December 1999
Network Testing: An Ounce Of Prevention...
BY VIC FORGETTA
When trying to ensure network QoS, an ounce of prevention is worth a pound of cure. The
key economic challenge for network managers is to determine when it is time to sprinkle a
few ounces of prevention into the network. The technical question is what kinds of
preventive measures can be deployed.
The answer to this question lies in proactive performance analysis. Proactive
performance analysis is aggressive testing of network components, both new and existing.
Network performance analysis lets managers create and generate the traffic of thousands of
network connected computers over a network or network device, then capture and analyze the
results to accurately measure network performance.
PERFORMANCE IS KING
No matter how much pounding it is expected to take, today’s network must deliver QoS
required for new applications, and it simply cannot fail. Increased dependency on the
network, support of intranets and extranets, and new applications such as VPN, voice over
IP (VoIP), and video conferencing place huge demands on network architecture. Many times,
the demands become too much, causing unnecessary delays for network users, degeneration of
customer service, and other problems.
Network managers must rapidly and effectively respond to these demands. With numerous
new technology choices designed to deliver network QoS, proactive testing is the only way
to determine how the network will react to new demands placed on it and what technologies
to use. While QoS is subjective, with the right tools it is measurable. The best way to
guarantee long-term QoS is through rigorous, standardized testing under a variety of
conditions at both the component and system levels.
One approach uses a methodology that defines all the measurements that collectively
determine the QoS of a network. It allows a network manager to evaluate the network at
several layers and measure vital parameters like latency and latency variation, throughput
packet loss and sequence, and examine the network’s ability to enforce QoS policy.
Aside from reactive testing (when users report problems), there are three times when
proactive performance testing is vital:
- To establish a baseline.
- When new applications place additional demands on bandwidth.
- When the mix of traffic is expected to change.
NETWORK IN TRANSITION
On top of a huge growth in data traffic, today’s networks are more complex. They use
multiple technologies, and they often grow to span hundreds or thousands of ports. The
market has also accepted the converged network, which must accommodate voice, video, and
data on a single system. New performance analysis systems and testing metrics make it
possible to predict when a network will fail and precisely which application is likely to
be the cause. This represents a giant leap forward in verifying a network’s
reliability and capability to provide QoS before going live.
Different elements of QoS are offered by different network layers and networking
technologies. In order to realize true QoS throughout the network, all of those pieces
need to work together.
When a Web-enabled call center adds extensive graphics or video to its customer service
mix, the stress on the existing network will be phenomenal. When a hospital decides to
send X-rays over its existing network, the pain will be felt by the network manager well
before the patient feels anything. Such a network must be tested proactively — ahead
of service deployment — to be sure that it can deliver the QoS required by the new
application without destroying the performance of existing applications.
TESTING ALL LAYERS
In addition to increased traffic, networks must also contend with the
delay-sensitive nature of voice and video. Knowing a network’s limits ahead of time
will save much user frustration down the road and keep the IT department’s good
reputation intact. A well thought out testing program like the multilayer approach to
performance analysis is designed specifically to address today’s complex set of
optimization and prioritization methods at all network layers. Such a program must
measure:
- VLAN, IP TOS, DiffServ, MPLS voice and video traffic flows.
- Calls and connections (signaling functions).
- Network applications (management capabilities).
- Transition of data between different technologies.
While the performance issues of Layer 2 LAN traffic are largely understood, the
performance loading of the new Layer 3 and 4 products are essentially untested.
Optimization of traffic can take place at Layer 2, Layer 3, Layer 4, and even up to
Layer 7. IEEE 802.1Q specifies a packet as part of a particular Virtual LAN, and 802.1p
assigns a priority to those packets to optimize traffic at Layer 2 (Datalink). DiffServ
and TOS prioritize, and MPLS and RSVP reserve resources for traffic at Layer 3 (Network).
A multitude of new QoS products, such as server load balancers and access control devices,
prioritize traffic based on criteria including TCP or UDP port number at Layer 4
(Transport). Even higher layer criteria such as URL or application type can be used.
Equally important are the verification of network connection capabilities, multicast IP
traffic behavior, firewall device performance under load and VPN load handling,
efficiency, and leakage in extreme loading conditions. Finally, there is the transition
between networking technologies as the data travels through the network. Since true
network QoS will require the cooperation of several technologies, all the network
components and applications at all layers must be tested.
Layer 2 tests of QoS are based on 802.1Q/p, where networks optimize based on
information placed in the VLAN tag in the Ethernet frame. A tester should determine for
VLAN performance (latency, throughput, and packet loss) and leakage across VLAN
boundaries.
Layer 3 testing verifies the proper operation of the different prioritization (DiffServ
and TOS, VLAN based on subnet or IP address) methods and reservation protocols, including
MPLS. It can help determine which protocols are receiving preferential treatment, evaluate
switch management performance, and determine how management overhead will affect QoS.
At Layer 4 we measure prioritization based on UDP or TCP port number, the
network’s TCP session capabilities, its ability to recalculate routes, and (for
applications such a Multicast I), how a system performs while processing multicast packets
and managing groups.
Above Layer 4, multilayer performance testing will generate different classes of
traffic (such as those required for voice and video) to determine how the network will
handle each under all possible conditions. Testing all network layers helps determine how
they work together to provide applications and users with the performance they need.
ASSURING QoS THROUGH THE NETWORK
To assure QoS, then, testing must also be done through the entire network. More firms are
migrating to service on public networks, often IP based. These networks are based on a
number of technologies, including DSL, cable modem, ATM, and SONET. Traffic typically will
be routed over a number of these before it completes its round trip. These transitions
must be tested to determine how, for example, an ATM to Ethernet conversion might affect
QoS, or what happens when the path changes in a packet-over-SONET backbone. Each piece has
to work with every other piece.
For the sake of security, network managers might want to add a firewall or VPN. A
loaded server will have slow response time, so load balancing may be required. For these
applications, throughput, latency, connection capabilities, scalability, and reliability
must be examined.
IT’S TIME TO START TESTING
Nobody denies that bandwidth is becoming more readily available, and as long as
the network is not crowded, response will be good. But economics dictates that management
will require IT staff to put as much traffic as possible on every link. Computer-based
education, multicasting, VoIP — these will all put nasty demands on bandwidth and
further stress the network. Unless the network manager has a baseline on performance, it
will be difficult to determine the effect of network upgrades or new technologies like
packet over SONET.
Voice traffic, while a priority, cannot be allowed to interfere with revenue-generating
transactions — like the sale of airline tickets or stock transactions. This requires
the network manager to have a firm understanding of which traffic deserves priority status
and which can live with best effort.
Proactive testing during the network design and deployment stage answers both the
financial and the technological riddles posed by today’s networks. It will save
millions of support dollars by avoiding service-related problems, and will allow
management to grow the technology in today’s network to meet tomorrow’s needs.
Vic Forgetta is director of market development for Netcom Systems, Chatsworth, CA.
The company’s SmartBits product, and its related suite of sophisticated software, are
among the leading network performance analysis tools available to network managers. For
more information, please visit their Web site at www.netcomsystems.com.
|
TMCnet LOGIN
Webinars
