A t the
end of the day, your enterprise network could make you tear
your hairs out ...
The rapid
development and acceptance of network technologies and applications
to gain a competitive advantage introduces significant traffic
management challenges. Keeping the network "up" is
no longer sufficient: the network must be an active component
in providing enterprise connectivity. Administrators must deliver
predictable and cost-effective performance levels for a diverse
set of corporate stakeholders. With organizations relying on
the Internet and corporate intranets to support core business
processes, network performance has become a bottomline business
metric. In addition, enterprise traffic issues may also impact
the integrity of the corporation's network security.
Without
consistent and reliable performance levels, the network cannot
provide the necessary infrastructure to deliver valuable information
to employees, suppliers, and customers. The result may be missed
business opportunities and lost revenues. The following are
a few examples of network-based performance problems and their
results:
- Customers
cannot access information on the company's web servers quickly
enough, resulting in lost business
- E1/64
kbps leased links are added to accommodate rising bandwidth
needs, but without a noticeable improvement in service
- Each
new network application results in a precipitous drop in performance
- Consolidation
of servers across the backbone results in performance degradation.
In addition
to causing serious business and productivity problems, enterprise
traffic issues may also impact the integrity of the corporation's
network security. All too often, when individual users become
frustrated with unreliable service, they attempt to establish
Internet connectivity outside the corporate infrastructure -
thereby compromising the enterprise security policy and putting
the entire network at risk. Enterprise traffic management is
an on-going challenge because it encompasses a number of variables.
Network
Performance Variables
Some of the variables that affect the performance of enterprise
networks include:
Criticality
Organizations depend on the corporate network to support core
business processes, such as accounting, sales, and operations.
Poor network performance can cripple these processes, dramatically
affecting business productivity. Criticality is magnified as
network resources are extended to external users, through an
extranet or by using the Web as a vehicle for advertising, product
information, and direct sales. These scenarios bring many more
users into the enterprise network environment, placing additional
stress on network performance. More than just a measure of availability,
criticality today involves performance requirements such as
response times, file transfer times, and network latency.
Complexity
The deployment of intranets and extranets adds significant complexity
to enterprise networks. Network administrators must manage traffic
not only on the backbone and on internal network segments, but
also at WAN access points. Matching the data rates of local
area networks (e.g., Ethernet or Fast Ethernet) to the limited
capacity of a corporation's wide area connection introduces
a potential performance bottleneck. In addition, the growing
number of Internet-based applications and services creates additional
performance demands. The wide range of content importance requires
dynamic traffic management to balance conflicting demands and
ensure adequate levels of service.
Return on Investment (ROI)
Over the
last decade, corporations have made massive investments in network
infrastructures. As with any corporate resource, this investment
must be actively managed to capture its full benefit. By effectively
managing the performance of LAN and WAN infrastructures, organizations
not only can support critical applications, but also can extend
the life of the network and maximize ROI. It is important to
note that organizations should avoid a "design by hope"
strategy that throws excess infrastructure and capacity at performance
problems in the hope of alleviating the problem. Too often,
the result is that the points of congestion or performance degradation
are shifted, rather than resolved. Dynamically managing the
performance of the existing network can obviate or defer the
need to make additional investments in hardware, bandwidth,
or server capacity.
Dynamism
The dynamic nature of networks creates additional challenges
to providing consistent and reliable performance levels. In
today's Internet-centric network, traffic patterns change rapidly
and unpredictably. This is becoming especially true as Web-based
services proliferate. The availability of new content on the
Web can shift traffic patterns and volumes dramatically in very
short periods of time. Other emerging technologies, including
audio and video streaming applications and "push"
services, add to the dynamic condition of networks. The inelastic
nature of real-time streaming requires guaranteed network performance.
Without adequate traffic management, the bandwidth necessary
to provide continuous streaming will not be available.
Similarly,
push applications typically transfer data in large bursts across
the network, rapidly consuming limited network resources. An
effective traffic management solution can assign appropriate
privileges for these applications so that more important services
or users are not starved of network bandwidth. In dynamic environments,
a network's performance history cannot be used to predict future
network needs. Traffic management for dynamic networks must
be policy-based, provide coverage for the entire enterprise,
and must actively provide analysis and feedback so the network
can adapt to changing network conditions. An effective traffic
management solution allows network administrators to balance
these often conflicting variables and provide a predictably
high quality of service.
... Get
this done right away, if you have to survive ...
Setting a Network Equilibrium Point
In order to provide a consistent quality of service for users,
network administrators must establish a "network equilibrium
point" that specifies a target state for the traffic management
system to maintain. A stable network equilibrium point gives
administrators a level of network control not possible with
traditional enterprise management approaches. For example, many
earlier traffic management tools simply "reserved"
a fixed amount of bandwidth for a given application or service.
Unfortunately, a reservation is a static control mechanism and
cannot adapt to the dynamic behavior of real networks. Maintaining
consistent network equilibrium requires dynamic and adaptive
control mechanisms. Achieving network balance also extends the
useful life of the current infrastructures by avoiding the need
for continual expensive upgrades. Current capacity and resources
are efficiently utilized to maximize return on investment. Network
equilibrium points are defined with management policies that
drive the traffic management system to deliver stable service
levels. Traffic management policies dynamically balance conflicting
network demands and optimize overall network performance.
... And
to be sane, follow up with ...
A Traffic Management Solution
Effective traffic management requires the close integration
of multiple control and analysis elements, including instrumentation,
bandwidth management, load balancing, and network layer tools.
All of these elements are integrated by the enterprise traffic
policy, which defines how the network should react to changing
conditions.
Policy-based
Policies are the rules, or constraints, placed on the traffic
management system. They define the decision-making criteria
and integrate all traffic management components.
Because
every organization has different requirements, the policy definition
should be flexible, allowing granular control of network traffic.
For comprehensive enterprise management, performance policies
should integrate with a corporation's security policy. Network
equilibrium can only be achieved when traffic is managed across
the entire enterprise, with a single traffic management policy.
This enterprise-wide policy should be centrally managed and
capable of leveraging a global directory service that provides
real-time information on network users and resources.
Centralized
management minimizes administrative burdens since the policies
are defined once and automatically distributed to multiple enforcement
points. Administrators are spared the drudgery of repetitive
configurations and gain the confidence that network performance
is managed coherently with a single policy. Enterprise-wide
policies are also more effective and secure than trying to synchronize
a set of independent management tools.
Furthermore,
policy-based traffic management must support the automatic distribution
of a single policy to multiple enforcement points, typically
all Internet and intranet access points. The enforcement points
provide active implementation of the policy by continually inspecting
traffic and utilizing real-time feedback mechanisms to maintain
network equilibrium. Logging of traffic flows and resource usage
also occurs at all enforcement points. Traffic management policies
must be defined to support dynamic network conditions.
For example,
by anticipating large bursts of discretionary network traffic,
the administrator can establish guarantees that ensure the availability
of bandwidth for important applications.
Traffic
inspection-based
Traffic management requires detailed information on all network
traffic. The various mechanisms which regulate network performance
require comprehensive traffic information in order to make intelligent
decisions consistent with the traffic management policy. Traffic
inspection should not only determine the source and destination
of a network communications, but it should also yield information
on application type, direction, and connection state.
Application
type-Flexible traffic management policies require that traffic
be classified by Internet service or application in order to
deliver the appropriate level of network resources for that
service. With this capability, applications with critical bandwidth
requirements can be granted priority over low priority services.
Direction-When
the traffic management policy applies different control criteria
to inbound and outbound traffic, the ability to identify traffic
direction becomes critical. For example, if an organization
supports electronic commerce, inbound HTTP traffic will typically
require a higher priority than outbound HTTP traffic.
State information-Arguably
the most critical aspect of traffic inspection is the ability
to derive state information from an individual IP packet. Understanding
the state of network traffic is crucial to applying the appropriate
management policy. Discrete packets yield little useful information
because they cannot be associated with an established or newly
requested network connection. State information allows the traffic
management solution to manage each packet in relation to the
entire transmission. Using this state information, traffic management
can be successfully applied to individual network connections.
Instrumentation-based
Network instrumentation provides the raw data for the traffic
management system. This data must then be analyzed to provide
pertinent information on network traffic flows.
With the
proper instrumentation and analysis tools, network administrators
can characterize traffic mix by application, heavy consumers
of network bandwidth, and the variations of traffic. This information
is necessary to establish the desired network equilibrium and
define the appropriate policies. Instrumentation also provides
the ongoing feedback to drive the adjustments that maintain
network equilibrium.
Bandwidth
management-based
Heavy Internet and intranet usage can result in network congestion
at a corporation's network access points. Even at moderate traffic
levels, the desired network equilibrium point can be upset and
the performance and availability of critical network services
can be degraded. Typically, this is a result of incorrectly
assuming that all network traffic has equal priority, which
is most often not the case. Furthermore, chasing the problem
with more bandwidth does not address the root problem, because
traffic will invariably increase to fill the available bandwidth.
The only viable solution is to actively manage bandwidth resources.
Policy-based bandwidth management can alleviate network congestion
by dynamically controlling traffic at all Internet/intranet
access points. The bandwidth policy defines the privileges granted
to specific users and applications.
Bandwidth
management solutions classify inbound and outbound traffic according
to user-defined classes, and then schedule traffic for transmission
based on its relative merit or importance. There are a number
of important points to consider when choosing and deploying
policy-based bandwidth management solutions.
Level of
bandwidth control-Bandwidth management should be provided on
an aggregate level, allowing the network administrator to establish
an equilibrium point that allocates available bandwidth based
on entire classes of traffic, not just individual connections.
For example, guaranteeing a minimum amount of bandwidth for
each HTTP connection will lead to problems if the number of
connections doubles suddenly. The preferred approach is to manage
traffic at an aggregate level that allows the administrator
to define the proportion of available bandwidth granted to all
HTTP connections. When bandwidth is controlled at an aggregate
level, variations in HTTP traffic impact other Web connections,
but bandwidth availability for other traffic classes is unaffected.
Control
criteria-In the past, bandwidth management meant establishing
limits and guarantees. An example of a guarantee is the Committed
Information Rate (CIR) for Frame Relay networking services.
CIR is a minimum bandwidth guarantee that an application or
user can expect to receive.
Limits can
also be placed on the maximum amount of bandwidth that a network
service consumes on a network link. Bandwidth guarantees are
useful for smoothing the burst-and-delay effect inherent in
most Internet traffic. Network users will observe an improved
quality of service due to the allotment of guaranteed bandwidth.
These static
mechanisms alone, however, do not provide a complete bandwidth
management solution for most networks. Weighted priorities are
necessary to support the dynamic traffic mix found in most network
environments. Weighted priorities allocate bandwidth based on
merit or importance in relation to other managed traffic. As
an example, outgoing Web traffic that is sent to satisfy external
requests can, if desired, be deemed twice as important as incoming
FTP traffic. When bandwidth resources are oversubscribed, the
weighted priorities will ensure that the ratio of outgoing Web
traffic to incoming FTP traffic be accurately maintained at
a 2:1 for all available bandwidth resources.
Unlike absolute
priorities, weighted priorities ensure that low priority traffic
is not starved of network bandwidth. Weighted priorities provide
an intuitive method of defining traffic policies and ensure
that even low priority traffic receives some bandwidth on a
consistent basis.
The combination
of weighted priorities, limits, and guarantees offers a new
dimension of flexibility for the network administrator. For
example, a specific Internet service may be assigned a large
weighting indicating a high criticality level. At the same time,
the service may receive a low bandwidth guarantee recognizing
that there is a minimal bandwidth requirement for basic operation.
When bandwidth
management is deployed at a WAN access point it must be able
to support not only all necessary applications, but also the
existing security functions as well. Foremost among these are
data encryption and network address translation. Data encryption
is a fundamental component of Virtual Private Networks (VPNs)
and secures the transmission of data across public networks
such as the Internet. Unless the bandwidth management solution
can intelligently handle encrypted data, it is useless in managing
bandwidth on the large volumes of traffic carried by VPNs.
In many
organizations, the number of users who require access to the
Internet exceeds the number of legal, or registered, IP addresses
assigned to the corporation. Network address translation can
overcome the IP address limitation by translating internal addresses
to one or more registered addresses. This is a very popular
implementation and must be supported by any bandwidth management
solution. Without this support, bandwidth management cannot
be deployed at any WAN access point.
Load-balancing-based
The ability to maintain a user-defined network equilibrium is
dependent on a robust network infrastructure. When a single
application server is relied upon to support a high volume of
client/server traffic, users may experience poor response times
or even connection timeouts. Such unreliable connectivity can
result in missed business opportunities.
Server load
balancing enables a single application server to be replaced
with a logical pool of servers sharing a common IP address.
All application data is replicated so that any individual server
can fulfill client requests. Load balancing permits incoming
connection requests to be shared among all of the available
servers. The deployment of multiple servers eliminates a single
point of failure in a server infrastructure and enhances the
reliability and availability of network resources.
Not only
do users experience noticeably improved response times, but
corporations are spared the expense of continually upgrading
to more expensive servers to meet the increasing demand for
service. Existing hardware can be utilized to increase the capacity
and robustness of a corporation's server infrastructure.
Flexible
load balancing solutions provide multiple balancing algorithms
to meet the specific requirements of an organization. For example,
algorithms are available that distribute connection attempts
to the server experiencing the lightest load. This method evenly
balances the load on all servers in the group.
To support
enterprise networks where network users are geographically dispersed,
incoming traffic can be directed to the closest server based
on domain name. This enhances response times for end users.
Other load balancing algorithms enable connection attempts to
be distributed to individual servers based on round trip delays,
round robin schemes, or random server assignment.
Server infrastructures
can be enhanced with load balancing to help corporations achieve
an optimal network equilibrium point.
... Finally,
you can look at these too.
Network-layer Management Tools
Managing traffic flows at the network layer offers several opportunities
to enhance overall network performance. Traffic can be inspected
at the network layer to extract complete state information and
provide optimal performance. Network-layer traffic management
also provides the flexibility to support a broad range of applications,
including compression and caching.
Network
Compression
Data compression can be performed at the network layer to reduce
the raw volume of data, thus improving performance and better
utilizing network resources. The key to data compression is
selecting efficient algorithms which introduce minimal latency.
Network
Caching
Network-layer caching increases performance by making frequently
requested information available as a local network resource.
Time consuming transactions between the client and the content
server are minimized reducing traffic flows at possible points
of network congestion.
For example, an administrator can cache multiple pages of a
remote Web server that is accessed frequently. Users experience
faster, smoother response time and bandwidth is conserved for
other activities.
Enterprise
traffic management is becoming an increasingly important issue,
especially as more and more corporations build Internet and
related technologies into their core business processes. With
the rapid development and adoption of new applications that
require large amounts of bandwidth, corporations must ensure
that they protect the performance of their business-critical
applications, whether they be client/server applications such
as Oracle or SAP, in-house developed applications, or Web commerce
applications. To do this, each corporation must evaluate its
traffic management requirements and implement an effective enterprise
traffic management solution that is flexible, centrally managed,
and provides the performance and access that customers and employees
need.
This
is based on a white paper on enterprise traffic management by
John McConnell, president, McConnell Consulting Inc.
(extracted from a white paper
written by the author)