New and highly demanding applications have emerged in the last decade creating the challenge of maintaining uninterrupted service availability and dependability even under network failures. This challenge has naturally sparked interest in Internet traffic engineering which deals with the issue of automatic multipath traffic control.
Current methods cannot efficiently utilize network resources and offer limited control capabilities for traffic engineering. Furthermore, when a network failure occurs, it takes a significant amount time until the network stabilizes again after such failures causing excessive delays, packet drops and service interruptions. They tend to have network-wide effect and can result in undesirable and unanticipated traffic shifts. In addition, current methods are reactive in nature and they usually require resources to be reserved that remain idle except during a network failure.
Multicast traffic over the Internet is growing steadily with an increasing number of demanding applications including Internet broadcasting, video conferencing, data stream applications, web-content distributions, and exchange of large data sets by geographically distributed scientists and researchers working in collaboration. Many of these applications require certain rate guarantees and providing such guarantees demands that the network be utilized more efficiently than with current approaches to satisfy the rate requirements.
Today's data centers are typically designed as an n-tier network with a number of peering points ranging from six to sixteen. Further, the network topology is highly connected as a partially meshed architecture so that there is enough redundancy to ensure availability in case of network failures. In such a network topology with a high level of connectivity, it is inevitable that while some of the network resources are highly utilized, others stay mostly idle due to the nature of traditional IP routing protocols. Consequently, the uncontrolled traffic distribution can result in significant network congestion while there are available resources in the network. Hence, data centers can significantly benefit from using active traffic flow control solutions not only to utilize network resources efficiently but also to provide traffic protection against network failures.
CertusNet's technology effectively targets all the limitations mentioned above through an effective multipath traffic control approach by considering the traffic control problem in IP networks. Specifically, we solve the multi-path optimal traffic control problem for both unicast and multicast sessions. The algorithm is measurement-based as opposed to existing solutions. With the use of the (Simultaneous Perturbation Stochastic Approximation) SPSA techniques, our solution is able to converge to the optimal network operating point very fast. Also, with making use of a constant step size policy and weak convergence theory, our technology is able to track and adapt to the changes in the network state continuously and in a fast manner.
With the use of a modified SPSA technology (Simultaneous Perturbation Stochastic Approximation), we provide an optimal, yet practical, solution that can effectively minimize network congestion and maximize the usage of network resources by only exploiting the locally collected noisy network measurements such as link utilization levels, packet drops or delay.
Due to its optimality, CertusNet's algorithm does not cause any instability or undesirable traffic shifts. An important advantage of CertusNet's solution over existing technologies is that CertusNet's generalized algorithm is able to guide and control traffic from sources with multicast nature as well as unicast.
Further, we do not require any changes in the existing network architecture, hardware or routing protocols.
