Site Loader
Rock Street, San Francisco

Based on the requirements highlighted in the previous chapter, the design of a virtual BRAS as a network function that contains the ability of scaling out across heterogeneous platforms is now presented. We establish the requirements and assumptions used for the design process. The system has to cope with the fundamental operational characteristics orientated to the access services creation in a virtualization environment. The fundamental design considerations are described in the following sections. At first, the fundamental goals expected from the system are presented, followed by a general overview of the architecture and later on, the design for a virtual BRAS network function which adapt scaling out capabilities is presented. Finally, the extit{two-step optimization model} is introduced.In general, the fundamental goal for a virtual BRAS network function relies on provide the design of a virtualized system that enables scaling in/out capabilities over heterogeneous platforms. In general, the idea is to have a multi-platform design (based on different processing technologies such as CPU, NPU or ASIC). Through this approach, we introduce the design of a software-based virtual BRAS network function which enables the ability for scaling-in and -out the mentioned platforms. As an important requirement and in alignment with the described scenario, the system has to provide creation of access services adapted to the traditional architecture of an ISP. It implies that the system has to be seen as an appended box in the network path of the current ISP infrastructure. It acts as an element that disaggregates the centralized classical BRAS system towards the access network. Each single disaggregated entity supplies the creation of access services without relying on the classical central system. This can be seen in figure
ef{fig:BRAS_disaggregation}.The system design has to cope with the session creation and the provisioning of resources required by the CPEs. We assume that the session initialization is always performed from the client side. At the reception of a client packet, the system must process it and identify whether it is a new customer request that ask for service configuration to join the network or is an already authorized customer and hence, proceed with a corresponding action that for example may be steering the packet towards the internet interface. When a new client request is received, the customer premises equipment must be authenticated by the system in order to be identified and therefore, proceed with the resource allocation including for instance a bandwidth profile and a routable layer 3 address among others. At this point, the customer must be recognized by the network although no additional services are delivered yet, e.g. provider-owned services. To that end, the accounting information of each user must be logged in a computer-based information system which allows the provider to acquire information about the customer, the used resources and the times when that happened. This information has different objectives such as the user control and surveillance of the utilized resources as well as its use for billing purposes. Time for the start and end of the connection, user identity and given profile is the minimal required logged information. The system must provide both access services and core services for the authorized users. It implies that the traffic is now able to access provider-owned network services as well as external system services e.g. internet access.   The traffic of the users is identified in general by the layer 3 resource provided at the authorization stage. The system must provide user identification and access control resulting in the allowance or rejection of the services requested. Furthermore, the virtual BRAS system has to provide layer 2 and 3 packet processing and forwarding. Other specific access services such as multicast support, quality of service and security are desired requirements for the system. However, they are not the main focus of this project and therefore will not be considered.The deployed system runs over commodity platforms and must be able to handle between 100 and 16500 user sessions over heterogeneous platforms. These numbers are based on the expected amount of users served at the main distribution frame, for example in the city Darmstadt. Nonetheless, the BRAS system will face the situation where an increment or a reduction of the number of users/sessions occurs and therefore it has to adapt the available resources to the number of handling sessions and in case of scarcity, a designed resource increment must be executed. Therefore, the services provided to the customers must consider the concept of elasticity, which reduces to build a procedure that works in an automated fashion and which bases in scaling the system resources to cope with the service needs. Its design must intend to give a better resource utilization as well as keeping or improving the service performance, which can be obtained from the NFV approach at a lower cost. Thus, a scale-in/out methodology is proposed over a scale-up/down, since a scale-out strategy offers better cost-performance, although at an increase in management complexity. Furthermore, scale-in/out presents also good results even on a scale-up platform citep{michael2007scale}. Additionally, we propose a extit{two-step optimization model} which is divided in two steps. The first, enable user sessions to be grouped in clusters which are afterward allocated through the available nodes. The second is inspired in a failover mechanism in which, failing nodes are modeled and clusters are re-distributed from where they were in the first place across the remaining operative nodes. For both optimization models, the used internal resources of the nodes i.e. processing units are minimized. It is proposed this way in order to use the minimum internal resources and leave room for adjacent network functions that might be collocated in the same nodes.    By using this approach, not only providers are able to move their BRAS services towards the end users but also an optimal usage of the resources deployed can be carried out. Thereby, current high bandwidth low-delay IP-based services could be placed in the aggregation/access network improving, in principle, the performance of the ISP network infrastructure.Next, the system architecture is introduced followed by the specific design of the scaling out mechanism and the subsequent functionalities.   

Post Author: admin


I'm Eric!

Would you like to get a custom essay? How about receiving a customized one?

Check it out