Signaled Provisioning of the IP Network Resources Between the Media Gateways in Mobile Networks Leena Siivola 10.12.2004 Rev PA1 10.12.2004 1 Problem Description • For circuit switched (CS) traffic the delay and the jitter requirements are strict. That is why the amount of voice calls must be controlled not only from radio networks (RN) side but also from IP multiservice backbone’s point of view. • The backbone edge nodes, i.e. the Media Gateway, must have ways to control the amount of traffic injected to the network – This must make it possible to give some QoS guarantees for the voice calls – The network resources will be used more efficiently Rev PA1 10.12.2004 2 Objectives and Scope • The objective of this Thesis is to – describe the current Call Admission Control (CAC) mechanisms in the 3G IP multiservice backbone – to evaluate the suitability of the NSIS signaling protocol framework for the CAC solution. Rev PA1 10.12.2004 3 The functional architecture of the 3G network Rev PA1 10.12.2004 4 The Call Admission Control Mechanism Rev PA1 10.12.2004 5 Provisioning Methods in the IP Multiservice Backbone . . MBAC = Measurement Based Admission Control MPLS = Multiprotocol Label Switching Rev PA1 10.12.2004 6 Static Provisioning Methods in the Media Gateway Rev PA1 10.12.2004 7 Signaled Provisioning Signaled provisioning is a tempting approach for CAC because it can give ’hard’ QoS guarantees for traffic flows and it can increase the network utilization. Many QoS signaling protocols exist: • Tenet & ST-II • RSVP with its extensions • YESSIR (Yet another Sender Session Internet Reservations) • Boomerang RSVP has been the most famous one • Has said to bee too complex and suffering scalability problems -> also other simulation results exist! The work with the NSIS signaling protocol framework was started, because there was a need for a more lightweight signaling protocol. Rev PA1 10.12.2004 8 The NSIS Signaling Framework NSLP = NSIS Signaling Application Level NTLP = NSIS Transport Level Rev PA1 10.12.2004 9 The NSIS Signaling for Quality of Service (QoS) The NSIS QoS signaling framework is based on a two layered architecture: • NTLP (NSIS Transport Layer Protocol) • NSLP (NSIS Signaling Layer Protocol) • QoS Model that is being signalled (e.g. Intserv or RMD) • NSIS without QoS Model is only a framework with many optional features. Rev PA1 10.12.2004 10 Comparison Between the NSIS QoS Signaling and RSVP • • • • NSIS can be both sender- and receiver-oriented NSIS does not support multicast Mobility support Bi-directional reservation possible Rev PA1 10.12.2004 11 NSIS(RMD) Architecture It is not possible to evaluate the NSIS signaling without taking the QoS model into account. The NSIS framework consists of several optional features that can be taken into use. • Resource Management in Diffserv (RMD) implemented with NSIS Rev PA1 10.12.2004 12 Successful Reservation Initiator Edge Interior Interior Edge Receiver Resv(QSpec) Resv(E2E ignore, QSpec) Resv(LQSpec) Resv(LQSpec) Resv(LQSpec) Resv(QSpec) Response Response Response Rev PA1 10.12.2004 13 One Possible Implementation of NSIS to the 3G Rev PA1 10.12.2004 14 Evaluation + NSIS framework is flexible and modular -> it can be used in different ways + There are several optional features that can be taken into use - The resulting QoS protocol is even more complex than RSVP -> what do we gain with the abstraction level? Rev PA1 10.12.2004 15 Evaluation: The NSIS(RMD) Implementation as an Example • Evaluation criteriors: – Per-hop Performance Metrics • Signaling message processing delay – Per-Reservation Performance Metrics • Signaling Bandwidth Overhead • Abortive Provisioning • Blocking Probability • Reservation Setup Time – Applicability of the NSIS(RMD) Signaling to the IP Multiservice Backbone SCALABILITY AND ROBUSTNESS Rev PA1 10.12.2004 16 Per-hop Performance Metrics: Signaling message processing delay • • • • ts = signaling message processing delay tS0 = the base parameter fR = a component dependent of the session load (LR) fT = a component dependent of the session (LR) and the signaling load (LT) Signaling message processing delay In the edge routers: proportional to the number of sessions In the core routers: a constant Rev PA1 10.12.2004 17 Per-Reservation Performance Metrics: Signaling Bandwidth Overhead Rev PA1 10.12.2004 18 Per-Reservation Performance Metrics: Abortive Provisioning Rev PA1 10.12.2004 19 Per-Reservation Performance Metrics: Blocking Probability Rev PA1 10.12.2004 20 Per-Reservation Performance Metrics: Reservation Setup Time Rev PA1 10.12.2004 21 Conclusions •The Intserv type (RSVP-like) per-flow end-to-end signaling brings nothing new when comparing to RSVP • The message processing times have been estimated to be approximately same (1 ms) • In the IP multiservice backbones some Intserv over DiffServ approach, such as RMD, could be the solution • The message processing time in the core routers is approximately 5 microsec. • The system bottleneck is the signaling load on the edge routers • There’s only approximately 0,9 msec time to process one reservation message in the edge router • The link utilization is the same than with per-flow reservations • The response time is lower because of the sender-oriented approach Rev PA1 10.12.2004 22 Conclusions (continued) • NSIS in itself has failed to meet its design criteria: • • It is not simple and ligthweight -> It is too modular There is a serious risk that NSIS will become only one signaling protocol amoung others • Too much politics involved in the protocol design work • The router vendors are not actively participating the work -> the possibility to implement NSIS in networks is dependent of the router implementation Rev PA1 10.12.2004 23 Future research • Router vendors’ interests • NSIS(RMD) / RSVP(RMD) with MPLS-tunnels • DCCP -> the adjustment of voice codecs with network congestion, ECN marking Rev PA1 10.12.2004 24 THANK YOU! • Any questions? Rev PA1 10.12.2004 25 ADDITIONAL INFORMATION Rev PA1 10.12.2004 26 Dynamic Provisioning Methods in the Media Gateway • Measurement Based Admission Control (MBAC) + CAC is fast + no extra signaling load + implementation costs low - cannot guarantee anything - the measurement result arrives always too late • Probing + no actual traffic will be lost - additional traffic -> the probe packets can overload the network - Setup delay - the routers do not support ? • Bandwidth Broker (BB) + high utilization - complex new node in the network Rev PA1 10.12.2004 27 RSVP vs. RMD Performance RSVP NSIS (RMD) Response time (bi-directional) 1 – 1.5 RTT 0.5 – 1 RTT Processing time < 1 ms Edge: < 1ms Int.: < 5 s Link utilization ~100% ~100% Scalability limited yes Cost High processing capacity is required in each nodes Edge nodes: same as for RSVP Int. nodes: simple functionality Source: A. Bader et al.:Presentation in the 11th International Telecommunications Network Strategy and Planning Symposium (Networks2004) Rev PA1 10.12.2004 28