250 likes | 371 Vues
An integrated switching strategy for ABR traffic control in ATM networks. Andreas Pitsillides, Petros Ioannou University of Cyprus University of Cyprus&
E N D
An integrated switching strategy for ABR traffic control in ATM networks Andreas Pitsillides, Petros Ioannou University of Cyprus University of Cyprus& University of Southern California
ATM based Multiservice and Multimedia networks • support various classes of multimedia traffic • different bit rates (continuous and variable) • voice, 64 Kbits/sec constant bit rate • data, on-off variable bit rate • MPEG video, say 2-15 Mbit/sec variable bit rate • different quality of service requirements, e.g. • data: no loss, but tolerance to delay • voice and video: cannot accept large delay, but tolerant to loss • make network control difficult
traffic control and resource management • crucial to guarantee desired grade of service. • Whilst recognised early that • for effective control essential to integrate several control mechanisms • not many works published. • focus on combined dynamic control of connection admission, flow, and bandwidth • consider ABR traffic in presence of competing real-time (guaranteed) traffic sharing the common switches along connection path.
control of connection admission, flow rate, and bandwidth • most studies • assume steady state conditions in problem formulation and/or solution • Adaptivity provided by assuming quasi-static loading conditions and resolve static steady state problem • each function looked in isolation and their interactions not analysed/considered • dynamic strategies • can improve network utilisation • little work reported
dynamic model for ABR traffic ATM switching node. ABR traffic path spanning M ATM switching nodes. guaranteed traffic appears across every ATM switch, requires some access to the shared resources of each link
"fluid flow" state model in differential equation form flexible in establishing performance measures for use in optimal control objective function lower order of complexity easy provision of controls in distributed form difficult to find a model which captures “essential” dynamic behaviour The full state space models prohibitively complex Chapman-Kolmogorov equaton for time-dependent state probability distribution notoriously difficult to solve analytically due to time varying coefficients. capture detailed behaviour dynamic model probabilistic model
dynamic fluid flow equation • From flow conservation principle: • consider a single queue and assume no losses • rate of change of average number of cells • rate of cell arrivals • rate of cell departures
extend to path (FIFO service discipline and shared buffer)
control philosophy • Integrate Control of: • Connection admission, ABR flow rate, and bandwidth allocation.
Derivation of Control strategy(Single node, finite buffer, finite capacity) zero guaranteed traffic x(to)=xo Select: Let: • xmax and a design variables • select xmax£ xbuffer capacity and xmax ³ xo
Derivation of Control strategy (cont.) • With this choice of C(t) and l(t) • since 0£x(t0)£ xmax • 0£x(t)£ xmax for all t • i.e. always upper bounded by xmax (and xbuffer capacity)
Derivation of Control strategy (cont.) But C(t) also needs to satisfy constraint • consider design parameter Cmax £ Cserver capacity • choose together with a so that C(t) satisfies constraint. Select a as: satisfies C(t) constraint ensures long term inflow demand is met • Selection of design variablesa, xmax, and Cmax • can be used to prescribe QoS delivered to user
Derivation of Control strategy (cont.) extension to non zero guaranteed traffic • Its satisfaction depends on xmax, and Cmax • If • control strategy cannot be effective
Extension of control policy to M ATM switch case: switching strategy • At each ATM switch i (i=1,..M) • fully decentralised and independent of any propagation delay
remarks • switching of control strategy (only required when Cmax is less than Cserver capacity) • xi(t)>ci to utilise any reserved capacity for guaranteed traffic • xi(t)ci obtained from earlier analysis. • ci chosen to guarantee bounds on xixbuffer capacity and CiCserver capacity are not exceeded. Switching only initiated when x(t)xmax, • hysteresis included for computational purposes (smooths trajectory)
Connection Admission Control (CAC) • CAC for ABR traffic • admission only limited by local buffer space (same consideration as today’s LANs). • average arrival rate can be declared to ensure long term demand met. • CAC for Guaranteed traffic keep less than • Guaranteed traffic rate must remain less than a portion of Cmax. given by ratio (xbuffer capacity-xmax) to xbuffer capacity+1. • to increase amount of Guaranteed traffic • increase buffer space (if the desired QoS allows), and/or • reduce buffer space allocated to ABR traffic (hence ABR traffic)
Simulations with ABR traffic over 3 ATM switches • ABR traffic over three ATM switches, propagation delays as shown • Cmax=10 at each ATM switch and Cserver capacity=12 cells/msec, i.e. 2 cells/msec can be reserved for guaranteed traffic. • xmax=100 and xbuffer capacity=140,i.e. 40 cell places can be reserved for guaranteed traffic • admission for ABR traffic set at lABRaxmax, a is set at the upper limit 10 msec(2000km) 2 msec (400km) 15 msec(3000km) =0 =0
4. Simulation: • a) No switching • b) With switching • c) switching strategy is inhibited
scenario with no switching between the 2 strategies • no switching since xi(t) never exceeds switching boundary
scenario with switching between the 2 strategies • capacity at switch 2 exceeded maximum, but not server capacity
switching between the 2 strategies inhibited • cannot prevent losses
Conclusions • integrated dynamic nonlinear control strategy for connection admission control, flow control and bandwidth allocation control under nonstationary network conditions. • extends earlier control solution for the multiple ATM switch path. • control behaviour evaluated using analysis, and demonstrated using simulation. • show scheme achieves effective server and buffer utilisation (as predicted by analysis) and zero cell loss
Conclusions (cont.) • proposed solution simple to implement feedback relationship. • offers proven (guaranteed) effective control performance, through selected design variables a, xmax, and Cmax can influence delivered QoS. • Extensions • investigation how above solution can be implemented within framework proposed by the ATM Forum Traffic Management Specification.