Special Focus: Testing & Monitoring
Scale smart to balance traffic
In recent years, McLean, Va.-based Cvent has
seen a 40 percent to 60 percent annual
growth in traffic to its hosted, Web-based
applications. Providing the infrastructure
to handle this rapid growth-before customers
feel any impact-is a top priority for Dwayne
Sye, vice president of technology. "Whether
or not the end-users notice anything is the
measure of how well my team does its job,"
he explains. "Our customers are paying us a
subscription fee to use the service, so
there is an expectation that the service be
available whenever they need it."
"If we were still on the old
controllers, by now we would be getting
complaints from end-users."
As one of the largest event-planning
software companies in the United States,
Cvent assists marketing professionals with
more than 100,000 events each year. Using a
software-as-a-service model, Cvent provides
on-demand strategic meetings management,
online registration, Web surveys and
e-marketing tools to corporations,
associations, nonprofits, educational
institutions and other organizations. Its
customers include Microsoft, Verizon,
Goodwill and the U.S. Postal Service.
In early 2007, the growing traffic began
to tax the F5 BIG-IP version 4.x devices
that Cvent had installed in 2000. "The
legacy controllers served us well for nearly
seven years, but with our tremendous growth,
they suffered from over-utilization," Sye
says. "It was clear that an upgrade was
needed to support projected growth."
Sye also wanted to implement an
application-delivery infrastructure that
would help enhance the performance and
availability of the company's services. "We
needed a solution that would provide
geographic site redundancy and load
balancing in order to deliver a higher level
of service to our customers," he says.
In his search for a solution, Sye focused
on five primary factors: functionality,
performance, reliability, ease of management
and total cost of ownership. After a
thorough evaluation of more than 12 other
products, Cvent once again chose the BIG-IP
product line from F5.
The ability to minimize the impact on
customers during maintenance windows was a
critical differentiator, Sye explains. Cvent
uses the BIG-IP system to direct customers
to different servers during these windows.
"We can control how the site functions when
we do a code deployment or application
release," Sye explains, "and not all vendors
were able to offer that same kind of
seamless user experience."
Another key deciding factor was the
potential impact on available resources.
"Some of the solutions we looked at
definitely would have required full-time
network engineering resources dedicated to
managing just those devices, versus doing
something else that might be more beneficial
to our customers," Sye reports.
In the first half of 2007, Cvent
initiated a major network infrastructure
migration, during which it replaced the old
F5 devices with new BIG-IP solutions.
Compatibility testing performed prior to
deploying the BIG-IP devices, and
implementation guidance from F5, "ensured
that the migration was smooth and
trouble-free," Sye says. "The F5 engineer
convinced me we could do the job ourselves,
and we did."
In June, the team deployed the
application-delivery networking solution
they had selected: three F5 BIG-IP Local
Traffic Manager (LTM) 3400s and two F5
BIG-IP Global Traffic Manager (GTM) devices.
Sye estimates that the migration process
took eight to 10 weeks of preparation. "By
doing all the planning up front, we were
able to do the migration seamlessly and with
virtually no impact to the customer. We kept
the window of intermittent access down to
less than 15 minutes," he says.
The BIG-IP GTMs-one located at each data
center-optimize and route Cvent traffic to
the appropriate data center, based on
traffic type and volume. The BIG-IP LTM
3400s sit in front of the company's 12 Web
and application servers, the edge of the
Cvent application infrastructure,
intelligently managing and delivering Web
and e-mail traffic.
These application-delivery networking
devices also provide secure sockets layer
(SSL) acceleration to improve the
performance of the application servers.
BIG-IP SSL acceleration offloads
CPU-intensive SSL encryption and decryption
from Cvent servers, migrating it to the
high-performance BIG-IP devices, which are
designed to handle SSL transactions more
efficiently.
The devices run on F5's TMOS
application-delivery platform, which
includes an intuitive graphical user
interface and remote-management capabilities
that simplify management and administration
tasks for Sye and his team.
Cvent also uses F5 iRules
technology-customizable commands that offer
granular control over application traffic-to
aid in the collection of customer Web
statistics using a hosted, third-party
model.
Cvent's former method, which required
tagging each Web page with a script that
referenced a data collector maintained by
the hosted vendor, was problematic in
security-conscious corporate environments.
"Customers often blocked access to our
hosted statistics vendor, resulting in some
fraction of usage not being logged, and, in
some instances, the security restrictions
interfered with site functionality," Sye
explains.
To remedy this problem, Sye implemented
an iRule to redirect traffic directly to the
statistics vendor host. He notes this
resulted in "more accurate Web traffic
statistics for us, and better application
compatibility for our customers. The iRule
also saves us the hassle and cost of
deploying and managing a complex, installed
Web application statistics collection and
reporting infrastructure."
"When we first deployed the devices, our
requirements were probably not all that
sophisticated," Sye adds. "But now, as we
run into new issues or want to add more
functionality, one of the first things we
ask ourselves is, 'Can we solve that problem
with an iRule or other BIG-IP capability?'"
For Cvent, the gain in network
scalability remains the most important
benefit of deploying the devices. "If we
were still on the old controllers, by now we
would be getting complaints from end-users
about slow performance or dropped
connections," Sye says. "In contrast to
that, we've been on the new devices for
almost a year and the CPU utilization barely
registers.
For more information from F5
Networks
(click here)
by Bill Allen
IT organizations are under constant
pressure to roll out more applications,
support more users, deliver higher levels of
service and handle more intense processing
workloads-with budgets that rarely increase
proportionally to these escalating demands.
As a result, they are exploring the
virtualization waters in varying degrees.
Even the most enthusiastic adopters,
however, are only applying virtualization to
some of their new production machines.
Virtual server technology can introduce
new management complexities and challenges.
Some of these challenges arise because
virtual servers are managed in conjunction
with servers in the data center that are not
virtualized.
For many applications-especially high
input/output ones like e-mail-virtualization
is neither practical nor advisable.
Therefore, IT organizations may always have
to deal with a mix of physical and virtual
server resources. IT organizations are still
figuring out how and when to implement
virtualization, and adoption rates can
depend on many non-technical factors, such
as competition for IT budget and whether the
department can deliver the appropriate
support for virtualization.
Challenges that arise from virtual server
technology include:
Giving specific IT staff
appropriate privileges for specific virtual
machines. When multiple virtual
servers are running on individual physical
machines, the administration of these rights
becomes even more complex.
Maintaining management
assignments as virtual servers move from one
physical machine to another. To
avoid giving systems administrators and
application specialists access to virtual
centers (because of potential inappropriate
control over the data center's virtual
infrastructure), IT organizations often use
a manual approach for the distribution of IP
addresses and/or management URLs to the
specific virtual machines for which each IT
employee is responsible.
Accessing multiple virtual
servers connected to different virtual
centers. At this time, IT staff
cannot unify views of multiple virtual
servers residing on different physical
machines if the service consoles of those
different physical machines are connected to
different virtual centers.
Protecting many virtual servers
from component failure in a single physical
server. Physical servers have a
one-to-one relationship between component
failure and application outage. When
multiple applications are running on that
same server, the stakes go up. IT
organizations implementing virtualization
have to be sensitive about the
vulnerabilities created by multiplexing
large numbers of virtual servers onto a
single physical machine, and should be sure
they have the out-of-band access to those
physical machines necessary to address BIOS-
and hardware-level issues.
Exclusive use of Microsoft Active
Directory. This can be problematic
for IT organizations that need to drive
management access using technologies such as
RADIUS and single sign-on in order to
support their broader
infrastructure-management architecture.
While understanding how virtualization
will impact management operations before
they are implemented can be difficult,
solutions that deliver integrated management
functionality across both physical and
virtual server platforms are available.
Bill Allen is director of product
management, Avocent Corp.,
Huntsville, Ala.
For more information
(click here)