Effcent Loa-Balance IP Routng Scheme Base on Shortest Paths n Hose Moel E Ok May 28, 2009 The Unversty of Electro-Communcatons Ok Lab. Semnar, May 28, 2009 1
Outlne Backgroun on IP routng IP routng strategy Traffc moels Requrement for IP routng Propose routng scheme Optmal routng formulatons Performance evaluaton Conclusons Ok Lab. Semnar, May 28, 2009 2
Routng n IP networks What s goo routng? To utlze the network resources effcently an ncrease the network throughput. Approaches Route selecton IP packets are forware on the selecte routes. Loa balancng Traffc emans are splt among source an estnaton noes. Mnmzng the network congeston rato leas to ncrease atonal amssble traffc. Network congeston rato Mamum lnk loa of all network lnks Ths s a man target n ths stuy. Source Congeston Destnaton Ok Lab. Semnar, May 28, 2009 3
Several routng strateges Mult-Protocol Label Swtchng Traffc Engneerng (MPLS-TE)-base routng Shortest-path-base routng Optmum lnk weght control Two-Phase Routng (TPR) Smart Open Shortest Path Frst (S-OSPF) Ok Lab. Semnar, May 28, 2009 4
MPLS-TE-base routng Label Swtch Path (LSP) tunnels are use. Traffc emans are eplctly route an flebly splt among source an estnaton noes. However, Legacy networks manly employ shortest-path-base routng protocols such as OSPF. Alreay eploye IP routers n the legacy networks nee to be upgrae. Network operators nee to confgure an manage LSP tunnels between all ege noes to form a mesh-lke logcal topology. The number of tunnels ncreases n proporton to N 2, where N s the number of noes. It s esrable that an estng IP routng protocol stll n use shoul be utlze. Ok Lab. Semnar, May 28, 2009 5
Optmum lnk weght control Set optmum lnk weghts n OSPF-base networks. Traffc s route on shortest paths base on the compute optmum lnk weghts. However, When traffc emans are change, optmum lnk weghts are recalculate an IP routes are change. Changng routes frequently causes network nstablty, whch leas to packet loss an the formaton of loops. Source w=1 w=10 Destnaton w=1 w=10 w=1 w=1 Ok Lab. Semnar, May 28, 2009 6
Two-Phase Routng (TPR) Presente by Antc an Smljanc et al. [ICC 2008] Performs loa balancng an each flow s route accorng to the estng OSPF protocol, n two stages across ntermeate noes. Optmum strbuton ratos are obtane by solvng a Lnear Programmng (LP) problem. Requres the confguraton of IP tunnels, such as IP-n-IP an Generc Routng Encapsulaton (GRE) tunnels, between source an estnaton noes. IP tunnel wth shortest path IP tunnel wth shortest path Source Destnaton Traffc loa balancng Ok Lab. Semnar, May 28, 2009 7
Smart OSPF (S-OSPF) Presente by Mshra an Sahoo. [Globecom 2007] Source noes strbute traffc only to the neghbor noes wth optmum ratos, whch are obtane form the LP problem. No tunnel confguraton s requre. Neghbor noe Source Destnaton Traffc loa balancng Ok Lab. Semnar, May 28, 2009 8
Summary of four routng strateges MPLS routng Shortest-path routng Lnk weght control TPR S-OSPF Tunnel LSP tunnel Not requre IP tunnel Not requre Stablty Yes Yes Yes Routng performance Ieal Hgh Hgh Hgh Scalablty Yes Ok Lab. Semnar, May 28, 2009 9
Optmal routng for traffc moels Ppe moel Hose moel Ok Lab. Semnar, May 28, 2009 10
Ppe moel Assumpton on traffc moel Traffc matr, T={ }, s eactly known. : traffc eman from source noe p to estnaton noe q. Routng scheme Traffc emans are assume to be eplctly route an flebly splt among source an estnaton noes, by usng Mult-Protocol Label Swtchng (MPLS) technologes Noe 1 Noe 2 Noe 3 T Ok Lab. Semnar, May 28, 2009 11 11 21 31 12 22 31 13 23 33
Ppe moel (cont ) Uner the ppe moel, the most effcent routng s acheve. However, It s ffcult for network operators to get an eact traffc matr Because Measurement of traffc eman for each source an estnaton par s ffcult when the network sze s large. Traffc eman s often fluctuate. Measurement of traffc eman Fluctuaton of traffc eman T 11 21 31 12 22 31 13 23 33 Noe 1 Noe 3 Noe 2 0 tme Ok Lab. Semnar, May 28, 2009 12
Hose moel Easy for network operators to specfy the traffc as only the total outgong/ncomng traffc from/to ege noe p an ege noe q. The hose moel s specfe by: Outgong traffc Incomng traffc q p 3 1 3 1 Noe 1 Noe 3 Noe 2 2 2 p q T Ok Lab. Semnar, May 28, 2009 13 11 21 31 12 22 31 13 23 33 1 2 3 1 2 3
Requrements for IP routng Hgh network utlzaton. The routng scheme shoul utlze network resources effcently an ncrease network throughput. Easy eployment. Utlzng a routng protocol that s alreay eploye n the estng network s preferre. It shoul be easy to be scale n terms of network sze. Stablty aganst frequent traffc fluctuaton. The routng scheme shoul not allow the network to become unstable by frequently changng routes, whch leas to packet losses an loops. Smple traffc nformaton. Complete traffc matr nformaton shoul not be requre. The routng scheme shoul use the hose moel. Queston: Is there any routng scheme that satsfes the above requrements? Ok Lab. Semnar, May 28, 2009 14
Routng strateges an traffc moels MPLS routng Shortest-path routng Lnk weght control TPR S-OSPF Ppe moel Solve Solve Solve Solve Hose moel Solve Not solve A routng scheme base on S-OSPF for the hose moel s propose. Ok Lab. Semnar, May 28, 2009 15
Network moel G( V, E) :recte graph V : set of vertees Q V :set of ege noes E :set of lnks, c T : porton of traffc from noe p Q to noe q Q through lnk (, E : capacty of lnk (, E { }: traffc eman r : network cogeston rato, mamum value of all lnk utlzaton rates n the network Neghbor noe c Congeston rato, r Noe p Traffc loa balancng G( V, E) Ok Lab. Semnar, May 28, 2009 16 Noe q
S-OSPF routng wth ppe moel Optmzaton problem Objectve : mn r Constrants : j:(, E, jospf p, qq 0 0 j' j':( j', ) E p, q Q, p, q Q, r 1 p, ancestor p c 0, q, OSPF r,(, E 1, p, q Q,(, j' j':( j, ) E E 1, nethop (1a) (1b) (1c) (1) (1e) (1f) Gven parameters :, c Decson varables: r, Ths s a lnear programmng (LP) problem. Ok Lab. Semnar, May 28, 2009 17
0 0 S-OSPF routng wth hose moel Optmzaton problem Objectve : mn r Constrants : j:(, E, jospf p, qq j' j':( j', ) E p, q Q, p, q Q, 1, r 1 p, ancestor p c 0, q, OSPF r,(, p, q Q,(, j' j':( j, ) E E E 1, nethop (1a) (1b) (1c) (1) (1e) (1f) Gven parameters: The range of q p Ok Lab. Semnar, May 28, 2009 18 p q, c s gven by : Decson varables: r, (1g) (1h) Ths optmzaton problem s a lnear programmng (LP) one. However, t s mpossble to conser all possble combnatons of specfe by Eqs. (1g)-(1).
S-OSPF routng wth hose moel (cont ) The optmzaton problem s solve by the followng property, whch s obtane by ntroucng the ual theorem an etenng Chu s property for MPLS-TE to S-OSPF [ICC 07]. Property: acheves congeston rato r for all traffc matrces constrane by the ntermeate moel f an only f there est the followng non-negatve parameters such that ) ) pq ( ) p p p ( p) c pq ( p) ( q), r, for each (, for each (, E E an every p, q Q Ok Lab. Semnar, May 28, 2009 19
S-OSPF routng wth hose moel (cont ) The optmal routng problem s transforme nto the followng regular LP formulaton. Optmzaton problem Objectve : mn r Constrants : j:(, E, jospf p, qq pq ( p) ( q), (, E ( p), ( q) 0 0 p 0 r 1 j' j':( j', ) E ancestor c ( p) 0, p, q Q, j' j':( j, ) E r,(, E pq ( p) c 1, p, q Q,(, E p p, q, OSPF 1, p, q Q, p r,(, E nethop Ok Lab. Semnar, May 28, 2009 20
Performance evaluaton Network congeston ratos are compare Propose scheme, classcal Shortest Path Fnng (SPF), MPLS-TE, an TPR Smulaton assumptons Lnk capactes ranomly generate wth unform strbuton n the range of (80,120) s ranomly generate wth unform strbuton n the range of (0,100), p q q p Ok Lab. Semnar, May 28, 2009 21
Network moels Sample networks (a) Network 1 (b) Network 2 (c) Network 3 () Network 4 (e) Network 5 Ranom networks Ranomly generate uner the conton that average noe egree D s satsfe for a gven number of noes N. Ok Lab. Semnar, May 28, 2009 22
Comparsons of congeston rato n sample networks The propose scheme ramatcally reuces the congeston rato compare wth the classcal SPF. proves comparable routng performance wth MPLS-TE an TPR, whch requre functonal upgraes an/or complcate confguratons. Normalze congeston rato 1.0 0.8 0.6 0.4 0.2 0.0 Network 1 Network 2 Network 3 Network 4 Network 5 Propose scheme Classcal SPF MPLS-TE TPR Ok Lab. Semnar, May 28, 2009 23
Comparsons of congeston rato n ranom networks When a topology becomes mesh-lke topology, the propose scheme outperforms TPR. Normalze congeston rato 1.0 0.8 0.6 0.4 0.2 0.0 3 4 5 6 Average noe egree, D Propose scheme MPLS-TE TPR Ok Lab. Semnar, May 28, 2009 24
Number of noes epenency The congeston rato characterstcs of the four schemes are kept when the number of noes s vare. Normalze congeston rato 1.0 0.8 0.6 0.4 0.2 0.0 Propose scheme MPLS-TE TPR 0 10 20 30 40 Number of noes, N Ok Lab. Semnar, May 28, 2009 25
Conclusons The propose scheme oes not nee to know the traffc eman between all the source-estnaton ege noe pars, unlke orgnal S- OSPF; only the total amount of traffc that a noe njects nto the network an the total amount of traffc t receves from the network s neee. The optmal routng problem was transforme nto a regular LP problem. The propose scheme reuces the network congeston rato compare to the classcal SPR scheme. proves comparable routng performance wth MPLS-TE an TPR, whch requre functonal upgraes an/or complcate confguratons. Ok Lab. Semnar, May 28, 2009 26