PDF of this tutorialBandwidth Management and Network Engineering
The foundation of bandwidth management in ATM technology is a traffic contract that is used to assign bandwidth to a specific customer requesting an IP–based service. ATM can accommodate this by allocating more or less cells (containing encapsulated IP traffic) within a given VC to the customer. Because the cells are identified as belonging to that customer, they can be monitored throughout the end-to-end connection to ensure that the customer is getting the service that it requested and to ensure that the customer's use of bandwidth does not exceed its contract and use up bandwidth that belongs to another customer on the same link. This bandwidth can be asymmetrical such that the user at one end can contract for more or less than the user at the other end. Because the service is connection oriented, ATM ensures that only the required bandwidth is reserved on each node, thus making the most efficient use of physical resources within the network and preventing the need for frequent upgrades to nodes. Additional bandwidth efficiency is created with ATM–based IP networks through the use of the ATM point-to-multipoint feature.
Within a connectionless PoS network, all bandwidth is available by all customers and all applications at all times. There are no end-to-end traffic guarantees that allow the user to be sure that the data will arrive in a timely fashion. If a node or a link becomes congested, packets are often dropped in a random fashion (depending on the features employed on the router). When this happens, the transmission control protocol (TCP) layer at the end node must ask for a retransmission of packets. When this happens, there are obviously considerable delays in receiving and reassembling the entire datastream. Network providers can manage bandwidth demands by over-provisioning their networks and ensuring that there are no bottlenecks at critical points. The general guideline is that the provider must over-provision by approximately 30 percent to prevent excessive packet loss. Point-to-multipoint connections are not possible in PoS networks as a result of the point-to-point nature of SONET connections.
Provisioning and Selling Services
Traditional Internet-access services are based on selling best-effort access for a fixed, often tiered, price. The focus for service providers who are providing this type of service is to reduce costs in order to increase or create margins. Now there is a new opportunity to sell differentiated IP services to business customers who want to outsource their wide-area-networking business. The focus of this opportunity is to provide secure, reliable, private, IP–based interconnection. For business customers, the alternative is costly private networks, so price is less of an issue: their priority is the quality of the service and their ability to prove the value received for the money spent.
Using an IP over ATM network, the service provider has an opportunity to sell differentiated IP VPN services. To start, the service provider can create a service contract that defines a CoS that is associated with a particular end-to-end data path. Because the data path is defined by an ATM VC, it is an easy matter for the service provider to then provision a VPN for an individual customer. VPNs that are provisioned this way are isolated from each other and run over independent ATM VC. Given this isolation of customer bandwidth and the deterministic nature of ATM traffic management, providers can deliver deterministic guarantees of service levels.
Such VPNs can be provisioned quickly by allocating the necessary resources on a network-management terminal, in a similar manner to the way that plain old telephone service (POTS) is provisioned through the signaling system seven (SS7) network today. Service-level monitoring can be provided by a service terminal located at the customer's site that allows the customer to monitor statistics for its specific data paths. Through similar mechanisms, it is also possible to provide service-level guarantees for Internet services.
PoS networks are well suited to selling access services to a broad range of customers who want the same level of best-effort service. By consolidating traffic onto bigger, more efficient routers, PoS networks offer the service provider the opportunity to reduce overall costs for bulk transport of IP traffic. These networks are not well suited, however, to providing differentiated VPN services, because they are not designed to set up and manage an end-to-end connection for an individual customer. There are several IP–protocol enhancements available that will improve this capability, including resource reservation protocol (RSVP), layer two tunneling protocol (L2TP), secure Internet protocol (IPSEC), differentiated services (DiffServ), and random early detection (RED). To date, the combination of these has not proven effective at providing the secure, reliable, manageable, and monitored services for which customers are looking when they outsource their wide-area network (WAN) business. For example, while IP tunneling schemes can provide isolation of customer forwarding tables, they require either the provider or end user to construct a set of point-to-point tunnels to provide connectivity. Managing this set of tunnels is equally as hard as managing the equivalent of a mesh of permanent virtual circuits (PVCs), an approach long ago rejected by all.
Network Management
The focus of network management in an IP–over–ATM network is to manage services rather than individual devices or point-to-point connections. This is accomplished by configuring policies for specific customers within the network management system and having these policies automatically propagated to all network elements that participate in providing a service to that customer. To maintain management flexibility within this environment, the network manager has the ability to preconfigure long-hold switched VCs or allow the end user to negotiate admission to the network under a specific traffic contract as part of a user-initiated switched virtual circuit (SVC). Because the management system associates a customer with an end-to-end connection and a policy, it can maintain the appropriate accounting and performance statistics that are vital for billing, general customer service, and the monitoring of IP SLAs. The focus of network management within an IP/PPP over SONET/SDH environment is to manage a collection of point-to-point links and individual routing devices. Because IP traffic is routed through the network as a connectionless service, individual datastreams are invisible to the network-management device (and the network manager), making it difficult to monitor or to offer SLAs. This system is, however, perfectly suitable for delivering network services to customers who are looking for best-effort services.
Scaling
Scaling IP over ATM networks to accommodate additional IP traffic is accomplished by growing resources within each of the following four areas as required: network management, routing management, data forwarding, and data transport.
- Network-management resources grow to accommodate an increased number of network elements that are part of a larger user base. The network-management system hides the complexity of individual links and nodes from the operator, who only needs to see this level for troubleshooting purposes. The majority of the commissioning work for individual nodes is done automatically when the node is brought on line.
- Routing management grows by adding more IP route–calculation devices (routing services control points or RSCPs) to the pool of redundant resources already within the system. As routing SCPs are independent of the forwarding elements, they can be scaled independently to provide capacity when required. They provide an effective means of aggregating routing resources to reduce the number of physical devices and hence failure points and entities that must be managed.
- IP–data forwarding scales by adding devices at the edge of the network according to the user demands within a particular geographic location. As these devices do not perform any routing calculations, they are relatively inexpensive to own and operate.
- Data-transport capacity is increased by scaling the ATM network, which uses an address scheme that allows for multiple hierarchies and a route-calculation mechanism that sets up connections through a hierarchy of groups of resources. As ATM data streams are divisible via VCs, the addition of new customers to the network does not necessarily require the addition (or the purchase) of new physical ports on the affected network nodes.
Scaling PoS router networks is accomplished either by adding more routers or increasing the capacity of core routers or both. As routing and forwarding capacity are provided on the same device, any upgrade to one implies an upgrade to the other, with the associated cost impact. Further, each additional routing entity places increased load on the routing system. This is particularly unfortunate if the addition was an unwanted side-effect of adding more forwarding capacity. Although each router and each of the links between them have traditionally been managed as separate entities, the management software from individual router vendors has improved recently to provide a more consolidated view of the network as a whole. Network design is critical as the network scales to prevent bottlenecks and to limit the amount of routing chatter associated with the networkwide propagation of routing-table updates.
Overhead Issues
IP achieves about 80 percent of the available line rate when operating over ATM, whereas it achieves 95 percent of the line rate when running over PPP/SONET/SDH. This is because, in the latter case, frames are transported directly into the SONET/SDH payload, thus eliminating the overhead required to support ATM (for example, the ATM cell header, IP over ATM encapsulation, and the partial fill resulting from the fixed-length nature of ATM cells). However, although there is this decreased efficiency with ATM, the cost expressed as a percentage is almost the same as the difference in utilization between ATM and PoS. As discussed, the benefits that come with the additional ATM overhead include the following:
- ability to sell differentiated services such as VPNs
- customer isolation without the management overhead of IP-level tunneling schemes
- ability to engineer the network and thus control the use of bandwidth specifically to ensure that it is used to derive maximum profits
- ability to provide deterministic guarantees of CoS and QoS, which in turn makes it feasible to offer SLAs the ability to manage services for individual customers, as opposed to nodes and links for a pool of customers
