HOME > Research > Traffic engineering > Principal research results
Traffic engineering
[Research results]
- Research result 1:
Traffic control techniques - Research result 2:
Anomalous traffic detection techniques - Research result 3:
Network performance evaluation/monitoring techniques
Research result 1
Traffic control techniques
With the appearance of content distribution techniques such as P2P and CGM (customer generated media) and the rapid growth in access to video-sharing sites, content is being distributed in greater quantities and in richer forms than ever before, and new content distribution applications are always appearing. Based on these trends, it is expected that service quality requirements will become more diverse and wastage of network resources will become more of a problem.
In the Traffic Engineering Group, we are therefore researching techniques for making traffic controllable. These techniques modify traffic to achieve a traffic flow that is better for network service providers and users by controlling the communication time, path routing, call destinations and communication modes within the range of quality demanded by users.
These techniques are visualized in Fig. 1. Network traffic and quality information are used to control streaming traffic, for example, so that it used better quality routes, while file-sharing traffic where throughput is more important is transmitted in vacant routes and vacant time bands, thereby implementing efficient forwarding of traffic according to the required quality. Our research results show that it is possible to substantially improve delay times by using quality information to control the routing of traffic.
In the future, we will continue to investigate methods for implementing more efficient and higher quality traffic forwarding by caching identical content within the network.
Fig. 1: Overview of technologies for making traffic controllable
Click on image for larger version.
Research results 2
Anomalous traffic detection techniques
To provide network services that are more reliable and dependable, it is essential that network anomalies capable of causing a pronounced degradation of the user's communication environment are promptly detected and suitably dealt with. Conventionally it is the operator's job to monitor the network for anomalies, but as the scale of networks increases, it is becoming harder to detect anomalies promptly due to the growing number of monitoring points and monitoring conditions.
At NTT Laboratories we have developed a dynamic threshold setting technique that automatically detects anomalies and reports them to an operator based on the amount of network traffic observed at multiple monitoring locations and under multiple conditions. This technique accurately predicts the current quantity of network traffic by statistically learning the characteristics of past fluctuations in network traffic. Then, by comparing the predicted values with the actual measurements, it detects network anomalies such as increased traffic levels such as DDoS attacks and reduced traffic levels caused by equipment failure. By continuing to predict normal traffic levels during anomalous situations, it can also determine whether or not the anomaly is continuing. In this way, by discriminating between momentary traffic fluctuations and serious long-term anomalies, it can provide operators with additional information such as the number of locations where anomalies are currently occurring. This was not possible with conventional techniques that only reported on sudden changes in traffic levels.
In the future, with the aim of expanding the information provided to operators, we will investigate schemes that incorporate analysis techniques to identify the causes of anomalies, and control schemes that can automatically apply first-aid measures to the network.
Fig. 2: Overview of anomalous traffic detection system
Click on image for larger version.
Research result 3
Network performance evaluation/Network configuration audit
Based on the router and switch configuration and traffic matrix information in an actual network, we can make the network simulation model on the computer as the virtual network environment using OPNET. In the network simulation model, the actual routing operation like OSPF, BGP is simulated accurately. Using this virtual network environment on the computer, several simulation scenarios are conducted. The "what-if test" simulation method enables us to evaluate the performance of network and application performance. (Fig. 3).
Network auditing technology enables us to detect the mistakes of the configuration. The inventory list of all nodes in the network is able to be generated from the network simulation model. Through the simulation experiments of the network models where some nodes or links fail, the following results are obtained; the unreachable traffic flows, the utilization of all links, the routes of the traffic flows. From the above performance evaluation activities, we will find the root cause of poor performance, and make a decision of the best network architecture among several network model scenarios. (Figure 4).
In the application performance evaluation, the communication path of application messages among several servers and clients are measured by the agent software that is running in each server. This measured application path information enables us to understand the cause of the delay of the application response time. Generally, the message sequence diagram is generated in designing the application. Using such a message sequence diagram, the communication path model of application messages is generated and this model is able to be imported in the network simulation model. We can conduct the simulation and evaluate the application performance for all scenarios. (Figs. 5, 6).
We can create simulation models on the OPNET for newly developed protocols and control mechanisms. We analyze the behavior of the protocol in detail and make process models which can simulate the behavior of the protocol. We have experiences to make the simulation models for following protocols, SIP, inter AS RSVP-TE, MPLS BGP VPN, 802.11, and message queuing protocol. The developed simulation models can be added in our OPNET library. (Fig. 7).
(Note) The system names and brand names shown on this page are generally trademarks or registered trademarks of their respective developers. They are mentioned here solely for descriptive purposes, and without any intention of infringing their owners' trademark rights.
Fig. 3: Virtual environment structure of computer model
Click on image for larger version.
Fig. 4: Example of a network audit
Click on image for larger version.
Fig. 5: Application performance analysis and modeling
Click on image for larger version.
Fig. 6: Simulation of application model
Click on image for larger version.
Fig. 7: Protocol modeling and evaluation
Click on image for larger version.