ENTERPRISE MANAGEMENT

Mark Ames
ICT Risk

Organisations putting their corporate data on-line want to make it available for business and still keep it secure. Managing the technology is challenging, but managing the business and operational risks is daunting. This talk discusses how business, security, and operational functions can be brought together to manage the risks effectively.

John Griffiths
ProActive Services

Service Desks are one place in IT that are very visible to customers. What happens there is a reflection, true or perceived, of how the customers view IT. Anything that can help reinforce a positive image of the Service Desk can only benefit all concerned. The process of introducing ITIL to your Service Desk could help, but how do you do it? This session will look at some practical issues such as where to start, what to consider and how to avoid some of the things that have caused others problems.

Mark Heers
EMC Australia

We are on the anniversary of the tragedy of the World Trade Center and the Pentagon. This singular event renewed the focus on business continuity with the senior management and boards of directors of companies around the world reviewing their company's strategies. Before then, Disaster Recovery (which more recently has been termed business continuance) for many companies was the existence of a document which described the steps to recreate their IT environment (or a key part of it) in a different location. For many, there was little will to regularly test the recovery plan nor to keep it up to date. For others, it was little more than a tick from the auditing section. One undoubted fact which emerged was that only two resources were truly irreplaceable - people and information.

This presentation will focus on some of the latest thinking with business continuity. Mark discusses several issues unearthed during the September 11 disaster. He also discusses issues such as keeping databases on diverse platforms in a consistent state and contrasts the difference between disaster recovery and disaster restart.

Chris Jones
Independent Consultant

The issues of IT metrics is a complex and integral part of any service management implementation. Within ITIL based implementations, metrics continuously emerge in every area, apart from Service Level Management. Metrics are the basis to reporting, but many times what we attempt to report, and what the customer/ management are expecting are two very different things.

INTERNET / E-BUSINESS

Tony Allan
APMS

Application instrumentation can significantly assist the processes of SLA measurement, performance analysis, capacity planning and problem determination. Unfortunately, you often end up drowning in a sea of data without finding the information that could be used for proactive management and decision making.

This paper suggests techniques than can be used to safely navigate this data.

Craig Linn
University of NSW (Nepean)

With a few specific exceptions response time performance on the World Wide Web is largely driven by proximity of the desired data. It is thus not surprising that an entire hierarchy of caches from the origin server to the final client browser has grown up - with this cache hierarchy not only reducing response time but also significantly cutting down on network traffic, improving availability, and distributing load. This paper first introduces the motivations and rationale for caching, and then provides a review of the structure of the traditional web cache hierarchy and how it maintains content freshness. Next it examines some of the features of the specialised web caching devices now available. It then goes on to consider the advent of edge-side network caching driven by Content Delivery Service Providers. In essence we look at the transition of web caching from a passive pull model for content distribution to what is fast becoming a very controlled and active pull model with a good deal of push thrown in. The paper is written on the premise that to manage caching (and related performance) one must first understand it. Therefore in addition to the presented material many references are included for sources that provide either more in depth material or practical guidelines as to how to enhance web system caching performance.

Brian Scott
Bunya Technology

XML has long been talked of as the next Good Thing. Others have taken the view that it could just be the most boring technology since ASCII. It isn’t particularly complex or sexy - it just works. But there are some pretty exciting ideas about how XML could make the world a nicer place. The author has been exploiting this underwhelming technology for 2 years. This paper aims to demystify XML and encourage you to enjoy its quiet charms too.

KEYNOTE

Jeff Buzen
Independent Consultant

Content Delivery Networks (CDNs) featuring "servers at the edge of the Internet" have achieved impressive levels of performance in a number of real world situations. Some content delivery providers have made bold claims about the complexity and value of their patented "rocket science" algorithms. This paper provides an independent look at these claims and examines cases where CDNs are, and are not, likely to deliver the performance benefits they promise. In addition, some experimental and analytic procedures for evaluating the performance of CDNs are briefly discussed.

Steve Samson
Candle Corporation

z/OS Workload Manager goes beyond what was available in OS/390 We'll examine critically the following topics:

• Intelligent Resource Direction
• WLM support of LPAR Defined Capacity
• ENQ enhancements
• Neat tips & tricks

OPEN SYSTEMS

Tom Beretvas
Beretvas Performance Consultants

In this session, the speakers will provide a methodology for identifying Linux performance problems. First the highlights are summarized. The emphasis would be on I/O problems. A case study is shown explaining the process.

Peter Enrico
Enterprise Performance Strategies, Inc.

UNIX System Services is now an integrated part of every OS/ 390 and z/OS system. In the past, for most shops it was usually something that was there but was not dealt with on a regular basis. Well, not anymore. UNIX Systems Services is a part the operating system that can (and usually does) affect every one of your workloads. Most new workloads (such as web serving and web application server) on OS/ 390 and z/OS run in the UNIX environment or is some way connected to the UNIX environment. Many of the traditional workloads are also been affected by UNIX System Services since the UNIX workloads (both large and small) are now in the mix contending for system resources with your traditional workloads. Regardless of the workload mix, UNIX System Services must be tuned for the benefit of all workloads.

This presentation is for those interested in the performance of their z/OS and OS/390 systems (that should be all of you).

Tuning UNIX System Services is a large subject, and it will not be possible to cover all topics during a single session. Instead the speaker will draw from his experiences with UNIX System Services to present a variety of relevant and important tuning recommendations. Items discussed will include how to gain an understanding of what is running in the UNIX System Service environment, and how to begin to monitor this work. The presenter will also discuss the new zFS file system, how it compares to HFS architecturally and in regards to performance.

A prerequisite for this session is an interest in UNIX System Service on z/OS. A goal of the presenter is for the attendee not just to learn a new recommendation, but why the recommendation exists. In turn, the attending will have a better understanding of UNIX System Services and how it could affect the performance of the system and the workloads running on the system.

Fil Faini
CPT Global

Increased visibility of an organisation's mission critical systems through the growth of Internet and web-based applications has seen the focus of capacity planning broaden. No longer can organisations continue to simply focus on the mainframe with the view of purchasing midrange capacity when the need arises. Procurement and installation lead times could see an organisation's mission critical application suffer poor performance for the lack of a capacity plan.

This paper will provide a framework for capacity planning in the midrange environment that can be modified and expanded to fit the diversity of any application. The paper will cover system and application data sources, strategies in collecting and storing data from multiple servers, methodology for combine system and business drivers into a meaningful capacity plan, building a simple capacity model, outlining some useful ROTs and threshold.

Ilyku Lee
Aurema Pty Ltd

A fair share scheduler allocates a share of a CPU to a task. A performance model tells whether the underlying CPU has enough capacity to handle given tasks within their required CPU response times. In this paper, a mathematical model for a fair share scheduler is proposed. First, an equation for estimating the worst-case CPU response time is derived. Then, a condition to tell if the underlying CPU has enough capacity to guarantee that all the required CPU response times are met is defined.

Barton Robinson
Velocity Software

Mainframes are dead? This presentation will illustrate the process of identifying those servers and applications that can and should be consolidated to s/390 (mainframe). Why? Server consolidation saves money in people costs as shown by many TCO studies. Server consolidation gives you the opportunity to provide better resource utilization and performance given certain conditions are met. The paper discusses the method of selecting the servers and applications to be so chosen.

Barton Robinson
Velocity Software

It is a big problem when managing a large network of rack servers to readily identify those servers that encounter performance problems. The paper presents a methodology and discusses the means of identifying the servers that have excessive CPU, storage or disk storage utilization. Deficiencies in the technology of performance instrumentation of these servers will be noted. A case study will be presented.

Richard Smith
Sun Microsystems

Solving performance problems is a bit like detective work. The initial evidence needs to be collected and analysed, but in itself may not be sufficient to solve the problem. Hypotheses may be formed and additional tests run to narrow down the possibilities. Part of the skill lies in being aware of the multiple sources of information available, and knowing how to interpret the data. In a typical operating environment there are many, many tools available for obtaining information, some well known, others quite obscure.

This paper is based on the author's varied experiences in tackling performance problems, based principly on SPARC/Solaris platforms. The experiences hopefully illustrate the practical application of a selection of tools and sources of data, as well as give some insight into the thought processes used to tackle these problems.

PLANNING AND REPORTING

Larry Chu
California State University

In an information system environment, Statistical Process Control (SPC) has become a performance tool for continuous measurement improvement. SPC helps to monitor workload activities, ensure user satisfaction, improve user productivity, and lower computing cost. Service Level Agreements (SLA) are a contributing key to the success of an effective and efficient information system department (ISD), and the Statistical Process Control concept has been adopted to be part of the SLA requirement. SLA reporting has become an important communication means between the ISD management and the users.

This paper describes our SPC methodology in targeting for 100% of Service Level Agreements, and shows the value of the SPC process in our information service environment, using a four sigma for the quality target. It details the one-job submission procedure to automate the SLA reporting to our intranet website, IBM Lotus Notes email system, the UNIX system, and our personal computer platforms.

Mike Tsykin
Fujitsu Australia

Service Level Agreements (SLAs) based on end-to-end metrics such as response time (ETE RT) are becoming common. For large networks, monitoring of these becomes very resource-consuming and trouble-shooting (identification of a probable cause of a problem) problematic. Automation of this process is necessary. This paper describes an approach to such automation. It describes the thery behind it, a sample implementation and the tools used. It also surveys the other tools available on the market.

S/390

Bradley Bruhahn
Sandpiper Data Systems, Inc.

True 'Global' IT continuous availability is now feasible with Geographically Dispersed Parallel SYSPLEX (GDPS) and related strategies. GDPS provides the control needed to maintain total data integrity within a 'mirrored' DASD and Tape environment and completely addresses the issues of a 'rolling disaster'. It allows z/OS and open systems to be completely available in less than one hour of a total data center loss, with a goal of no perceived downtime to the user. In this session the speaker will discuss the functionality of GDPS and detail the planning issues involved with mirroring two physical data centers. He will also cover several configuration options and performance aspects of GDPS that you should consider before embarking on this or similar projects in your environment.

Peter Enrico
Enterprise Performance Strategies, Inc.

OK - you've successfully migrated to goal mode and have even been there for a while. Your workloads are running well, goals are being met, and it appears your resources are being used efficiently. Life is good. Then one day, things just seem to run a bit differently. Maybe a goal is being missed. Or perhaps you notice that a particular set of address space are running a lower CPU dispatching priority or maybe even getting swapped out. Maybe a user calls up complaining that response times seem slower. Your natural reaction is to look at the standard workload activity data, but this analysis only shows you so much and does not give you the answer you are looking for. What you really want to know at this point is 'What is WLM doing?' or 'What is WLM's view of my workloads and resources?' or 'Why did WLM make that decision?'. This presentation will make the leap from the standard workload activity measurements to analysis of the SMF 99 records.

Ron Ferguson
Mainstar International

This session will explore several performance tuning aspects concerning your ICF catalogs for MVS. Topics will include catalog definition attributes that affect performance, Catalog Address Space buffering considerations, catalog integrity support with VVDS and Enhanced Catalog Sharing. The session will also show internal

Neil Gunther
Performance Dynamics Company (SM)

Amdahl's law and the Multiprocessing Factor (MPF) are two scaling models used throughout industry for sizing multiprocessor capacity in the presence of the multiprocessor effect. Both sizing models express different laws of diminishing returns. Amdahl's law identifies diminishing capacity with a constant amount of serialization in the workload, while the MPF model treats it as a constant ratio (compound interest). Each model, however, predicts very different results for large processor configurations that are so important for today's applications. Which one should you use? We turn to queueing theory for more insight and show that Amdahl's law corresponds to a synchronous bus model, while MPF represents a uniform Coxian server. The latter exhibits unphysical effects, such as SUBLINEAR response times(!?) hence, we caution against its use for large multiprocessor configurations.

Neil McMenemy
McMenemy Consultants

A major UK bank was about to introduce a new core banking application but needed to prove that it was scalable and performant. Rather than rely on modelling, the bank decided to invest in a ‘stand alone’ stress testing environment. This paper describes the issues in collecting performance data, analysing it, making tuning recommendations based on that analysis and the regression testing of those changes. Both on-line and batch will be covered as will the techniques and tools used to collect and manipulate the performance data. The exercise concludes with a discussion on how significant the new application is going to be on the existing production infrastructure.

Tony Shediak
Ampdev P/L

This paper covers practical performance tuning tips and experiences primarily aimed at the MVS environment using Cobol and PL/I with some consideration to IMS, VSAM and DB2. The emphasis is not only on tuning existing applications but also on developing performance considerate applications

SEMINARS & WORKSHOPS

Jeff Buzen
Independent Consultant

New pricing models for both hardware and software are radically reshaping the landscape for capacity planners and performance analysts. On the hardware side, companies no longer have to pay in advance to stockpile equipment needed to meet future surges in customer activity. Instead, they can activate hardware components "on demand" and postpone paying for these resources until they are actually required. Changes in software pricing may have an even larger budgetary impact. The Variable Workload License Charge (VWLC) pricing model implemented under the IBM License Manager (ILM) is already saving some companies many thousands of dollars per month by ensuring that software costs are based on required (defined) capacity rather than total installed capacity. Setting ILM pricing parameters so they accurately reflect actual processing requirements will become an enormously important function over the next few years. The workshop covers the highlights of the pay as you go revolution and explains the details of VWLC pricing under ILM including entirely new concepts such as soft capping, defined capacity, license certificates, white space, and the reconciliation log.

Neil Gunther
Performance Dynamics Company (SM)

Synopsis:

• Application scalability is key to success for many operations
• Shops use tools like LoadRunner (LR) to measure applications
• Tendency to focus on stress testing
• Scalability analysis requires a modified approach
• Manual Scenarios with steady-state windows
• How many load test points should you choose?
• How do you know you have a consistent set of test data?
• You need a model to sanity check the measured data
• What kind of model?
• Queueing models are too hard, and time consuming to set up (never gets done)
• Worse, they do not easily model effects like overloading
• I will provide a simple regression model in EXCEL
• Requires relatively few measurement points (4-6)
• Contains all major effects
• I've used it successfully for a decade
• Reduce the cost of benchmarking prior to deployment
• Based on Chap. 6 of my book: http://books.iuniverse.com/viewbooks.asp?isbn=059512674X1&page=fm5
• I will present several examples showing both good and bad LR data sets

Stephen Samson
Candle Corporation

You've put off migrating to Workload Manager Goal Mode for years, but now it is time to bid farewell to your beloved IPS, ICS, and other arcane parmlib members you probably never fully understood or exploited anyways. Steve Samson will give you a roadmap for implementing Goal Mode, with tips and techniques garnered from the experience of previous implementations at shops of all sizes.

STORAGE

Tom Beretvas
Beretvas Performance Consultants

In this session, the speaker will provide a methodology for identifying and curing storage processor (DASD) performance problems. First the highlights are summarized, then the procedures are described. The symptoms of tuning problems are identified, the method of finding them are explained and potential remedies are suggested. A case study is shown explaining the process. The case study illustrates the reduction methods that lead to rapid identification of problem areas to analyze the data. In addition, further tuning steps to be used are also discussed.

Ron Ferguson
Mainstar International

This paper will be on point-in-time copies for mirrored volumes, with respect to the performance aspects of how the copies are taken, and what the timing and performance implications are for getting applications and onlines back up and running.