Engineering a multiservice IP backbone to support tight SLAs
Introduction
When using public transport, the traveller may benefit from contractual commitments from the transport service provider, for example that 95% of journeys will arrive within 5 min of the scheduled time. The commitments may include other parameters or metrics such as number of stops en route and any meals included. The more competitive the market for the particular service, the more comprehensive and the tighter the commitments or service level agreements that are offered. In the same way, the increase of competition between IP service providers (SPs) together with the heightened importance of IP to business operations has led to an increased demand and consequent supply of IP services with tighter service level agreements (SLAs) for IP performance.
The IP technical community has developed a set of technologies that enable IP networks to be engineered to support tight SLA commitments:
- •
Differentiated services (Diffserv). The Diffserv architecture allows differentiated delay, jitter and loss commitments to be supported on the same IP backbone for different types or classes of service.
- •
Faster IGP convergence. New developments in Interior Gateway routing protocols (IGPs) allow for faster convergence upon link or node failure, hence enabling higher service availability to be offered.
- •
MPLS traffic engineering. MPLS traffic engineering (Diffserv-aware or not) introduces constraint-based routing and admission control to IP backbones. This allows optimum use to be made of the installed backbone bandwidth capacity, or conversely allows the same level of service to be offered for less capacity. It can also be used to ensure that the amount of low-jitter traffic per link does not exceed a specified maximum.
- •
MPLS traffic engineering fast reroute. MPLS traffic engineering fast reroute is an IP protection technique that enables connectivity to be restored around link and node failures in a few tens of milliseconds.
For an SP IP service, the SLA commitments are based on delay, jitter, packet loss rate, throughput and availability. This paper focuses on defining these SLA parameters, and describing why and how the above technologies should be used in order for these SLA parameters to be tightened. Consideration is also given to how networks using these technologies should be operated. There are other factors that will affect a service provider's ability to offer tight SLAs in addition to the technologies described, but which are not covered in this paper; these include: network security and BGP convergence.
In focussing on service provider IP backbone networks, it is noted that the mechanisms employed at the edge of the network to deliver tight SLAs may be different from those used in the core. In the backbone, where traffic is aggregated, SLA requirements for a traffic class can be translated into the appropriate bandwidth requirements, and the problem of SLA assurance can effectively be reduced to that of bandwidth provisioning. At the network edge, other considerations, such as serialisation delay, become significant. The mechanisms employed at the edge of the network are not considered further in this paper.
Section snippets
Customer requirements––what an SLA should commit to
The following metrics are highlighted as the most important for specifying the quality of an IP backbone service [15], For each metric, typical values are given for different types or classes of service.
IP backbone diffserv overview
The Diffserv architecture [3] is the preferred technology for large-scale IP QoS deployments today, such as service provider backbone networks. Diffserv achieves scalability through performing complex QoS functions such as classification, marking, and conditioning operations at the edges of the network. Traffic is classified and then marked using the Diffserv code point (DSCP) [21] into a limited number of traffic aggregates or classes. Within the core of the network, scheduling and queuing
Fast IGP convergence
Link or node failures in an IP backbone cause packet losses until the network has reconverged around the failed link or node. These packet losses directly impact the availability that can be offered for SLAs across all classes. To assess the significance of this one can compute the amount of downtime corresponding to different availability commitments: the often-quoted 99.999% or “five-nines”-target figure for network availability potentially equates to less than 1 s of downtime per day.
Another
MPLS traffic engineering
In conventional service provider IP networks routing protocols such as OSPF and IS–IS forward IP packets on the shortest cost path to the destination IP address of each IP packet. The computation of the shortest cost path is based upon a simple additive metric, where each link has an applied metric, and the cost for a path is the sum of the link metrics in the path. Availability of network resources, such as bandwidth, is not taken into account and, consequently, traffic can aggregate on the
MPLS traffic engineering fast reroute
In Section 4, we highlighted that link or node failures in an IP backbone can significantly impact the availability that can be offered for SLAs across all classes. Whilst sub-second convergence for IP routing protocols is a realistic prospect, it is expected that IGP convergence will not be able to match the capabilities of SDH/SONET networks, which use the capabilities of Multiplexer Section Protection (MSP) and Automatic Protection Switching (APS) respectively to recover around failures in
Conclusion
In the context of ever more competitive service provider offerings and ever more demanding requirements from their customers, this paper has analysed which SLA parameters are significant for IP service performance (delay, jitter, loss bandwidth/throughput, per-flow sequence preservation and availability) and has listed the targets usually set for these parameters for the typical backbone aggregated classes of service.
We have reviewed the technologies that can be used to tighten these SLAs and
Acknowledgements
The authors would like to thank Olivier Bonaventure, Thomas Telkamp and Bruce Thompson for their valuable feedback and constructive comments.
Clarence Filsfils has an Engineering Degree in Computer Sciences from the Institute Montefiore of the University of Liege, Belgium and a Business Degree from the Solvay Business School, Brussels Belgium. He joined Cisco in 1996 and as a Distinguished Engineer; he focuses on IP Core Routing and Capacity Management (IP QoS/Traffic Engineering) designs.
References (28)
- D.O. Awduche et al., RSVP-TE: extensions to RSVP for LSP tunnels, RFC3209, December...
- D. Awduche et al., Requirements for traffic engineering over MPLS, RFC 2702, September...
- S. Blake et al., RFC2475, An Architecture for Differentiated Service, December...
- T. Bonald, A. Proutiere, J. Roberts, Statistical guarantees for streaming flows using expedited forwarding, INFOCOM...
- R. Braden et al., RFC2205, Resource Reservation Protocol (RSVP) Version 1 Functional Specification, September...
- S. Casner, C. Alaettinoglu, C.-C. Kuan, A fine-grained view of high-performance networking, Packet Design, NANOG 22,...
- A. Charny, J.-Y. Le Boudec, Delay bounds in a network with aggregate scheduling, in: First International Workshop on...
- B. Davie (Ed.) et al., An expedited forwarding PHB, Internet draft, April 2001 Charny et al., An expedited forwarding...
- N.G. Duffield et al., A flexible model for resource management in virtual private networks,...
- Le Faucheur et al., Requirements for support of Diff-Serv-aware MPLS Traffic Engineering, Internet...
Random early detection gateways for congestion avoidance
IEEE/ACM Transactions on Networking
Cited by (14)
A priority-based queuing model approach using destination parameters for real-time applications on IPv6 networks
2020, Turkish Journal of Electrical Engineering and Computer SciencesSDN-based QoS aware network service provisioning
2015, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)Optimization of network redundant technologies collaboration in IMS carrier topology
2011, Lecture Notes of the Institute for Computer Sciences, Social-Informatics and Telecommunications Engineering, LNICSTAn auditing system for multi-domain IP carrying Service Level Agreements
2011, Proceedings of the 12th IFIP/IEEE International Symposium on Integrated Network Management, IM 2011Interdomain quality of service routing: Setting the grounds for the way ahead
2008, Annales des Telecommunications/Annals of Telecommunications
Clarence Filsfils has an Engineering Degree in Computer Sciences from the Institute Montefiore of the University of Liege, Belgium and a Business Degree from the Solvay Business School, Brussels Belgium. He joined Cisco in 1996 and as a Distinguished Engineer; he focuses on IP Core Routing and Capacity Management (IP QoS/Traffic Engineering) designs.
John Evans received a B.Eng. (Hons) degree in Electronic Engineering from the University of Manchester Institute of Science and Technology (UMIST), UK in 1991 and an M.Sc. in Communications Engineering from UMIST in 1996. He joined Cisco Systems in 1998 and as a Consulting Engineer; he focuses on IP network design and development with special interests in core routing and traffic management, including IP quality of service and traffic engineering. Prior to joining Cisco, he worked on the design and development of large-scale networks for the financial community and for the military.