3Com NEWS:
-------------------------------------------------------------
techweb.com
Target Your Apps
By Amy K. Larsen
In this unpredictable Webbed world, getting core business applications
to hit acceptable service level targets is no simple task.
Managing multiple protocols while the company migrates to IP is
tough enough. Trying to keep costs under control at the same time
really tests your skills.
So, if you expect to get your intranet and E-commerce projects moving
forward, it's more important than ever to define and demonstrate a
bottom line impact on the business. Fortunately, a new class of tools
that measures end-to-end IP applications has emerged to compile
availability statistics and clock application response times for highly
distributed networks.
But tools aren't enough. It's up to you to target which apps are
business-critical and, ultimately, to define acceptable service levels.
"In every industry-but especially in established industries like
utilities-cost containment is essential," says Bryan Bates, president of
Supportnet Consulting Inc., a Calgary, Alberta-based firm that
specializes in application management for utilities. "Business and the
IT groups have to communicate what their priorities and resource
limitations are."
Honeywell Space Systems in Phoenix, a unit of $8 billion Honeywell
Inc., takes this lesson seriously. The company's space technology
products division is bent on deploying Oracle across its intranet in a
way that's meaningful to the business.
The division's first step was recognizing that with users accessing
Oracle apps from distributed servers across the corporate WAN,
response times were almost guaranteed to be inconsistent. "We needed
a way to qualify and quantify performance," says Boyd Conner, a
network manager with the space technology products division.
Conner's goal was to identify slowdowns before they had a negative
impact on operations and to prove how IT is making it easier for
employees to make queries and send information. In the past,
Honeywell's support team relied on protocol analyzers to capture
traffic data, identifying the app by port number. Conner says that
approach delivered inconsistent margins of error, so he looked
elsewhere for a tool to fill the void.
He was looking for a way to measure overall response time, but he also
wanted to break out network latency and factor in information like
network volume to get an accurate performance profile. "There are
tools out there to tell how long a query takes at the server, but we
wanted to be able to see what the total turnaround time was," Conner
says.
Honeywell's search led to Ecoscope, a turnkey solution from
Compuware Corp. that compiles app performance and network
utilization statistics. Ecoscope uses a series of probelike devices that
listen in on the network, capturing traffic data as packets traverse the
network.
The application totals point-to-point response times and looks at the
impact the application has on network resources. The Compuware
product also automatically recognizes hundreds of apps-a key feature
Conner says helped Honeywell get it running quickly.
Conner says although his group has some "generic" service level
agreements in place, it just passed the infant stage of the Oracle project
and still has a way to go before it has a full-fledged service level
management program in place.
The space technology division's short-term goal is to define acceptable
performance levels and identify potential fixes.
The idea is to improve service levels continually, increasing
productivity and reducing costs-delivering results the business side can
understand.
Happy Together
Using tools to set metrics and define service levels is important. But
nothing moves forward unless corporate IT is working in step with the
business side.
Granting business groups ownership of technology gives them direct
responsibility for equipment and a better understanding of the resource
limitations IT faces. This alliance also gives IT personnel insight into
how they support business processes--and which ones take precedence.
In turn, that helps network managers set goals, devote technical
resources and guarantee recovery time for high-priority applications.
"Getting the business and IT sides talking has helped us set a better
expectation of performance because both sides understand the
limitations from the beginning," says Alan Nguyen, a senior network
management engineer for $5.6 billion 3Com, which is spearheading
service level contracts for all new transaction-based apps on its internal
network.
An increasing number of businesses are even starting to employ
application managers, technical personnel responsible for the
performance of a few corporate applications, according to Steve Foote,
vice president of research strategy at Hurwitz Consulting Group Inc.
Although these managers have roots in the IT organization, they receive
bonuses from individual business lines based on application
performance.
These managers apply technical knowledge to assure service quality.
That includes both pre-deployment planning and post-production
follow-up.
"For successful application service level management, you need to
cover both the micro- and the macromanagement, and not get caught in
between," Bates explains.
Macromanagement means anticipating how much of the enterprise
resources an application will consume and how well it will perform
once it's deployed.
Bates points out that to anticipate an application's resource
requirements and predict response time, application developers and
capacity planners need to test applications adequately prior to
production.
Bates uses Application Expert from Optimal Networks Inc. to estimate
response times based on different bandwidth allotments and other
resource variables.
Micromanaging applications amounts to keeping a tight handle on
availability and response times once they're in production. That also
includes recognizing falloffs in performance before they impact
business processes.
Of course, benchmarking performance is an inexact science. "It's
difficult to project what type of response time we should be getting
without having some kind of baseline to compare it against,"
Honeywell's Conner says.
Bates faces similar dilemmas. "By itself, all the modeling and
simulation in the world can't tell you exactly how an application is
going to perform on the production network or how it will impact
other resources," he says.
He cites the case of a utility client that tested an application in the
development phase and deemed its 14- to 20-second transaction
response time adequate for the business process it was running at the
time. Once the customer rolled out the application on the network,
response time rose to an unacceptable minute-and-a-half average.
After considerable analysis, Bates discovered that none of the necessary
drivers were resident on the user's NT workstation, so every time a
client machine initiated a request, the machine had to download the
drivers from a network server.
Today, most companies are focused on basic metrics. Only a few, such
as 3Com, actually are writing agreements.
"We draft an official document that all the managers from the business,
data, network and application sides sign off," Nguyen says. "Basically,
we don't roll out a new application now without doing one."
3Com is using VitalSuite, a set of performance tracking and analysis
products from VitalSigns Software Inc., to track and diagnose
response-time issues with critical business applications such as
PeopleSoft and SAP R/3.
Performance Results
3Com's network operations center monitors intranet application
performance at the networking company's headquarters in Santa Clara,
Calif., and its branch offices in Boston, Chicago, Singapore and
Europe.
Nguyen reports an overall positive result from the SLAs so
far-including a quicker resolution to long-term performance issues
such as insufficient bandwidth allocations. However, he does offer
some criticism of the nature of the available tools.
"There isn't any fully integrated product for planning and
management," he says.
As a result, his team stitches together data from the VitalSuite,
homegrown tools and products such as Make Systems Inc.'s Netmaker
XA for bandwidth impact analysis. Nguyen also says there are no real
commercial applications for automated problem resolution.
In the meantime, network managers will continue to cobble together
data from their apps to define acceptable service levels and the
resources required to achieve them.
The reality is that although tools are improving, there's no single
product out there that will do it all for you. If you really want to
demonstrate corporate IT's value to the business, be prepared to roll up
your sleeves-and target your apps.
SIDEBAR: Four Ways To Measure Performance
Network managers typically use four basic approaches to measure and
assess application performance: ghost tran-sactions, point-to-point
packet inspection, client-based agent measuring and application
response measurement. Here's a quick rundown of each:
- Ghost transactions. App developers can apply application simulation
products that use ghost transactions to mimic activity and record
response times. Products like Chisel from Network Tools Inc. emulate
actual transactions and are largely used for stress testing or capacity
planning.
- Point-to-point packet inspection. This approach, used in products
like Ecoscope, monitors packets as they travel between network points.
Although it's an easy way to track apps, there is a drawback:
Point-to-point packet inspection can miss the client side of the
equation.
- Client-based agent measuring. Applications like ETE Watch from
Candle Corp. and VitalSuite from VitalSigns Software Inc. equip client
workstations and PCs with software agents that clock response times.
The advantage to putting a timing device on the client side is seeing
performance from the user perspective-a metric IT managers can align
with business productivity. This approach also offers a more granular
view of query and transaction response times.
- Application response measurement. ARM is a set of application
program interfaces that reports performance data back to a management
application. By using the APIs, an application can leave a trail of its
activity and compliant software products can then determine the
specific path taken by each request to get a read on response time.
-Amy K. Larsen |
