QoS Frmework for SIP Signlling Alexnder A. Kist nd Richrd J. Hrris RMIT University Melourne BOX 2476V, Victori 3001, Austrli Emil: kist@ieee.org, richrd@ctt.rmit.edu.u ABSTRACT The Session Initition Protocol (SIP) is widely ccepted s the IETF lterntive the ITU-T H.323 teleconferencing protocol to enle cll nd medi session mngement nd control. It is lso used in crrier grde environments, such s the IP Multimedi Susystem (IMS) of the 3rd Genertion Prtnership Project (3GPP) in emerging Universl Moile Telecommunictions System (UMTS) networks. This pper proposes frmework tht enles the Qulity of Service (QoS) provisioning for signlling messges. The contriutions re twofold: Firstly, it defines n overll frmework to enle QoS provisioning for SIP signlling messges in crrier grde networks. Secondly, it outlines existing work tht fits into the frmework nd identifies res tht require further investigtion to implement the concepts in crrier grde networks. KEY WORDS Session Initition Protocol (SIP), Qulity of Service (QoS), Signlling, IP Multimedi Susystem (IMS), Universl Moile Telecommunictions System (UMTS) 1 Introduction Recent yers hve seen the IETF s Session Initition Protocol (SIP) (RFC 3261 [1]) ecome the premier protocol choice for user loction, session setup nd session mngement tsks in IP environments. For exmple, the 3rd Genertion Prtnership Project (3GPP), which is glol inititive to develop stndrds nd specifictions for next genertion Universl Moile Telecommunictions System (UMTS) networks, uses SIP s the signlling protocol for its IP Multimedi Susystem (IMS) in Relese 5 (3GPP Technicl Specifiction 23.228 (R5) [2], 24.228 (R5) [3], 24.229 (R5) [4]). Other exmples include, ut re not limited to, the use of SIP for ir trffic control pplictions [5] nd in IPv6 environments [6]. Where Qulity of Service (QoS) issues for medi trnsport in IP networks hs een in the focus of the reserch community for mny yers, QoS provisioning specificlly for signlling messges hs received less ttention. Signlling messge dely nd session initition dely ws ddressed y Eyers [7] nd Curcio [8]. Erlier work [9] outlined the need for QoS considertions on the SIP lyer to provide equivlent telephony services, in prticulr, if SIP is to e used in 3GPP crrier-grde networks. This need is minly sed on two spects: signlling network resources re shred with other services nd gurntees to customers re only elievle if service levels re defined. This pper introduces n overll frmework tht enles QoS provisioning for SIP signlling trffic. It consists of severl prts, including: Virtul SIP Links (VSLs), which llow the definition of virtul SIP networks, methodologies to clculte the size of signlling flows, dynmic resource lloction schemes nd SIP routing methods. Note tht the IMS is used s n exmple network in this pper, ut the concepts cn e used in ny environment tht requires QoS SIP signlling. 3GPP uses it s own nottion nd introduces numer of SIP proxy servers clled Cll Session Control Function (CSCF). Commercil service providers require these servers to control session signlling messge flows nd enle uthentiction, illing, service provisioning, etc. Proxy CSCFs (s) re the network entry points for the User Equipment (UE). Serving CSCFs (s) hold copy of the user profile, record session stte informtion nd provide higher level session hndling functions. Interrogting CSCFs (s) re network entry points for terminting sessions nd decide the messge routing to s. The contriutions of this work re twofold: Firstly, it outlines the motivtion nd it defines n overll frmework tht enles QoS provisioning for SIP signlling messges in crrier grde networks. Secondly, it summrizes existing work tht fits into the frmework nd identifies res tht require further investigtion to use the concepts in crrier grde networks. Both tsks id the ultimte gol of QoS provisioning for SIP service users in crrier grde networks. The pper is orgnised s follows: It ddresses the SIP signlling lyer first; nd then introduces the Virtul SIP Overly Network in Section (VSON) 2. Section 3 summrises the frmework nd how the different concepts interct. Virtul SIP links re prt of the VSON definition nd re discussed in Section
4. VSLs cn use Dynmic Resource Alloction (DRA) which is explined in Section 5. VSLs nd DRA rely on trffic estimtion nd flow nlysis which is discussed in Section 6. Messges on the VSON cn tke lterntive pths; SIP messge routing is discussed in Section 7. Section 8 outlines res tht require further ttention. 2 SIP Lyer Astrction In existing SIP signlling configurtions, IP networks provide trnsport service for SIP messges. Any node tht is connected to the IP network nd hs n IP ddress, is glolly routle. In this sitution, ll nodes re logiclly fully meshed. The sme is true for SIP signlling nodes tht re connected to IP networks. An exmple of such network is shown in Figure 1. It depicts the Trnsport/Network Lyer nd the SIP/Appliction Lyer. The cloud symolises the underlying trnsport network, nd the nodes symolise SIP servers. SIP/Appliction Lyer UE1 UE2 Trnsport/Network Lyer Figure 1. SIP nd Network Lyer The virtul network is reduced to well-defined links nd nodes. Relevnt issues of the underlying network hve to e mpped onto this lyer. This includes, ut is not limited to, delys nd it errors. The VSON defines signlling environment tht enles the gurntee of service levels. Known methodologies cn e pplied nd new strtegies cn e developed for this virtul overly network. The following sections outline issues tht re relevnt to the VSON. These include the definition of the connections etween SIP nodes, possile resource lloction strtegies for this link, signlling trffic mngement nd SIP messge routing. 3 Frmework The concept to provide QoS for SIP signlling consists of severl models nd methods tht interct with ech other. The overll gol is to define virtul SIP overly network with service gurntees s well s predictle nd ccountle ehviour. The relevnt res include: the SIP nodes, the connection etween the nodes nd the messge routing in etween the nodes. SIP nodes re server implementtions tht re similr to other existing Internet services; the connections nd the routing re specific to the VSON. Figure 3 depicts the different locks of the frmework nd their interction. The VSON (4) consists of SIP nodes nd Virtul SIP Links (2). The VSL re dimensioned nd set up y the Dynmic Resource Alloction DRA. The trffic model (1) is used y DRA nd VSL to dimension these resources. Since the VSON defines it s own virtul network, routing strtegies (5) re required on this level. To e le to use such The IP resources of generl-purpose trnsport networks re lso used y services other thn signlling, viz: This network configurtion provides no dedicted signlling trffic resources. To define QoS levels solely for SIP, the SIP lyer hs to e seprted from the trnsport lyer. QoS mesures on the SIP lyer should e uncoupled from used trnsport technologies. On this SIP lyer, service levels cn e defined nd n integrted QoS service concept cn e developed. Figure 2 depicts such trnsport independent virtul SIP overly network. 4 2 1 VSON 3 5 VSL DRA Routing Trffic Model 6 Figure 3. SIP QoS Frmework SIP/Appliction Lyer UE1 Figure 2. VSON Lyer UE2 scheme in n environment where signlling trffic nd other services shre network resources, it is importnt to gurntee QoS. To enle QoS on the ppliction lyer, the trnsport erers hve to provide QoS methodologies to protect trffic requirements for different services. The service level greements, which define these needs, should e of dynmic nd dptle nture. The detil of the Service Level Agreement (SLA) ssignment process nd implementtion is not in the scope of this work.
4 Virtul SIP Links Virtul SIP links re the strction of connections etween SIP nodes. VSLs re logiclly locted on the SIP lyer nd connect two SIP nodes. As outlined erlier, this is possile since the full connectivity provided y the IP network is not necessry to fulfil the functionl requirements of VSONs. The VSL definition is motivted y three min resons: To define required resources, to enle trffic clcultions nd to enhnce the performnce of VSONs. To e le to introduce qulity spect to VSONs, comprle nd predictle prmeters re required to clssify the SIP signlling trffic. A well known method of clssifying trffic is the Leky Bucket (LB) methodology which is used to define VSLs. VSLs do not only define the flow size, ut lso limit the mximum queueing dely. Trffic models require the input of messge loss proilities. If resources re limited nd clssified y LBs, messge loss oundries cn e estimted nd trffic flows cn e clculted [10]. VSLs hve no physicl equivlent. They re logicl origin-destintion ssocitions with distinct trffic requirements. VSLs re defined y their Trffic Specifictions (TSpec), i.e. the men rte, the pek rte, the minimum policed messge size, the urst size nd the constrints of the messge loss proility. These trffic specifictions define leky ucket which enles trffic policing nd shping. Thus, VSLs implement de fcto dmission control for signlling trffic. VSLs mke connections comprle nd ccountle nd they define cler interfce points in the network. VSLs re dimensioned for mximum numer of messges nd defined mximum messge loss proility. Models tht re ddressed in Section 6 id in the clcultion of the LB prmeters. Once the men flow size is known, the required pek rte nd uffer size cn e clculted to comply with minimum messge loss proility. All defining VSL prmeters re known. To deploy VSL these prmeters hve to e ccepted y the trnsport network. The trnsport network requires sufficient QoS technologies nd hs to e le to provide QoS resources. VSLs require no dditionl hrdwre. Softwre modules in SIP servers cn implement their functionlity. In principle the use of VSLs for connections etween SIP nodes is ritrry. Connections with nd without defined QoS conditions cn coexist. If end-toend QoS gurntees re required, ll connections hve to support VSLs. The mechnism of dimensioning nd deploying cn e executed on the fly y Dynmic Resource Alloction (DRA). 5 Dynmic Resource Alloction The im of dynmic resource lloction is twofold. Firstly, it is methodology to enle the QoS provisioning for the virtul SIP signlling network. Secondly, it chieves the dimensioning utomticlly on the fly. DRA uses cpilities tht mixed services IP trnsport networks provide. If the trnsport network supports dynmic service level greements, nodes hve the ility to reserve more/less ndwidth from the trnsport network to dpt to new requirements. Dynmic resource lloction requires severl supporting technologies. The underlying trnsport network hs to support dynmic SLAs i.e. the ility to negotite the resources with the network. It lso requires tht VSLs e used on the SIP lyer. To keep the perceived QoS constnt, the DRA process dimensions VSLs for SIP trffic. To e le to dimension VSLs ppropritely, knowledge out network trffic flows is required. In such context, the dynmic resource lloction methodology llows the utomted configurtion of resources nd ensures QoS for signlling. DRA works s follows [11]: DRA oserves connections of SIP servers nd dpt the resources s required. To chieve such functionlity severl su functions re required: Dt out the VSL stte re collected s messge rrivls per time unit nd verge messge size. If these vlues cross upper or lower threshold limits, the scheme tries to increse or decrese the resource suscription. To ensure tht ctul trends re detected nd not rndom fluctution, the two thresholds need to hve minimum distnce. The distnce is estimted y known sttisticl methods. The ctul size of the resources tht re suscried cn e clculted y using flow model nd methods for VSL dimensioning. The resource suscription is executed if thresholds re crossed. Additionl resources re requested or resources re returned to the trnsport network. If the trnsport network cnnot fulfil n increse-request, the resource suscription remins unchnged. Both dynmic resource lloction nd dynmic routing (Section 7) re processes tht re executed on the fly, so it is importnt tht they operte on different time scles. Frequent fluctutions in user/messge numers cn e hndled y dynmic routing schemes, long-term trffic shifts re hndled y DRA. DRA cn yield considerle resource svings, since unused resources re returned to the trnsport network. This is in prticulr useful if the resources re illed on SLA ses. 6 SIP Flows nd Trffic Estimtion To e le to use the methodology of VSLs nd to get fundmentl understnding of trffic flows in SIP overly networks, it is necessry to hve models tht descrie SIP signlling messge trffic. These models investigte the influence of messge loss on the trnsmission medi on SIP trffic nd ccount for the fct
tht SIP messges grow in size when they pss through the network. A SIP fluid flow model does not tke different SIP messges into ccount, ut investigtes how their trffic ggregtes cn e used to nlyse flows. It llows lso the clcultion of the minimum it error requirements for trnsport network connections. Such flow model ws introduced in [12]. A simplified version of this model relies on two prmeters: the numer of messges tht re sent nd the verge size of these messges. Both prmeters re loclly ville in nodes. Thus it is possile to estimte the men flow size. The reminder of this section outlines simplified model tht llows flow size estimtion etween two SIP nodes. Firstly the clcultion of the messge size is shown. The signlling trffic etween two nodes depends on the numer of messges tht re sent etween these nodes, including the numer of messges tht hd to e resent. Figure 4 depicts n exmpleconnection etween two nodes nd. The men size of messges leving node is denoted y m, the numer of originl messges sent on the connection under the ssumption of zero loss is denoted y n,. The connection hs messge loss proility of P E (, ). The numer of messges tht were dropped nd resent on connection, nd eyond re denoted y d, nd the numer of messges tht were dropped eyond nd trversed, is denoted y d,. Depending on d m n, P E (,) d Figure 4. Exmple Connection the locl knowledge, the flow on link, cn either e clculted for node y Eqution (1) or for node y Eqution (2) F L, = m (n, + d, ) In the cse of node, d, node, d, is known. ( F L, = m n, (1) is known nd in the cse of 1 + P E(, ) + d, 1 P E (, ) ) (2) The estimtion of the verge messge size m is discussed next. SIP messges consist of messge heder nd messge ody. The messge ody is usully used for the session description which uses the Session Description Protocol (SDP). SIP messges pper with nd without messge ody, nd therefore vry considerly in size. Stte full servers tht re trversed y requests re recorded in the vi heder field nd in some cses in the route-record field of the messge. This cuses the SIP messges to grow while they pss through the network. Since the oservtions here re locl the messge size chnge hs no impct in this context. The current messge size estimte is of interest. A locl node cn clculte recursive men of ll processed messges online. Eqution (3) shows the clcultion. m(t + 1) = m(t) + M(t + 1) m(t) n (3) The verging count is denoted y n, i.e. the numer of smples tht re considered, the current messge size is M nd m is verge. This vlue is simpler to clculte thn the moving verge. It does not require memory for the pst n vlues. More recent mesurements hve lrger impct thn older mesurements. This simple model llows the trffic estimtion etween two SIP nodes y dynmiclly mesuring strightforwrd prmeters, i.e. messge size, messge numer nd the numer of dropped messges. The estimtes help the dimensioning of signlling resources etween these nodes s outlined in Section 5. 7 SIP Messge Routing Messges on the VSON lyer cn tke lterntive routes to their destintion, nd therefore messge routing is required. Most SIP messge routing decisions re mde during the registrtion of users, since intermedite nodes record registrtion stte nd/or session stte informtion. Once ssocitions re formed they re persistent for susequent requests. In principle, messge routing depends on two fctors: Firstly the connectivity on the VSON lyer, i.e. VSLs re defined etween nodes nd secondly, it depends on opertor-specific policies nd routing strtegies. The first depends in generl on functionl specifictions nd the network structure, e.g. for 3GPP, the defined network entry point for UEs re s, the lter cn e specific to opertors or single networks. SIP messge routing cn e divided into two res: Routing decisions tht resemle lod lncing schemes/server ssignment prolems; nd routing prolems tht cn e solved y shortest pth like routing methodologies. For 3GPP, the first re relevnt for routes from s to s nd the lter concern routing decisions tht re required for routes from UEs/s/s to s. First lod lncing nd overflow routing is ddressed. The chllenge of efficient messge routing is similr to existing server/resource lloction prolems which include server lod lncing, hsh routing nd we cching schemes. In recent yers, these res hve
een mjor reserch trgets. The SIP Messge Overflow Routing Scheme (SMORS) is introduced [13]. It trgets efficient messge routing for high volume SIP servers, in the 3GPP IMSs context nd it specificlly ddresses messge routing etween s nd S- CSCFs. SMORS ssigns users/messges to servers on the sis of generic routing informtion. If threshold is reched, new users/messges re overflown to dditionl servers. The principl concept of overflow routing is not new nd hs een extensively studied in the context of Pulic Switched Telephone Networks (PSTNs). SMORS hs severl enefits: It is fst nd requires little processing for susequent messge routing, it hs lod lncing, ckup provisioning nd overflow routing cpilities. One mjor dvntge of this scheme is tht the user-to- ssignment is flexile. s cn e ssigned, y user priority, service suscription or lpheticlly y user nme. The ssignments do not hve to e clculted t runtime; these ssocitions cn e pre-clculted. Shortest pth-like routing protocols nd methodologies in the IP routing context re well-understood nd remin continuing focus of the reserch community. Shortest pth routing lgorithms find pths etween n Origin-Destintion Node Pir (OD-pir) tht stisfy miniml optimistion metric. Trditionl routing protocols such s Open Shortest Pth First (OSPF) use sclr metrics to optimise the pths. Mny possile lterntive metrics re known, ut the inverse of the cpcity is most commonly used in IP networks. Fctors tht hve n impct on the messge routing on the VSL lyer re the hierrchicl nd logicl setup defined y the network structure, the vilility of resources in nodes to hndle dditionl requests, resource vilility for dditionl signlling trffic on the trnsmission medi, the close proximity of nodes nd the qulity on the connection. The routing process should consider these fctors. [14] proposed multidimensionl metric. During defined stges in the network opertion one metric within the multidimensionl metric will dominte the others. If, for exmple, two pths hve the sme numer of ville users nd similr delys of cceptle length, the pth with the mximum reliility should e selected. On the other hnd, core network connections re likely to hve similr reliilities. In this cse, the pth selection is minly sed on dely which is mesure for distnce in this context. If the numer of ville users on certin pth gets smller, routing hs to tke this into ccount. It is importnt to note tht in the comined metric, one metric domintes the others t given time, the metric is not simple weighted verge clcultion. 8 Further Work This pper mentioned numer of models nd methodologies tht cn work together to provide QoS for SIP signlling messges nd improve the service experience of users. This section outlines res for further reserch directions. 8.1 Trnsport Protocol One issue tht hs considerle impct on QoS issues is the trnsport protocol tht is used. In principle, the SIP protocol is trnsport protocol independent. The drwck of the defult trnsport protocol UDP is its unreliility nd the disdvntge of TCP is the protocols long connection set up time. UDPs unreliility cn cuse prolems in SIP overly networks with the high numer of intermedite links nd SIP s end-toend reliility [9]. The Strem Control Trnsmission Protocol (SCTP) [15] could e n lterntive, with considerle dvntges for the session setup times nd throughput. Internet Drft [16] discusses the SCTP protocol in conjunction with its use in SIP. It suggests tht most of the enefits of SCTP occur under loss conditions. The sitution in QoS signlling domins is different from the sitution in generl Internet environments, therefore further study is required to investigte the impct of trnsport protocols on SIP QoS. 8.2 QoS Network VSLs ssume tht the trnsport network cn provide QoS resources to signlling trffic. Further study should investigte the interction etween VSLs nd the trnsport network, nd possile implictions. Furthermore, VSLs use dynmic SLAs. DRA ssumes tht these exist nd cn e used. Further study hs to focus on the exct implementtion of these dynmic SLAs, which my include SLA negotition nd further issues. 8.3 Generic Overly Networks Recent yers hve seen mny new peer-to-peer services emerge, mny of which define their own overly network. Overly networks re lso wy to provide dvnced services vi the pulic Internet tht re not provided y the network. In future IP networks, these services will e le to use QoS functionlities s well. Results such s the VSL principle, dynmic resource lloction nd routing might e pplicle in this context s well. These efforts cn lso include dislocted corporte computer networks which use IP resources to connect vrious sites, s well s pplictions tht define their own overly network nd intend to pply QoS
principles. Oviously, this outlines wide reserch re with numer of interesting reserch prolems to solve. 9 Conclusions This pper introduced frmework tht enles QoS provisioning for SIP signlling service in dvnced crrier grde networks. It outlined numer of technologies nd schemes tht cn id the QoS effort. Future work hs to investigte the prcticlity of these models in live networks. The mjor im of this pper ws to introduce some of these issues nd possile solutions. These issues might gin importnce, s more nd more commercil services will use SIP s session control nd mngement. Providing QoS for SIP signlling is core issue in providing QoS for consumers of future emerging (wireless) networks. Acknowledgements The uthors would like to thnk Ericsson AsiPcificL Austrli nd the Austrlin Telecommunictions Coopertive Reserch Centre (ATcrc) for their finncil support for this work. References [1] J.Rosenerg, H.Schulzrinne, G.Cmrillo, A. Johnston, J. Peterson, R. Sprks, M. Hndley, nd E. Schooler. SIP: Session Initition Protocol. IETF, June 2002. RFC 3261 (Osoletes RFC 2543). [2] 3rd Genertion Prtnership Project. IP Multimedi (IM) Susystem - Stge 2 (Relese 5), Jnury 2004. 3G TR 21.905 V5.11.0. [3] 3rd Genertion Prtnership Project. Signlling flows for the IP multimedi cll control sed on SIP nd SDP - Stge 3 (Relese 5), Decemer 2003. 3GPP TR 24.228 V5.7.0. [4] 3rd Genertion Prtnership Project. Internet Protocol (IP) multimedi cll control protocol sed on Session Initition Protocol (SIP) nd Session Description Protocol (SDP) - Stge 3 (Relese 5), Jnury 2004. 3GPP TR 24.229 V5.7.0. [5] K.Drilion, W.Kmpichler, nd K.Gschk. Eventsed rdio communiction signlling using the session initition protocol. In The 11th IEEE Interntionl Conference on Networks (ICON 2003), Sydney, Austrli, Septemer 2003. [6] J.Fiedler D.Sislem. SIP nd IPv6: Why nd how? Interntionl Symposium on Applictions nd the Internet (SAINT2003), Orlndo, USA, Jnury 2003. [7] T.Eyers nd H.Schulzrinne. Predicting Internet Telephony Cll Setup Dely. In In IPTel 2000 (First IP Telephony Workshop), April 2000. [8] I.D.D. Curcio nd M. Lundn. SIP cll setup dely in 3G networks. In The Seventh IEEE Symposium on Computers nd Communictions (ISCC 02), Tormin/Girdini Nxos, Itly, July 2002. [9] A.A. Kist nd R.J. Hrris. SIP Signlling Dely in 3GPP. In Proceedings of Sixth Interntionl Symposium on Communictions Interworking of IFIP - Interworking 2002, Perth Austrli, Octoer 13-16, Octoer 2002. [10] A.A. Kist nd R.J. Hrris. Using virtul SIP links to enle QoS for signlling. In The 11th IEEE Interntionl Conference on Networks (ICON 2003), Sydney, Austrli, Septemer 2003. [11] A.A. Kist nd R.J. Hrris. Dynmic resource lloction in 3GPP SIP overly networks. In Fourth Interntionl Conference on Informtion, Communictions & Signl Processing nd Fourth Pcific-Rim Conference on Multimedi (ICICS- PCM 2003), Singpore, Decemer 2003. [12] A.A. Kist nd R.J. Hrris. A simple model for clculting SIP signlling flows in 3GPP networks. In Second IFIP-TC6 Networking Conference 2002, Pis, Itly My 19-24, My 2002. [13] A.A. Kist nd R.J. Hrris. SIP messge overflow routing scheme (SMORS). In 2003 Austrlin Telecommunictions, Networks nd Applictions Conference (ATNAC), Melourne, Austrli, Decemer 2003. [14] A.A. Kist nd R.J. Hrris. SIP routing methodologies in 3GPP. In First Interntionl Working Conference on Performnce Modelling nd Evlution of Heterogeneous Networks (HET-NETs 03), Ilkley, West Yorkshire, U.K, July 2003. [15] R. Stewrt, Q. Xie, K. Morneult, C. Shrp, H. Schwrzuer, T. Tylor, I. Rytin, M. Kll, L. Zhng, nd V. Pxson. Strem Control Trnsmission Protocol. IETF, Octoer 2000. RFC 2960. [16] J. Rosenerg, H. Schulzrinne, nd G. Cmrillo. The Strem Control Trnsmission Protocol s Trnsport for the Session Initition Protocol. IETF, Novemer 2003. Internet Drft <drft-ietfsip-sctp-04.txt> (work in progress).