Packet Loss and Delay obined Optiiation for Satellite hannel Bandwth Allocation ontrols Igor Bisio Meber, IEEE, Mario Marchese Senior Meber, IEEE DIST - Departent of ounication, oputer and Syste Science University of Genoa, Via Opera Pia 13, 16145, Genoa, Italy phone: +39-1-35386, fax: +39-1-353154 e-ails: { igor.bisio, ario.archese }@unige.it Abstract The paper studies the bandwth allocation process over satellite counication systes as a Multi Objective Prograing (MOP) proble and evaluates an allocation ethod called obined Utopia Miniu Distance (obined UMD). The entities of the syste are earth stations and, for each of the, a set of perforance etrics (represented by specific analytical functions), which copete to access the satellite channel. obined UMD is aied at approaching the perforance obtained for each perforance etric when there is no conflict aong the to access the channel. In short, it assigns the bandwth so to approach a non-copetitive situation where each etric sees the overall channel bandwth availability as close as possible. In ore detail, in this work two kinds of perforance etric have been consered: the Packet Loss Probability, which is a typical QoS etric for the TP based traffic and the Average Delay, which is typical for UDP based traffic. The allocation ethod is tested through the ns siulator by using TP and UDP traffic generators and by varying the fading level of the satellite channel over tie. obined UMD has been copared with other approaches taken fro the literature in the field. Keywords-Satellite Systes, QoS Metrics, Multi-Objective Prograing, Bandwth Allocation, Perforance Evaluation. I. INTRODUTION In Satellite environent [1] bit errors caused by noise and atospheric conditions (e.g., rain fading) are ajor issue and object of investigation. High Bit Error Rates (BERs) affect the ability of the satellite channel to offer reliable data transission. Forward Error orrection (FE) oding, typically eployed in these systes, is able to copensate for such errors, trading bandwth for effective data (as shown in [, 3] and synthetically reported in Section II.A.). On the other hand, Quality of (QoS) is the ability of a network eleent (e.g. an application, host, router or, in this case, satellite gateway) to have soe level of assurance so that its traffic and service requireents can be satisfied. In the consered environent counication detrient due to BER and Fading causes QoS degradation. Allocating the bandwth properly aong satellite earth stations, which can be affected by different noise or fading levels, is iportant to itigate the proble and to increase the proved QoS level. The rationale under this paper is consering bandwth allocation as a copetitive proble, by extending the concept proposed in [], where each station is represented by a set of perforance functions, each of the representative of a traffic with different QoS sensitivity, which needs to be iniied at cost of the others. All functions ust be iniied siultaneously. In ore detail, in this work, data traffic has been categoried into two ajor types with different QoS sensitivity: delay sensitive traffic (real-tie) and loss sensitive traffic (non-real-tie) traffic. applications are ultiedia applications such as veo conferencing, and Internet telephony (VoIP) typically based on the UDP transport layer protocol. Loss sensitive applications are those not involving ultiedia data, such as FTP, that use the TP transport layer protocol. In practice, stations, and for each of the, traffics with different QoS requireents copete for bandwth. It is the definition of the Multi-Objective Prograing (MOP) class of probles, which is the base of the ethod eployed in the paper. obined Utopia Miniu Distance (obined UMD) is aied at approaching the eal perforance, which theoretically happens when each single perforance function of an earth station is consered alone and has the availability of all channel bandwth. obined UMD approach allows, differently fro other ethodologies in the literature, the cobined optiiation aong copetitive and heterogeneous QoS etrics. As previously entioned, two traffic classes have been consered: Traffic and Traffic. Each of the is supposed to have a dedicated buffer. In the heterogeneous QoS etric sensitivity case consered, previously proposed approaches in the literature, such as the optiiation of the su of the perforance functions, would not work well. For exaple, if the buffers have equal and sall sies the packet loss probability of the loss sensitive traffic is very high. Vice versa, being low the nuber of packets in the buffers the delay (seen as queue waiting tie) of the delay sensitive traffic is very low. If the optiiation approach, used to allocate the bandwth aong stations, is based on the su of all etrics the delay ight not play any role. The MOP approach solve this possible inconvenient. The paper is structured as follows: Section II introduces the Network Structure and the control architecture. The foraliation of the bandwth allocation as a MOP proble and the obined Utopia Miniu Distance criterion is presented in Section III. Section IV reports an introductive perforance evaluation obtained through the ns siulator. Section V lists the conclusions. 978-1-444-75-9/8/$5. 8 IEEE 195
II. NETWORK STRUTURE AND ONTROL ARHITETURE A. Geostationary Satellite Network The network consered is coposed of earth stations connected through a satellite link. Each user requests a TP/IP service (e.g., Web page, File transfer or a VoIP session) by using the satellite channel itself (or also other counication edia). After receiving the request ISPs send traffic through the earth stations and the satellite link. To carry out the process, each earth station conveys traffic fro the directly connected ISPs and accesses the channel in copetition with the other earth stations. In the consered satellite network, as entioned in the introduction of the work, noise and fading effects, which ay affect the QoS perforance of the network, are odelled as bandwth reduction. Fro the practical viewpoint, it eans using a FE code where each earth station ay adaptively change the aount of redundancy bits (e.g. the correction power of the code) in dependence on noise or fading, so reducing the real bandwth availability. Matheatically, it real eans that the bandwth available for the -th station is coposed of the noinal bandwth and of the factor β, which is, in this paper, a variable paraeter contained in the interval [, 1]. Forally, as in [, real 3], = β. A specific value β corresponds to a fixed attenuation level seen by the -th station. An exaple of the apping between the arrier Power to One-Se Noise Spectral Density Ratio ( N ) and the β paraeter is contained in reference [3]. B. ontrol Architecture The control architecture is based on the presence of decision entities, also called Decision Makers (DMs) as reported in Fig. 1. In general, it can be used one DM for the whole syste in a centralied way, where an earth station or the satellite itself, if switching on board is allowed, represents the single DM that anages and proves stations with a portion of the overall bandwth (e.g., TDMA slots). Alternatively, ay be used a distributed ipleentation of the bandwth allocation process where one DM for each station is eployed. In this work, the second approach has been proved: each station has a DM that anages the bandwth distribution independently of each others (Fig. ). Fro the structural viewpoint, each station has a battery of buffer, one for each kind of traffic. In this case, one buffer is dedicated to Traffic (TP based) and one dedicated to Traffic Source (UDP based). The first buffer is eployed to store and, consequently send, the TP packet (e.g., traffic generated fro TP based sources such as FTP sessions or Web Browsing applications); the second buffer is used by UDP traffic (e.g., traffic generated fro UDP based sources such as VoIP sessions or Veo Streaing applications). Each buffer of a station has a service capacity that is a portion of the capacity allocated, by the DM(s) to the station itself. Fig. 1. Station Station Z-1 Bandwth Allocation Syste ontrol Architecture. III. apacity apacity apacity apacity Decision Maker Decision Maker Satellite hannel LOSS AND DELAY OMBINED OPTIMIZATION A. Multiple QoS Metrics Proble Forulation (MOP) The practical ai of the allocator is the provision of bandwth to each k -th k [, K 1 ], k buffer server of each -th station [, Z 1 ], by splitting the overall capacity available aong the buffers (the copetitive entities of the proble). Each buffer is eployed to serve a specific kind of traffic and, as a consequence, it is represented by a specific perforance etric. Analytically, the bandwth allocation defined as a Multi Objective Prograing (MOP) proble ay be foralied as: opt = opt {,,... opt, 1 ;... opt,,... opt, 1 ;... opt 1,,... opt K K Z Z 1, K 1} { ( )} ( ) Z K Z K = arg in F ; F : D where: = {,,...,, K 1 ;...,,,...,, K 1 ;..., Z 1,,..., Z 1, K 1}, D, is the vector of the capacities assignable to the earth stations buffers; the eleent, k,, Z 1, k, K 1,, k is referred to the k -th [ ] [ ] opt buffer of the -th station; D, is the vector of the Z K optial allocation; and D represents the doain of the vector of functions. The solution has to respect the constraint: (1) 196
where tot Z 1K 1, k = tot () = k= is the available overall capacity. ( ) dependent on the vector, is the perforance vector { ( ) = f, (, ),... f, K 1 (, K 1 );... f, (, ) fk, 1 ( K, 1 ) fz 1, K 1 ( Z 1, K 1 ) fz 1, K 1 ( Z 1, K 1 )} k -th [ ] [ ] F...,... ;......,... F, The single,, Z 1, k, K 1 Z, K perforance function is a coponent of the vector. Each perforance function f, k (, k) (or objective) of the syste is defined here as the average packet loss probability for the TP traffic (if k = ) and as the average delay for the UDP traffic (if k = 1 ). As a consequence, K = possible perforance etrics have been consered. Actually any other convex and decreasing function of the bandwth ay be used. The packet loss probability for TP traffic and the delay for UDP sees a reasonable choice but it ay be regarded also as an operative exaple for the theory presented. B. TP Packet Loss Probability (PLP)Function The used TP packet loss probability Ploss () is a function of the bandwth (, ) as well as of the nuber of TP active sources ( N ) and of the fading level ( β ), for each P = f ay be expressed as: station. () () loss, (,,, β ) loss P N = (,, ) = 3N 3b r+ 1 RTT + Q ( ) β 1 where: b is the nuber of TP packets covered by one acknowledgent; r is the reduction factor of the TP transission window during the ongestion Avoance phase (typically r = 1 );, is the bandwth seen by the TP aggregate of the -th earth station expressed in packets/s (, =, d, where d is the TP packet sie, always fixed in this paper); RTT is the Round Trip Tie; Q, is the buffer sie, expressed in packets, of the -th earth station dedicated to the TP traffic. The used PLP ( Ploss () ) is a onotone decreasing and convex function,, [, Z 1 ],, and it consers the effect of the channel state because it is also a function of the β paraeter. The odel is val at regie condition of the TP senders. (3) (4). UDP Average Delay (AD) Function The UDP average delay, defined as the delay spent in the UDP buffer of the -th earth station, D () is a function of the bandwth (,1 ) as well as of the nuber of UDP active sources ( M ) and of the fading level ( β ), for each station. D () = f,1 () ay be expressed taking as reference the well known M/M/1/X odel, where X is equal to the overall storage capacity in packets (buffer sie plus packet in service), of the UDP buffer. In this case, for each UDP buffer of earth stations X = Q,1 + 1. Starting fro the M/M/1/X odel hypothesises the average delay is: ( Q,1+ 1 Q,1 ) ρ ρ + D (,1, M, β) = µ 1 Q,1 ρ + (5) 1 ρ where: µ ( β) = d is the UDP buffer service,1 1 capacity of the -th earth station; d 1 is the UDP packet sie, always fixed in this paper; ρ = λ µ is the offered load to the UDP buffer of the -th earth station; λ = λ is the M 1 = overall arrival rate of the UDP packets in its dedicated queue ( λ is the generation rate of the single -th UDP source); Q,1 is the UDP buffer sie, expressed in packets, of the -th earth station. Also the used AD ( D () ) is a onotone decreasing and convex function,1, [, Z 1 ],, and it consers the effect of the channel state because it is also a function of the β paraeter. The odel is val at regie condition of the UDP senders. It is worth noting that the odel proposed, whose coputation has been oitted for the sake of synthesis, has been used by relaxing the hypothesis of exponential distribution of the service tie. In this paper a deterinistic service tie has been applied. D. Multiple QoS Metrics Allocation (obined UMD) In general, the proble defined above, is a Multi Object Prograing proble where each consered function f, k (, k) represents a single copetitive cost function. In other words, a single perforance function copetes with the others for bandwth. The optial solution for MOP probles is called POP-Pareto Optial Point, coherently with the classical MOP theory. The Utopia Miniu Distance ethod is a flexible ethodology that allows the resolution of the allocation proble (1). It bases its decision only on the eal solution of the proble: the so called utopia point. In ore detail, the eal perforance vector, in the case of this work, is: 197
where F { ( ) = f, (, ) f, K 1 (, K 1 ), (, ), K 1 (, K 1) Z 1, ( Z 1, ) Z 1, K 1 ( Z 1, K 1 )}... f,... f ;...... f,... f ( ) ( ),... ;... f = f, k, k in, k, k,, k, tot k, Fro equation (7), called single objective proble, it is clear that the optial solution is given by k, = tot, [, Z 1 ], k [, K 1]. So, = {,,..., } tot tot tot (6) (7). Obviously it is a physically unfeasible condition that can be only approached due to constraint (). Starting fro the definition of the eal perforance vector, the proble in equation (1) can be solved with the following allocation (coherently with []): opt = arg in F F (8) ( ) ( ) where is the Euclean nor. The proposed technique allows iniiing the distance between the perforance vector and the eal solution of the proble. Obviously, the iniiation is carried out under the constraint (). IV. PERFORMANE EVALUATION The ai of this perforance evaluation is to evaluate the bandwth allocation ethod functionalities in ters of PLP and AD. The action is fulfilled by using an ns based siulator, where the optiiation procedures have been ipleented. In the following tests, the coparisons have got by varying the fading conditions, in practice a given behaviour of the β paraeter over tie has been used in the siulations for each earth station consered. The allocator acts periodically (each T a [s]). In each allocation instant, each DM (Fig. ) knows the fading level and the traffic paraeters, related to its earth station, through a specific signalling procedure. After that, DMs prove the bandwth allocation, by solving equation (8), in a negligible coputation tie T c ( T c T a ). The network scenario consered is coposed of Z = 4 earth stations: Stations fro to are always in clear sky condition ( β always equal to 1 [, ] ), Station 3 varies its fading level, according to real fading levels taken fro [3], over tie as ade explicit in the following figure: BETA Value 1.9.8.7.6.5.4.3..1 5 1 15 5 3 35 Fig.. Fading Level Variation. Tie [s] Each station gathers traffic fro TP and UDP sources and transits it to the terinal users through the Satellite syste. The nuber of active TP and UDP sources is set to N = M = 5, [, Z 1]. The overall bandwth available tot is set to 8 [Mb/s] and the TP and UDP buffer sie Q, = Q,1 is set to 1 packets (of d = d1 = 15 bytes) for each earth station. The Round Trip Tie ( RTT ) value eployed in the coputation of the traffic odel is supposed fixed and equal to 51 [s] for all the stations. The allocation control acts each T a = 3 [s] and in all cases the siulated tie is always fixed and equal to 36 [s]. The TP sources actives a FTP session at the beginning of the siulations. Each FTP transfer has been set as a persistent session for the overall duration of the siulations: in practice, sources have always packets to send. The UDP sources are consered Poissonian packet generator with fixed λ equal to 1 [packets/s]. Each buffer is ipleented as a Dubbell topology with a single coon receiving node. The topology is coposed of source agents (which are nodes): 1 agents active 1 TP connection and the others 1 UDP connection. They send their packets to earth stations by using not congested and weband links that do not represent bottlenecks during siulations. An earth station is, in practice, a pair of buffers with storing capacities equal to Q and Q 1 packet, where packets sent fro sources are conveyed and forwarded if no congestion events are experienced. The service capacity, in [b/s], of the buffers of an earth station, is the bandwth allocated to it and the effect of the fading is consered by using the odel entioned in Section II.A: the fading is supposed copletely copensated by using FE schees (no channel errors are consered in the siulations) and their ipact is a ere bandwth reduction represented by the β paraeter. In Figs. 3 and 4, the easured Packet Loss Probability has been reported. In Fig. 3 a clear sky station has been consered (ore specifically Station ); in Fig. 4 the PLP perforance of the faded station (Station 3) has been reported. The obined UMD technique described in previous sections has been copared with a siple STATI approach (each buffer of the earth station receive a fixed quantity of bandwth equal to 1 Z K of the overall available capacity) and with a ethod, 198
taken fro the literature [4] here called VALUE, where the allocation is obtained by iniiing the su of the perforance functions. The three techniques have siilar PLP perforance if the earth stations see good channel conditions ( < t 5 ). When the fading becoes severe ( 5 < t 36 ) VALUE and UMD obviously have better perforance with respect to the STATI approach and, in ore detail, obined UMD as slightly preferable PLP, for the faded station, aong all the proposed techniques. Measured Packet Loss.6.5.4.3..1 5 1 15 5 3 35 Tie [s] STATI VALUE UMD Fig. 3. Packet Loss Perforance oparison (lear Sky Station). Measured Packet Loss.1.9.8.7.6.5.4.3..1 5 1 15 5 3 35 Tie [s] STATI VALUE UMD Fig. 4. Packet Loss Perforance oparison (Faded Station). Figs. 5 and 6 report the sae coparison above described but related to the Average Delay. In this case the advantage of the obined UMD ethod is really outstanding: in all cases the AD perfored by the proposed allocation is better than the other consered approaches and, in particular, consering the faded station (Fig. 6) in the tie period 5 < t 36, the obined UMD allows reaching AD of about [s] while the VALUE has AD of about 3 [s] and the STATI approach perfor AD of about 7 [s]. The differences are, obviously, very significant. Measured Delay [s].14.1.1.8.6.4. 1 3 Tie [s] STATI VALUE UMD Fig. 5. Delay Perforance oparison (lear Sky Station). Measured Delay [s].8.7.6.5.4.3..1 5 1 15 5 3 35 Tie [s] STATI VALUE UMD Fig. 6. Delay Perforance oparison (Faded Station). V. ONLUSIONS The paper describes and analyses the obined UMD allocation schee for satellite counications. It is suited to be used in Satellite Networks where heterogeneous perforance etrics have to be siultaneously optiied. The theoretical fraework consered is the Multi Objective Prograing Optiiation. The paper investigates the behaviour of the obined UMD schee by consering the traffic as a superposition of TP and UDP sources, opportunely odelled, and copares the results. REFERENES [1] S. Kota, K. Pahlavan, P. A. Leppänen, Broadband Satellite ounications for Internet Access, Kluwer Accadeic Publishers, Boston, 4. [] I. Bisio, M. Marchese, Miniu Distance Bandwth Allocation over Space ounications, IEEE ounications Letters, vol. 11, no. 1, pp. 19-1, January 7. [3] N. elandroni, F. Davoli, E. Ferro, Static and Dynaic Resource Allocation in a Multiservice Satellite Network with Fading, International Journal of Satellite ounications, vol. 1, no. 4-5, pp. 469-488, July-October 3. [4] R. Bolla, F. Davoli, M. Marchese, Adaptive Bandwth Allocation Methods in the Satellite Environent, In Proc. International ounications onference (I 1), Helsinki, Finland, June 1, pp. 3183-319. 199