Here is a peek into the evolutionary idea of service aware path computation to improve the cloud supporting wide area networks. By Dhruv Dhody
Cloud is all the buzz these days and it’s often said to be the next big thing. In fact cloud is almost here, right at your doorstep. More and more enterprises are cutting their IT budgets and exploring moving their applications to the cloud. So how does “cloud” impact our routers?
Applications which were available over the LAN would now need connectivity over a WAN, making the distance between the user and the application to drastically increase. Routers providing this WAN connectivity would be a key to deliver the cloud functionality to the enterprise.
To improve the cloud supporting WAN networks, there are some revolutionary ideas (software defined network or SDN) and then there are some evolutionary ideas (service aware path computation using PCE – Path Computation Element). Software Driven Network aims to improve network behavior by controlling the network devices using a central orchestrator; it opens up the control of data flows through the network to customizable software that is independent of the hardware that forwards the flows. OpenFlow is one such revolutionary approach to program the dummy box replacing the decades old control plane function of the router. Another evolutionary approach is to make the current control plane service aware — where the network elements (routers) are fully aware of delay (latency), jitter (latency variation) and packet loss.
There existed many mechanisms to measure these parameters after the path has been established (ex VOIP), but this is inefficient — these service parameters should be considered during the path selection process itself.
Concern for quality
Quality of Experience (QoE) is essential for all users and can only be achieved if these service parameters are met to best optimize the network. There exist mechanisms in MPLS (Multiprotocol Label Switching) based network to reserve some bandwidth with a mechanism to add more bandwidth when service disrupts. This way of providing a best path and hoping the service parameters would be okay, and taking action only when these parameters are not satisfied is unacceptable. Service parameters should guide the path selection process optimizing the network based on the QoE of user rather than some arbitrary cost metric.
Different applications have different service requirements; some may want high bandwidth, some low jitter, some minimal packet loss and its various combinations.
The network should be aware of the service parameters, and the routers should measure and make these available to the PCE. The PCE server would use these parameters in path selection process, optimizing the paths and overall network, thus guaranteeing better QoE.
Other major technology trends include big data (large complex data set used by business intelligence) and real-time data analytics. Hadoop is a framework for distributed processing of such complex unstructured data in a cluster set. Big data has unique service-parameters which should be considered in path calculation.
Thus, we can assure traffic that is delay-sensitive. For instance, in financial stock trading, data may follow a different path in the network than regular data backup. The best effort connectivity or just reserving some basic bandwidth is not efficient. It’s a must to differentiate traffic from enterprise to cloud to be well aware of the application and its inherent service requirements.
More evolved
PCE-based service aware path computation is more evolutionary as it retains much of the investment in widely deployed technologies, migrating only the key service aware path computation/selection function to a centralized PCE server. The path computation process is key in managing the network behavior by control traffic engineering policies with service –aware parameters. The PCE architecture delivers this crucial piece without modifying every network element. This is a gradual mechanism to a complete centralized application aware software driven network.
We along with other vendors and operators, have been leading this effort in the Internet Engineering Task Force (IETF) PCE Working Group (WG) with WG adoption toward RFC standardization. The work is aimed at allowing the PCE to provide path optimizing new service parameters (provide a path with minimum delay) or to set boundary conditions (provide a path with minimum jitter and max packet loss allowed 0.001%). Close co-operation with multiple WGs is planned to achieve a standard complete service aware path computation – OSPF (Open Shortest Path First) & IS-IS (Intermediate System – Intermediate System) Routing Protocol extension to carry link service aware properties to populate the TEDB (Traffic Engineering Data Base); MPLS (Multiprotocol Label Switching) for framework and use of these service parameters along with CSPF (Constrained Shortest Path First) algorithm updates for the new constraints.
Cloud Service Optimization (CSO) is an area with key requirement for service awareness. This is because cloud balancing (geographically dispersed data centers) puts pressure on the service provider network. As more and more organizations move toward the cloud, there is a need for communication and joint optimization of network (bandwidth, latency, etc.) and data center application IT (server, memory, CPU…) resources. Service aware PCE is a key component of such optimization.
Dhruv Dhody is a System Architect with Huawei India R&D.
