The design of the scaling capabilities is introduced in this section. Although the basic functionalities are the operational basis of any BRAS system, the scaling in/out design is presented first because it is the main contribution of this thesis.

The author in citep{blendin2016position} presents an overview of the resources that are considered performance-relevant for a data path element. It comprehends the CPU Core, the last-level cache (LLC), the main memory space, the memory access bandwidth, the disk storage, the PCIe bus (direct access), the NIC (virtual switch), the NIC (direct access) and the VM monitor process. This set of physical components is considered part of the resource utilization problem whose objective is to estimate the resource costs related to the number of users/sessions associated with a data plane of the system. The result of the measured cost must lead to determine whether scaling out of the system is required and which hardware solution can be considered adequate for the system into consideration. According to the conception of the access service, it is proposed to evaluate first the operation of the system when only one customer session is handled.

In this case, a straightforward design of the system is suggested. It incorporates a single platform deployment in which all virtualized functionalities provided by the BRAS system data plane run on top of the same platform. In principle, this approach is preferred over a distributed one because it leads to the lowest-cost design able to run on commodity software. To provide the design of the ability to scale out a data path element, a second session is considered. Hereafter a stepwise design procedure is introduced. First of all, we assume that the two sessions make use of the same number of resources of the system. It means that both of them are received through the same input interface, utilize the same capacity of memory and CPU to be handled, an equal bandwidth is expected by the two sessions, and use exactly the same services or more specifically, the chain of virtualized functions employed is alike in number and in length.

Likewise, both sessions are assumed to be already established and be operative in the first data path platforms whereby it is granted that there is no needed communication with the control plane in order to offer certain service to the session. A general view of the established sessions is observed in figure ef{fig:InitialSessions_scaling-out}. It illustrates the services used by the two sessions in which the blue modules make reference to the previously mentioned physical resources and the white ones to the virtualized functions. It is relevant to notice that between the forwarding unit and each of the VNFs exist interfaces that connect them. They are referred as the network As stated before both sessions are already established and operating over the data path element 1. The “scaling-out” process must consider different situations and recognize the procedure to be employed for each of them. From the recognition of the managed data path elements to the migration of sessions.

As considered previously there are four scenarios that are evaluated in our design. They are separated by the number of input and output interfaces that the system produces through the implementation, and for each, a different setup for the interfaces is deemed and an adjustment in the “scaling-out” procedure is proposed.The general “scale out” methodology for a data path element of the BRAS system is introduced. The author in citep{bogineni2016sdn}, which leans its work on the NFV reference architecture framework defined by the ETSI, presents a couple of use cases as part of a VNF management example based on an architectural framework for IP multimedia services. One of those use-cases details a methodology for a VNF scale out. It depicts a message sequence chart for the automated scale out of a VNF and the interworking of the different components of the NFV architecture. We decide to use this representation as a guideline for our design because it involves many of the modules from the SDN-NFV reference architecture, which is backed by numerous top telecommunications organizations as well as the similarity and suitability to our research.

Therefore the following “scale-out” design for the BRAS system adapts general liniments of the previous research to establish the operation method from our work. Table ef{tab:Messages_Init_protocol} sums up the protocol messages used for initialization, resource negotiation, and VNF confirmation as well as the error and time-outs messages which are next introduced and explained. First, the control plane of the system must identify the additional data path element. For this, it initiates an inclusion protocol that basically reduces to a request-acknowledge communication whose response is generated by a virtualized function which is in charge of the scaling out activities. It must provide status information and resource availability to be able of joining as an additional platform of the data plane. This protocol is proposed as a modification of the normal recognition protocol, which must be expanded in order to carry a flag that enables the control plane to ask for the intention of including the new data path element to its data plane.

 Once the join packet is received the new platform must send an acknowledge join intention packet which confirms that the platform recognizes the possibility to operate as an additional platform of the BRAS data plane. As a further step, and as part of the advanced module of the BRAS control plane, the information about the resource availability for the new element must be evaluated. Therefore the control plane sends an information packet request directed to the identification address of the element. As a response, the element must enclose the resource availability in an information packet reply, which is afterward, sent back to the controller. The information must contain whether the system is virtual or physical, total and available CPU, total and unused memory, number of network interfaces with the respective interface capacity, system implementation type among others. After the reception of the reply message, the control plane must evaluate whether each of the characteristics of the platform fulfill the requirements, which have been set manually or automatically, and if so, it sends a resource-acknowledge request packet and waits until receiving a resource-acknowledge reply from the new element in order to confirm that the negotiation for the resources suits the configurations requirements.The next step is to confirm if the necessary virtual network functions are deployed in the new data path element.

With that in mind, a special hash identification scheme is suggested so that the control plane is not only able to check the virtualized network functions running on the platform but also capture their versions in order to evaluate the compatibility with the currently used in the BRAS system data plane. Thus, the module for advanced services of the control plane must send a VNF information request which asks for the type, function, and hash identification of the virtualized network functions deployed in the system. The protocol follows the same behavior of the resource negotiation part. Once the request packet is received, the new data path element must append the required information in the payload of a VNF information reply packet which is subsequently sent back to the controller. Afterwards, a pair of request/reply VNF acknowledgment messages is exchanged between them as long as the negotiation has been successful. The visualization of this initialization protocol can be observed in figure ef{fig:Scale-out_recognition}, note that in this case, all requests have been successfully replied.Taking into account the time constraints, it is relevant to notice that this initialization and set up process is not needed to be executed in a short period of time due to the scaling out process is assumed to be carried out with enough planning and configuration time frame.

It means that a swift resource and VNF negotiation is desirable rather than mandatory. Time-out responses for the reply negotiation packets as well as error response packets must be considered. They enable the ability to recognize whether the system resources are insufficient or a communication problem occurs between the control plane and the data path element. 


I'm Erica!

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

Check it out