In-Depth
Brave new performance management world
- By Jack Vaughan
- January 1, 2004
In recent years, mid-range servers have been wheeled into and out of IT shops
like so many refrigerators at one of Crazy Eddie's appliance stores. In the
heady days of e-commerce, ''tuning'' these servers to run optimally usually meant
wheeling in additional machines, many of which ended up on eBay when the
Internet gold mine dust settled.
Looking on morosely from the alcoves were the mainframe operations folks.
These are the people for whom capacity planning and performance management
became a science. In the rush to client/server computing, and then e-commerce,
the hard-won skills of system virtualization, performance modeling and workload
characterization were somewhat cast aside.
But skills that were cast aside are in demand once again. Unless there is a
fabulous spending bust-out a la the first Internet era, IT and app development
managers will be compelled to make the most of what they have.
That means a new emphasis on the kind of careful research spread by the
Computer Measurement Group (CMG), which, at its yearly meeting in Dallas last
month, saw an abundance of new papers on performance-related issues as they
applied to Java- and .NET-class platforms. At the same time, the automation of
performance monitoring and management is a special focus for vendors with roots
in mainframe systems management, as well as for players with origins in software
test and development.
The goals are to have better views into systems; better indications of
problems' root causes; better breakdowns between network, database and CPU
problems; a better sense of how the user perceives system performance; automated
monitoring of performance vis-a-vis service-level agreements (SLAs); and,
sometimes, more intuitive management cockpits that allow chief technology
executives to see how fully a company's technology portfolio is being
exploited.
Vendors in the space are plentiful and assorted. Count among them BMC,
Candle, Computer Associates, Compuware, Hewlett-Packard, IBM, Ixia, Mercury,
Microsoft, Nexvue, Parasoft, Qwest, Segue, Shunra, Sun, TeaLeaf, Veritas, Wily
and others.
Candle targets WebSphere
Throughout software history, vendors
have made hay in the aftermarkets of IBM. Among these is system management
specialist Candle, which has built out its traditional mainframe and distributed
systems expertise to now include Java servers. Recently, the company added Path
WAI packages that accelerate IBM WebSphere J2EE EAI initiatives.
In recent years, said Steve Lindeman, Candle's VP of services, ''a lot of
applications have been built, and a lot of servers rolled out without a lot of
best practices.'' Most of these ''new'' apps are still pulling data from mainframes
in one way or another, Lindeman noted. A lot of these applications were also
built with limited consideration for the impact of the application that would
eventually need to pull information from subsystems, he added.
Lindeman said Candle's push for WebSphere performance management grows out
of its background in mainframe and MQ System management. ''We call it the
WebSphere Whole Life Cycle,'' he said. ''Too often, development groups are focused
on the application and not on the whole life cycle.''
As customers add applications, they sometimes hit a wall of scalability. The
tendency, said Lindeman, is to blame the new application. But in a recent
encounter, he said ''when we analyzed the system, it wasn't a problem with the
second app per se, but a problem with best practices. The developers hadn't been
thinking about the application life cycle, how the second application tied into
the system.''
Vantage meets SLAs
Compuware has bought a number of performance
point tool vendors over the years, and continues to enhance its offerings for
the new age. ''The challenges have evolved from a server-centric focus to a
network-centric focus; but now the emphasis is shifting to services delivered to
the business, rather than just on the infrastructure,'' said John Williams,
launch director for Compuware's Vantage 9 application service management
software.
The new product now provides client-network-server time breakdowns with
baseline comparisons, exception-based analyses that allow operators to measure
network behavior through the course of a transaction, and comparison views
replete with English language interpretations of performance bottlenecks.
One of the challenges of the distributed environment is that it is
heterogeneous, Williams said. ''You can deliver deep functionality on a single
architecture like J2EE, but you have to be able to deliver a solution that spans
an entire software portfolio. As well, the industry needs to focus not just on
the needs of the business, but allow technical people to drill down and solve
problems.
''The help desk is still too often the monitor,'' he chided.
One way Compuware is meeting evolving performance needs is by directly
monitoring the end-user experience -- something that is too often forgotten in
the push to monitor. SLAs can truly prove useful only if the user experience is
correctly characterized. Too often, SLAs are met at the same time that trouble
tickets increase.
''You have to talk to the business and figure out which applications are the
most important,'' explained Williams. Where you start is important in managing
enterprise application portfolios.
''We have an integrated approach to service analysis,'' maintained Williams.
With Compuware's Vantage 9, this goes beyond the traditional notion of technical
troubleshooting, he indicated. In the future, when operators see degradation,
they will drill down and such problems can be tackled on a systematic basis. The
goal is to characterize problems, perform intelligent handoffs and give
technical staff deep analytics, Williams noted.
Mercury, the messenger
Mercury Interactive is notable among
performance software players, as it has parlayed its early test tool
near-hegemony to even greater benefit in the operations space. Last year, the
company began to push the idea of Business Technology Optimization (BTO) as an
answer to the problem of exploiting and justifying technology expenditures --
more than just a niggling problem for today's CIOs. At the same time, it pursued
significant mergers (with Performant and Kintana) to improve its ability to
drill down into Java problems, and to roll up software portfolio statistics to
management for review.
''The idea of BTO is to optimize technology to meet the needs of business,''
said Rajesh Radhakirshnan, director of product marketing, Mercury. ''In essence,
it is making sure the technology really delivers what it needs to and, if it
doesn't, using diagnostic capacity planning to optimize it to make sure that it
does.''
Radhakirshnan agreed that mainframe practices have been slow in coming over
to the Web systems space. ''In the mainframe world they understood how to do
capacity planning. But there are no standard practices in the Web world.''
Inter-team communications is growing, said Radhakirshnan. ''The testing team
is now working more closely with the capacity planning team and with developers
to identify performance problems.''
''Testing, planning, tuning, diagnostics -- we are trying to put these
together with the best practices,'' he said. Central to Mercury's effort, he
indicated, is Performance Center. Look for this software to be further enhanced
in months to come with a Delivery Dashboard, based in part on technology
acquired along with Kintana.
''It shows the goals -- for example, response time for business transactions,
or the number of supported concurrent users -- to the whole organization,''
Radhakirshnan said of the dashboard.
One difference in today's environment is a higher rate of updates. With
upcoming performance technology such as Mercury's, Radhakirshnan said, groups
can now more carefully prepare and compare patches.
Veritas: True to form
Backup and operations software powerhouse
Veritas updated its own diagnostic portfolio last year with its purchase of Java
performance specialist Precise. The fruits of that merger should become more
apparent in the months to come.
Capacity planning methods of the past have limits in the present day, said
Tom Murphy, Veritas. ''We would not be in business if people could resolve
problems just using load generators. People can't anticipate,'' he said.
''There is a gap between what is architected, what is developed and what
happens in production,'' Murphy explained.
''We identify the shortcomings of something that was architected,''
said Murphy, who hails from the former Precise group within Veritas. He is
familiar with network and system modeling capacity planning alternatives, but he
said what people really want is ''a quick way to react vs. capacity planning aspect
modeling.''
Precise's claim to fame was its ability to correlate from tier to tier; for
example, to follow SQL statements from the database through JDBC to the J2EE
server, said Murphy. Just looking at a database does not do much good these
days.
In upcoming releases Precise optimization software is likely to benefit from
better links to other Veritas software suites. Look for Precise diagnostic data
rollups into the Veritas Command Central Service suite. Look for connections to
business reporting software as well.
Note that .NET tools, not just Java optimizers, are likely to be part of the
mix at Veritas as well as among other major third parties. In addition,
Microsoft is not sitting on the sidelines as Java servers gain performance
enhancers. The Redmond, Wash.-based giant is prepping an Enterprise
Instrumentation Framework set to appear with its Everest software suite.
User story: At West Bend Mutual
System-level controls were of
mixed value when it came time to measure the user's experience, but updated
tools proved something of a cure at West Bend Mutual Insurance of West Bend,
Wis. There, Craig Walker, enterprise architect, used Compuware Vantage tools to
pinpoint outages and degradation before they became trouble tickets.
''I handle both infrastructure and application development. I meet with the
designers for every product,'' said Walker, who said that he writes the technical
recommendations on how to accomplish systems for a company that has an
environment composed of Z/OS mainframes, Windows 2000 servers and some RS6000s
running AIX. Languages used include Cobol, ASP and .NET.
''A lot of solutions in the past were technology-driven. Now we have to be
much more frugal in our solutions. We need to look for new and better ways of
doing things,'' said Walker. ''Part of that is SLAs, which show the value of the
services that are most important to our end users.'' Walker and crew worked with
these end users to define those apps that were the most critical to them.
''Three or four years ago, we put in place a nice service management process
with our business units. We defined certain classes of applications to establish
what was acceptable in terms of downtime,'' he said. However, Walker and company
needed to keep track of how they met those agreements.
''We could monitor routers, servers and CPU usage, but it was hard to
translate that into terms our business users could relate to. We had to jump
through hoops to establish availability, and it wasn't a true measure of what
our end users were experiencing,'' he explained.
With Compuware Vantage tools, Walker said, he gained a nice communication
tool for reporting back to users how IT was performing. A more effective
end-user monitor was especially needed.
''We wanted to see things from an end-user perspective,'' he said.
The data provided by Vantage helped to isolate trends in application response
time. In monthly cross-organization meetings, Walker is now able to point out
trends that are going in the wrong direction.
''I can ask developers to fix them before we cross that [SLA] threshold. The
information is fed back to the tech staff, and [then] fed on up to the business
staff,'' he said.
Walker sees room for improvement in improved drill-down from problems to
causes, and this, in fact, is a key part of Compuware's Vantage 9, which was
released last month.
''Compuware, in the past, had been a very loose collection of tools,''
commented Walker. ''What I see in its next generation is more of an integration
of the tools into a service management process.''
Also see:
NexVu's Herriman on remote performance management - 2/10/2004
Room for improvement in Java app performance - 1/13/2004
Load emulation for better deployment results - 1/6/2004