet the SLA for new volumes. In this paper, we showcase the use of performance modeling to analyze ETL processes for different volumes with/without parallelism using a simple case study.
8136 - Computer Performance Evaluation
Attribute-based Workload Characterization (AWC) combines qualitative models for performance with quantitative analysis of the relationships between workloads and performance measures. An initial case study is shown applying the AWC methodology to Enterprise Service Bus system performance. Documenting the performance quality attributes helps clarify the dimensions of application usage, and raises important questions about the system architecture. The preliminary results show all workload factors were identified by AWC and there is good correlation of the workloads with measured performance.
According to IDC, Business Transaction Management software will be a key factor in the successful deployment of new computing platforms based on virtualization. This presentation illustrates how BTM in a virtual environment allows organizations to: - understand the relationship between the configuration of virtual environments and the service levels such configurations provide for transaction flow - monitor business transactions to identify, isolate and resolve host/VM contention issues - gain insight into IT transactions to meet service level goals
8134 - Computer Performance Evaluation
Methods to find a resource bottleneck in a single machine or tier that suffers from a resource crunch are well known. However, is the crunch the real bottleneck or is it a symptom of another problem? An unresolved resource crunch will repeat since root causes haven’t been addressed. Application bottlenecks impact business transactions’ throughput and service levels, but tracking transaction flow guides us to real bottlenecks. This presentation shows which parameters are needed in order to find actual bottlenecks, parameters that hide real bottlenecks and how to interpret the data we find.
The quest for Key Performance Indicators for specific e-Governments Initiatives is a search as old as e-Government. However, research has provided quantitative and qualitative sets of indicators related to various dimensions of e-Government. This paper tries to organize these propositions based on a life cycle and provide a comprehensive framework to a local or national e-Government initiatives to select the indicators needed to track its progress and impact.
Most load balancing environment today focus on distributing the load between the various services. They are in fact performing a localized optimization so resources are equally utilized and no specific server/resource is over utilized. But, in doing so, are we actually improving the end user service? Can we use load balancer to optimize the business service and not just a local resource optimization? This presentation will focus on how attendees can effectively measure load balancing and explore ways to fine tune them so it has a positive impact on the end user business service.
8131 - Computer Performance Evaluation
To provide a better ROI for IT, system administrators need to make sure that an enterprise does not loose its customers due to non-availability or unsatisfactory performance of IT applications. Such good performance and optimal utilizations of existing IT infrastructure can be obtained by running processes, online or batch, in parallel, scheduled to exploit the workload trend. This paper proposes a framework which allows system administrators to study the performance behavior of parallel processes and schedule them accordingly.
Server IO performance is a critical factor in the data center. With server consolidation through virtualization, it is interesting to understand impact of this on the IO performance of virtual machines. In this study we attempt to model the relationship between IO performance of virtual machine and the number of virtual machines on consolidated hardware. This model is expected to help in deciding how many virtual machines to be hosted meeting desired IO performance for a given hardware.
8129 - Computer Performance Evaluation
Business organizations face a daunting task of maintaining the peak performance of their applications, on failing which they have to incur heavy revenue loss in terms of increased IT investments and abated reputation. To alleviate this problem, performance engineers often turn to various performance testing solutions, which, on the other side, introduce their own hefty expenditure in organization’s overall IT investment. This paper proposes an approach that allows organizations to leverage open source technologies for economical and effective performance testing of applications.
Nowadays, the service oriented architecture implemented in the form of webservice technology is "hot" and its application in business critical environments is growing. In the context of a chain of composite services we investigated the effects from severe overloads and possible countermeasures, targeted at keeping service performance at agreed service levels. In this presentation we present and evaluate an intelligent request discard mechanism that resolves this problem.
Although medical processes are supported by international medical standards, they are independent and with few interconnection points between them. The problem can be seen when a medical process, BPEL orchestrated, needs different services with different standards to run. Inside this process, a workflow may be invoked by a HL7 event that invokes a web service to get the last medical study made in another hospital department using the DICOM standard. Using the IHE technical framework with the Open ESB, in this work a possible solution for this problem is presented.
8126 - Computer Performance Evaluation
We needed to upgrade our servers because the load on our system began to increase unexpectedly. However, at the same time, we needed to add new functionality. We conducted measurements four times because we needed an accurate gauge of the performance requirements and potential performance problems as early as possible. The tests were two simulation measurements and two load test. We will discuss the differences between these results and what efforts were made to adjust performance prior to the successful completion of the upgrade.
8125 - Computer Performance Evaluation
High availability of Enterprise Applications is of paramount importance. With increased business competition, downtime of a few hours would equate to huge losses in revenue. Due to software anomalies and their effects avoiding downtime is a difficult proposition. Many enterprise applications have such inherent flaws which usually cause downtime once the application is released to production. This paper describes an approach to reduce downtime in spite of anomalies caused due to aging effects, particularly memory leaks and thus ensure HIGH AVAILABILITY of Enterprise Applications.
8124 - Computer Performance Evaluation
Profiling is important for ensuring efficiency of the code. In SDLC the code passes through development, test and production. In each phase a stakeholder has different requirements, hence most profilers support only certain phases. Jensor, our home grown Java Profiling solution is low-overhead and provides rich data under high workload which makes it “fit-for-purpose” for all stakeholders. We present Jensor architecture and benchmarks along with a few innovative analysis modules. These modules help in doing Performance, Scalability, Reliability and also Business Analysis using technical data.
8123 - Computer Performance Evaluation
It is evident that the design patterns in application provides multi fold benefits when properly used and deployed. Much have been discussed for J2EE/.Net design patterns through various books and forums but design patterns for databases are rarely discussed, published and not available in standardized format. This paper hightlights design patterns for databases based on the authors’ experiences on database modeling and designing of various projects where the usage has shown significant improvement in mantainability, performance and scalability.
8122 - Computer Performance Evaluation
One of the challenges in analyzing performance data is discovering inter-relationships between different attributes. We present a visualization technique that compares multiple software releases by displaying performance attributes gathered on running different workloads on a storage server. We construct salient visualizations that help analysts, experts and engineers to explore, analyze, see trends, patterns, and identify performance bottlenecks. We use simple geometric objects that vary their spatial position, color and texture appearance to encode different performance workload attributes.
Business considerations for cost allocations sharing resources and who really has the obligation to pay. Since the early 70s this has been discussed both inside and outside the datacenter as business units wanting their "own" systems continue to voice their concerns for fiscal responsibilities. As the pendulum swings back toward consolidation we once again find ourselves searching for acceptable practices to bring the sharing of resources in line with the cost of these resources. Even more heated are the discussions of how upgrades are acquired and who should pay.
IBM introduced Parallel Access Volumes (PAVs) in 1999 to facilitate mainframe customers moving to larger logical volume sizes. This presentation will review the history of PAVs and describe some test results that compare the performance benefits of Static PAVs, Dynamic PAVs, and HyperPAVs.
8119 - Computer Performance Evaluation
Network workload models rely to some degree on certain invariant properties of internet traffic. In this paper we investigate some implications that such invariants introduce in the accuracy of network performance models. We also investigate factors that impact the validity of invariance property assumptions and the range of scales in which such invariants apply. The aim is to establish sensitivity of ensuing performance models on these assumptions. Finally, some practical problems are discussed.
8118 - Computer Performance Evaluation
This paper describes a methodology for modeling workload for distributed systems. It derives concise and descriptive models of workload from the actual workload traces. The job inter-arrival time, runtime, CPU required, and daily arrival pattern are modeled. This methodology has been successfully used in several cases studies to model workloads from enterprise grid systems to supercomputers in academic environments.
The concept of striping has been around for many years as a method to improve throughput. The idea being that increased parallelism will provide superior benefits than increased speed. RAID provides the default level of striping in most modern Disk Subsystems. SMS Striping is another popular method that can provide a great deal of control over data placement. Now IBM is coming along with Extent Pool Striping, where logical volumes will span RAID Ranks. This paper will examine the various software/hardware methods available and discuss the strengths and weaknesses in each.
Virtualization & physical networks are diverse environments that need to co-exist; as a new ecosystem, challenges that can impact the operational performance of the overall network are presented. This session will look at: •How to manage a combined virtual & physical environment •What it takes to ensure VMware doesn’t tax the network using real-time infrastructure reports •Visibility & metrics for pre-consolidation planning, allocating resources, capacity planning, maintaining end-user performance, resolving troubleshooting issues quickly & correctly & the management tools needed to do so.
We built a Grid Capacity Model and have used it for the past two years for; 1) reporting on capacity used, 2) forecasting future capacity needs, and 3) reducing current unused capacity. I will discuss how we built the model, fine tuning the monthly reporting, and the results we have experienced.
This paper investigates feasibility of using UHF RFID readers and passive tags for identification of persons. Specifically we performed this experimental study in order to better understand the limits, costs and performance of UHF RFID as a tool for identification of individuals. Results showed the technical feasibility of this approach but also highlighted that individuals’ collaboration is necessary for tag reading to be successful. Keeping in mind certain technical precautions UHF RFID has a high percentage of success and is more widely accepted by users than active sensor technology.
Large Scale Virtualization technology available from today has made it much easier to prove significant cost savings and to implement rapid application deployment processes in the IT organization. By applying a contemporary Capacity Management approach to the myriad of virtualization and consolidation technologies now available the authors show how to keep the IT and Business groups happy. The approach extends to datacentre Capacity Planning (facilities, space, greening and power) providing an End-to-End Capacity planning approach.
How does an Australian-Wide Network servicing 30,000 Government employees transform to cater for a complete re-write of over 130 legacy mainframe applications to only a handful of mainframe and midrange applications? This paper details the tale of 2 Network Capacity Planners who were given the task to baseline, scope and model the current and new network within a challenging time-frame. Network Capacity Reporting, forecasting and application tuning techniques will be detailed. Particular attention is given to minimizing the extent of the network upgrade during the application roll-out.
8111 - Computer Performance Evaluation
Precise measurement of task execution time is central to resource management and performance characterization of systems. Tasks are often repetitive with short execution times depending on workload characteristics. CMPs facilitate these tasks to execute concurrently. In contrast to sequential case, precise & repeatable execution time measurement of concurrent, symmetric, & short duration tasks is a challenge. We review known time measurement methodologies, used in benchmarks, and introduce a precise & repeatable methodology for measuring concurrent symmetric tasks on CMPs.
Different workloads utilize system resources at different levels. Depending of the resource utilization pattern some workload may be better suited for hosting on virtual platform compared to others. The study is intended to analyze how different workloads such as Online Analytical Processing (OLAP), Online Transaction Processing (OLTP), Web and Email application perform on Virtual Machines (VM) compared to non VM environment.
8109 - Computer Performance Evaluation
The principal objective of this paper is to demonstrate the use of statistical techniques in undertaking Capacity Planning for midrange systems. It will show how, by using appropriate statistical techniques, one can overcome typical data related problem. The paper will draw on the authors experience in various forms of capacity plans and performance analysis, focussing on the application level. It will provide insight on how to create a capacity model and an integrated database, which are all capable of being delivered in short timeframes, with minimal input from application support staff.
8108 - Computer Performance Evaluation
This article will describe methods of extracting SMF data on a daily basis for specialized monthly reporting. A practical method of collecting data, storing and saving for processes like MSU reporting, security audits, software auditing and sub capacity reporting. This will show examples of data extraction and selection based only on the data needed for a given time period. Most examples will use JCL, the IFASMFDP program and programming examples as a method of automating SMF utility control parameter creation.
One of the primary problems facing performance engineers and others who work with customer systems is initial assessment of a “performance problem”. We have over 4700 systems for which we which to answer that question. To do so we produced a monitoring repository to house primary performance information. Reports from this repository allow us to quickly assess the reality of a performance problem and help pinpoint the cause. This paper discusses the creation of such a repository, the reports, and what we have learned from the experience.
A system using the VI3 SDK for the collection of performance metrics will be shown. The methodologies of this system augment the performance metric collection capabilities of Virtual Center (VC) by providing a centralized metric collection solution, more variation of metric averages, and flexibility in metric storage and access for large enterprises consisting of multiple VCs. As an example of the benefits, the calculation of aggregate guest Ready-Time for a host will be shown. Ultimately, this system provides a valuable tool for enterprise capacity management.
The number of guests that an ESX host can optimally sustain is an important capacity measurement for a virtual infrastructure. This paper explores the relationship of various host attributes and performance metrics to aggregated Ready-Time (RT) performance. Using guest RT as a host capacity indicator other host capacity indicators, such as CPU utilization, I/O, and distribution of guest types, i.e. single, dual, or quad virtual CPU, are quantified. This paper provides insight into how the quantification of host attributes and performance capacity relate to an optimum guest capacity.
Considering taking an advantage of pureXML applications on DB2 9 for z/OS? This paper will discuss what you can expect on performance and general practices on capacity planning for XML applications. These best practices are based on the laboratory performance measurement results. It includes the expectation and consideration applying zAAP and zIIP redirection for XML System Service. The paper should bring you up to speed on understanding performance impact of XML applications, basic performance tuning points that you need to watch out.
8103 - Computer Performance Evaluation
This session will show how to do Mainframe LPAR Capacity Planning using SMF Type 70 Records. By accessing only one z/OS LPAR on each physical machine, you can gather statistics for every LPAR and engine type in your data center. This session will demonstrate the metric collection for regular CPs and IFLs into the capacity planning database. SAS and MXG will be used to decode the SMF records and create the capacity planning databases. You can utilize any tool of your choice to do the same. After this session, you should be able size and trend your LPARs for the entire data center.
Virtual I/O server is widely adopted in server virtualization solutions, for it provides the sharing of physical disks and network adapters in a flexible and reliable manner. Traditional performance models for physical server lack the ability to represent such environment. On the basis of analyzing the architecture and characteristics of a variety of virtual I/O servers from different vendors, this paper presents a practical analytical model for measuring the performance of applications and servers involving virtual I/O server. It also introduces method leveraging the commodity modeling tools.
VMware ESX Server is a leading solution for server consolidation on x86 platforms. Most performance analysis so far focused on CPU primarily. In reality, memory poses more severe constraints to the capacity. This paper discusses the major performance concerns regarding memory based on the architecture of ESX Server memory management. Although VMware Virtual Center provides rich memory metrics for both host and each individual guest, there is no clear explanation on the meaning. With benchmark testing results, the paper studies these metrics and gives guideline on memory analysis.
8100 - Computer Performance Evaluation
Analytical models are constructed and used by capacity planners to predict and study computing resource requirements related to workload behaviour,volume changes measure effects of hardware and software changes. The paper also contain details about defined number of approaches to represent various systems & subsystems (like Memory)and device loadings representing the service required by jobs, and are the variables & parameters needed by the computational algorithm which calculates performance measures for the model.
Storage energy efficiency is becoming more important in our data centers, not only because energy efficient storage is environmentally friendly, but also because it takes less space and provides relief for data centers that are maxing out on their power and cooling ability. This paper provides a comprehensive overview of techniques that can be used to save power without sacrificing performance. The potential of FATA drives, the impact of the RAID technology, the use of multiple drive types, and the influence of mirroring will all be discussed in the context of potential power savings.
8098 - Computer Performance Evaluation
Performance baselining of applications is increasingly becoming essential before deployment. It requires extensive and detailed planning and simulation of Production scenario. This paper covers my experiences and recommendations that can be adopted when planning of production workload simulation and performance baselining of applications is being done. These experiences might help in eliminating rework and achieving best results from the exercise thus reducing the overall cost of the application development.
8097 - Computer Performance Evaluation
The summary dashboard is growing in popularity for showing the status of capacity management at major IT sites. It is best supported by a “Service Assurance Matrix” or “Risk Matrix” to incorporate more technical detail. This paper describes the implementation experiences of a “leading edge” site. Firstly the paper reviews the capacity management process and its auditing. Secondly, the implementation approach used at Euroclear is discussed, showing the challenges and concluding that dashboards for capacity management can be made to work to good purpose, despite some inherent weaknesses.
8096 - Computer Performance Evaluation
Although management consultancies offer audit services based on ranking sites’ IT processes against their peers, many companies prefer to identify any gaps in their adoption of good practice for capacity management. This is more objective, reveals “unknown unknowns” and relies less on attitude surveys. This paper reviews a number of such gap analyses. Irrespective of the enterprise culture (performance “just adequate”, or “mission critical”) the same approach is applied. Good capacity management measures workloads and performance to ensure that business needs are met now and in the future.
This paper describes in detail ITIL v3 Maturity Framework developed using author’s ITIL v3 experience and usage of this framework as reference for ITIL v3 implementation within Infosys. The 5-leveled ITIL v3 Maturity Framework has been based on maturity guidelines provided in ITIL v3 books. Practices have been further categorized based on different parameters such as general, organization, technology, processes, etc. Three ITIL processes have been traced on this maturity framework. The uniqueness of this paper is the maturity framework architecture and ITIL v3 process tracings on it.
8094 - Computer Performance Evaluation
In case of large database, it is difficult to guarantee the performance. The SQLs in such applications which are performing well in development, may not perform efficiently in the production because of volume mismatch. The database emulation technique manipulates the database statistics which gives a feel of high volumes without having actual data. The automation of emulation technique helps to integrate performance checking process with the development process and helps in •Detecting the queries which may not perform well in production •Tune the queries to get optimized execution plan
Xen is a rapidly developing alternative virtualisation technology to VMware. This paper describes the background to Xen, how it works and what it can achieve. It includes practical details of how to load, run and configure Xen, and it describes how well it performs. Special attention will be paid to the performance monitoring of Xen hosts and their virtual machines. A comparison will be made between the architecture and capabilities of Xen and those of VMware.
The key challenges the industry faces today in the adoption of SOA revolves around Performance management and Service Governance. Traditional approaches to performance management have been highly technology oriented. Given that SOA is mostly an enterprise-wide business-led IT initiative, tuning them for performance needs an enterprise outlook to provide tangible benefits. In addition to optimizing the applications for performance, the key to performance management lies in understanding the ecosystem that involves optimizing the business-processes & information flow across the enterprise
Growth patterns change frequently in z/OS computing systems. Capacity managers are often challenged as to perform future projections based on expected trends, or whether modeling studies should be undertaken to expand predictions to include performance implications. Workload categorizations, transactional volumes, analysts’ time, cost options, and overall benefits play into the decision to trend, model, or, to do both.
This paper will provide some ideas on how to ensure the performance and capacity needs of your business are met when some or all of your operations are outsourced. The success of performance and capacity management is largely determined by the outsourcing contract, so specific ideas on what should be considered during contract negotiations will be provided. Next will be an overview of governance and what the analyst should focus on during the life of the contract. Finally the paper will conclude with a discussion of the steps required to ensure a smooth transition at the end of the contract.
8089 - Computer Performance Evaluation
Most DB2 monitors can load accounting and statistics data into DB2 tables. We can use spreadsheets as an adhoc reporting against this data. Statistics rows show usage of critical resources during peak times and accounting rows show which applications are consuming the critical resources. We will build an inventory of tables and column names and generate SELECT statements from a Rexx exec. Each of our spreadsheets will be focused on a related group of metrics over a meaningful time frame. We will then subtotal, summarize and graph the data.
Many of us have co-workers, family or friends who may not understand safe web surfing. Once a PC is infected, it may be running intended applications so slowly that it is nearly impossible to use. How can we reclaim our computers from malware infestations? This paper will look at the methods this author has used to clean several infected computers. It will focus on using open source and freely available tools to clean the viruses and spyware and regain a useable PC without losing programs and data.
8087 - Computer Performance Evaluation
Dynamic instrumentation of message-driven systems offers an insightful window into recent performance behavior when statistics are calculated at defined intervals. Simple dimensionless metrics for service utilization and service stability are essential distillations of performance characteristics that may be used as predictors of changing systems behavior or workload characteristics.
8086 - Computer Performance Evaluation
The TPC-E is the new benchmark recently approved by the TPC. It is designed to exercise an On-Line Transaction Processing workload of a brokerage firm and be representative of current database server work. In this paper we present DBT-5, a fair usage open-source implementation of the TPC-E benchmark. In addition to reporting on the design and implementation details of the tool, experimental results on a system running the PostgreSQL database are also described. The significance of this work is that it provides an environment where recent innovations in the OLTP workload field can be evaluated.
8085 - Computer Performance Evaluation
A widely used medical Software Product does not provide Disaster Recovery (DR) capability. This paper details the process and the activities that led to the development of DR capabilities and tperformance considerations. The he process was iterative. Some of the iterations were: 1. Formulate DR Strategy, 2. Restore one server, 3. Restore to isolated network, 4.Restore some Software Package functionality, 5. Restore full Software Package functionality, 6. Restore from Client Network (rather than from development shop), 7. Restore to DR site (rather than development site).
8084 - Computer Performance Evaluation
How often does the need arise for modeling and forecasting? Should it be done manually by ad-hoc,by project requests or automatically? What tools and techniques are best for that?When is trending forecast enough and when is a correlation with business drivers required?The answers to these questions are presented in this paper.The capacity management system should automatically provide a small list of resources that needs to be modeled or forecasted;a simple spreadsheet tool can be used for that.This technique method is already implemented on the author’s environment with thousands of servers.
8083 - Computer Performance Evaluation
At times the analysis of a performance problem will identify a particular DB2 application as the culprit. You may even have the problematic SQL. But do you have any idea what to do next? Changes to application software are costly and time-consuming. Here is a documented set of steps to apply DB2 tuning from the most passive levels to the most invasive. A rehash of wisdom from the experts, this basic triage process shows how to perform effective performance tuning for the least cost and effort. This process has been applied to legacy applications, in-house development and purchased products.
8082 - Computer Performance Evaluation
This presentation will show how one analyst uses the pivot table feature of Microsoft EXCEL to arrive at a quick performance analysis for DB2 and BATCH mainframe applications. At our enterprise, we use pivot tables and charts as monitoring tools, performance analysis, and management communication. An explanation of the basic concepts to create and manipulate pivot tables will be combined with a real-time demonstration of the tables and charts. At times in your career you find a tool that is truly worth learning about. This is one of those times and one of those tools!
8081 - Computer Performance Evaluation
Workload analysis on the mainframe can be complicated if your business runs operations 7/24 with hundreds of applications. At our enterprise, we have a success story where the Z-series MIPS capacity has been held flat for more than two years. Each month every application is measured for current activity and compared to a baseline. Those applications that exceed a variance threshold are detected as “out of standard” or OOS for short. The OOS applications receive scrutiny to determine a named cause of the variance. This paper provides basic instruction and lessons learned.
8080 - Computer Performance Evaluation
Scaling up to the growing demand, the existing depository system faces challenges of keeping the cost per transaction low and ensuring that the performance is within the permissible limits. During re-design of existing depository application, the most widely used business functionality was redesigned and implemented by using J2EE technologies. Later, benchmark tests were conducted to validate the performance, scalability and ease of portability of the application across Mainframe and UNIX platforms and portability across Oracle and DB2 database.
As data centers continue to grow in capacity and complexity, and the businesses they support demand 24-7-365 operation, virtualization provides a key part of the solution, toward the goal of total ILM automation. This paper provides a high-level comparison of the three enterprise class storage virtualization offerings by IBM, EMC, and HDS; their architectures, and functionality. Then, details are profiled on the industry’s only control unit based solution: Hitachi’s USP V, highlighting the functionality necessary today, along with requirements for tomorrow’s total ILM automation.
8078 - Computer Performance Evaluation
Many data sources exist capable of providing an in-depth view into resource consumption and workload details of various application, database and system software packages. Typically these data sources are stored in logs, database tables, or formatted reports and do not lend themselves to efficient analysis. Parsing can be used to extract key elements of this data and store them in a common data structure, SAS tables, to be used for subsequent analysis. This paper explores a sampling of existing data sources and how parsing can be used to mine information nuggets for workload analysis.
By looking at the math behind the barycentric coordinates system, the simplex and the sum rules, it appeared impossible to produce a visual representation of barry5 with a good impedance match. This paper addresses the visualization of such problem domains using a few creative techniques from your Barry founders. If you want to see your performance data in new and more visually compelling manner, this paper is certain to at least wet your appetite.
8076 - Computer Performance Evaluation
This paper will showcases the methodology that can be used for performance analysis of the system by identifying hotspots, benchmarking application performance and optimizing the configuration of application. This will allow application owners to predict and the optimized application behavior in the production environment. This strategy is best suited in condition where non functional requirements are not well defined and in conditions when you need to benchmark the application for capacity projections.
8075 - Computer Performance Evaluation
The Universal Scalability Law (USL) is an analytic function used to quantify application scaling. It is universal because it subsumes Amdahl’s law (AL) and linear scaling (LS) as special cases. Using simulation, we show (1) that USL is equivalent to synchronous queueing in a load-dependent machine repairman model, and (2) how LS, AL and USL can be regarded as boundaries defining three performance zones. Typical throughput measurements lie in all three zones. Simulation scenarios provide insight into which application features should be tuned to get into the optimal performance zone.
8074 - Computer Performance Evaluation
Performance analysts and capacity planners should have at least a rudimentary understanding of HTML coding. While poorly coded HTML can lead to performance problems, you are probably more likely to be interested in using HTML to present your own data. While there are many reporting tools that produce HTML output, sometimes it is preferable to modify the HTML from these tools or create your own HTML so you can present the data in the exact manner you chose. Fortunately, the basics of HTML are not hard to learn and this paper is designed to start you on that journey.
Optimizing the data center environment can bring substantial and immediate savings with some incremental investment. Unfortunately, many IT organizations are not proficient at utilizing the levers available to reduce costs. During this interactive session we will explore opportunities for optimization in both outsourced and insourced environments. For each major cost category we will cover: Magnitude of potential savings Investments required to achieve the savings Risks / impediments to achieving the savings An action plan to achieve the savings
8072 - Computer Performance Evaluation
This paper describes a process that employs user simulation, capacity planning modeling, and software engineering techniques during the development and testing of a new or existing application. The paper describes how virtual user simulation is used to create a workload model on virtual servers and uses that data to reduce development costs, isolate and correct problem software components, define the correct system configuration, and predict the impact on existing SOA applications.