Service level agreements for DiffServ-based services' provisioning

Abstract

The evolution of mechanisms for providing Quality-of-Service (QoS) over the contemporary network infrastructures has introduced the need for regulation and management of the emerging QoS services with the use of Service Level Agreements (SLAs). SLAs define the qualitative and quantitative characteristics of the services provided from a network provider to peering networks or customers. In this work, we define a template for the SLA structure to support the provision of a QoS service between two peering domains and then we proceed with the definition of an end-to-end SLA across consecutive domains, based on the bilateral ones. We also propose a model for the service provisioning procedures.

Introduction

A Service Level Agreement (SLA) is an explicit statement of the expectations and obligations that exist in a business relationship between two entities: a service provider and a customer (Rajan et al., 2000). Bilateral SLAs can also be defined among organizations that have a symbiotic relationship, with each being a customer of the other's services. The SLA provides a means of defining the service. It specifies what the customer wants and what the supplier is committing to provide. It defines the standards for the quality of the service provided, setting performance objectives that the supplier must achieve. It also defines the procedure and the reports that must be provided to track and ensure compliance with the SLA. In the field of telecommunications networking, SLAs play a significant role, reinforced by the latest advances in differentiated services' provisioning.

The availability of high-speed transmission media and networking equipment, as well as the evolution of quality-demanding applications has focused research interest on the provision of Quality-of-Service (QoS) in addition to the traditional best-effort model of the Internet. A number of alternatives for service differentiation and QoS provision have been proposed and standardized, but in the case of IP-based backbone networks the Differentiated Services (DiffServ) architecture has prevailed, due to its scalability and deployment feasibility. The provisioning of IP services according to the DiffServ framework has introduced complexity in the corresponding business model and has raised the requirements for controlled resource allocation and management, definition, monitoring and verification of the quality provided. At this point, the appropriate definition of SLAs between customers and service providers is envisaged to provide the controlled environment required. In this framework, SLAs will act as mediators for mutual service provisioning between peering domains.

The DiffServ framework stands out for attempting to provide service differentiation to traffic in a scalable manner, by suggesting the aggregation of individual application flows with similar quality needs. It introduces the definition of different service classes to which such aggregates are appointed and the implementation of mechanisms for differential treatment by network elements (Per-Hop-Behaviour, PHB) of the packets belonging to each service class. A PHB is thus describing the treatment of aggregated traffic in a manner that ensures the quality guarantees provided by the corresponding service class.

Although, DiffServ has been initially confronted with a positive attitude, due to its scalability, the DiffServ framework mechanisms have proved difficult to deploy and monitor in a large scale in production networks. Based on the DiffServ framework, a number of service models constructed by a combination of DiffServ mechanisms have been proposed and experimentally evaluated up to our days. However, real-world implementations of DiffServ-based services in production networks have not successfully operated in large-scale yet. Missing DiffServ functionality from IP routers, translation to last-mile (end user access links) QoS, considerable operational and economical paradigm shifts required from operators, lack of a flexible business model, inadequacy of service verification infrastructure, inadequate standardization and architectural gaps are some of the major deployment problems analysed in (Paxson et al., 1998).

We believe that the DiffServ framework and DiffServ-based services do have a significant potential in upgrading the best-effort service model in today's Internet. However, due to its probabilistic rather than deterministic differentiation mechanisms, the provisioning model of DiffServ has to be thoroughly specified and standardized as soon as possible. Among the deployment problems already mentioned, a flexible business model for intra-domain development and peering agreements according to the existent agreements are considered crucial for the successful deployment of DiffServ-based services in production networks. We believe that SLAs are of crucial importance on this basis.

Today, the installation of SLAs between customers and providers is a rather static and labor-intensive task. The procedures involved in this process are proprietary to the provider and, in many cases; these procedures are invoked on a low frequency basis (e.g. when updating a Virtual Private Network (VPN) topology). By its proprietary nature, such a process does not allow for an open-service architecture, such as that of DiffServ-based services, to be built upon interconnected IP networks. It should be understood that standardization of the technical parts of the basic process may allow for a highly developed level of automation and dynamic negotiation of Service Level Specifications (SLSs) between peering providers or customers and providers. This automation may prove helpful in providing customers (as well as providers) with the technical means for the dynamic provisioning of QoS guaranteed transport services. SLAs and SLSs are essential for delivering the obtained QoS from one end-user to another across multiple domains.

So far there have been several efforts towards the standardization of definition of SLAs and their instantiation in QoS-enabled networks (Bouras et al., 2002, Fankhauser and Plattner, 1999, Neilson et al., 1999, Nichols et al., 1999). In production networks, service providers use their own customized SLAs for qualitative IP services provisioning. Moreover, the provisioning model usually concerns a backbone network service provider and its directly attached customers. The case of multi-domain QoS provisioning in which quality parameters of the network have to be mapped along neighbouring domains and need to be consistent along the end-to-end path is rarely addressed. Furthermore, processes for the establishment of end-to-end SLAs in such cases are not mature enough yet. Our proposal attempts to address these issues.

In this work, we are initially presenting the basic principles for the deployment of DiffServ-based service contracts (SLAs) in a bilateral fashion (between peering domains). For this purpose, we use the case of a service based on the Expedited Forwarding (EF) DiffServ PHB (Salsano, 2000), for serving high-priority and quality demanding traffic. Furthermore, we propose a methodology for establishing an end-to-end SLA based on the bilateral SLAs, using as our reference architecture that of the GEANT core pan-European research network, interconnecting the European National Research and Education Networks (NRENs) and through them campuses and institutions, user groups and end users all over Europe. Apart from the end-to-end SLA establishment, we propose a provisioning model for the set-up and coordination of SLA-based service deployment and operation.