LectureLecture 12.12.

IntroductionIntroduction toto IPIP RoutingRouting

G.Bianchi, G.Neglia, V.Mancuso WhyWhy introduction?introduction?

ÎRouting: very complex issue Öneed in-depth study Öentire books on routing Îour scope: Ögive a flavour of basic routing structure and messaging Ögive an high-level overview of IP routing protocols

G.Bianchi, G.Neglia, V.Mancuso RoutingRouting

131.175.21.2 131.175.21.3 143.67.1.4 143.67.1.9

131.175.21.1

143.67.1.4 131.175.21.2 208.11.3.3 131.175.21.2 ????

Router: must be capable of 208.11.3.2 addressing every IP network

G.Bianchi, G.Neglia, V.Mancuso RoutingRouting TableTable

Destination network Next router ………… ………… 131.175.0.0 144.21.32.4 ………… …………

G.Bianchi, G.Neglia, V.Mancuso RoutingRouting tabletable

10000.32 20000.5 20000.6 30000.12 30000.7 40000.8 Net 40000 Net Net Net 10000 R1 20000 R2 30000 R3 Net 50000 R2 routing table Destination Network Next Hop ROUTING TABLE: Other nets 20000 Direct fwd Î Router NEEDS to know 30000 Direct fwd which direction to forward the datagram 10000 20000.5 Ö to let it reach the final 40000 30000.7 destination 50000 30000.7 Î But DOES NOT NEED to default 20000.5 know the detailed path! Ö It stores only the NEXT Three cases: 1) direct forwarding HOP router. 2) Indirect forwarding (explicit) 3) Indirect forwarding via default router (when available)

G.Bianchi, G.Neglia, V.Mancuso DefaultDefault routeroute ÎFrequent in small and medium size networks Ögenerally administrator sends to router of higher network hierarchy Öe.g. our 131.175.15.254 (tlc) router defaults to elet router, which defaults to polimi, which defaults to Cilea router ÎLarge networks (class B sized) should default only when strictly necessary Öto avoid traffic increase and suboptimal router ÎTOP LEVEL ROUTING DOMAINS Ömaintain routing information to most Internet sites, and do not use any default route Ö5 in 1993: NFSNET, CIX, NSI, SprintLink, EBONE

G.Bianchi, G.Neglia, V.Mancuso RoutingRouting operationoperation assume router with IP address X 1) extract destination IP (Y) from datagram 2) if Source Route Option, forward accordingly 3) if Y==X, deliver packet to specified protocol 4) decrease TTL; if TTL=0 throw away datagram and send ICMP “time expired” message 5) if X.and.Netmask==Y.and.Netmask, direct forwarding of datagram (use ARP) 6) extract next hop router from routing table, and forward packet to next router 7) If no next hop, forward to default router 8) if no default route, declare route error and send notification via ICMP

G.Bianchi, G.Neglia, V.Mancuso ICMPICMP hosthost andand NetworkNetwork unreachableunreachable errorserrors

ICMP type 3 errors, codes 0 (network) and 1 (host) ÎHost unreachable Önetwork found, but packet could not be delivered to host ÎNetwork unreachable Öroute error (network not found in routing table)

G.Bianchi, G.Neglia, V.Mancuso TypicalTypical redirectionredirection casecase Campus address host ???? No route info for this address: go to default router R1

Router R1 Router R2 (to the Internet) (to campus network)

Clearly, host should have used R2 immediately…

G.Bianchi, G.Neglia, V.Mancuso redirectredirect ÎR1 operation Ölooks up routing table, and determine that R2 is the proper path Öin the mean time, it realizes that packet comes from same interface on R2 network Öthis makes R1 understand that redirection is possible Öthus sends a ICMP redirect error message ÎHost: Öwhen receiving a redirect message, it updates its routing table Öbasically, host LEARNS from redirects (easier task for admin that does not need to correctly configure all hosts)!

G.Bianchi, G.Neglia, V.Mancuso ICMPICMP redirectredirect 07 815 31 Type (5) Code (0-3) Checksum Router IP address that should be used

IP Header + first 8 bytes IP of original datagram data

REDIRECT CODES 0 Redirect for network The only one used in practice 1 Redirect for host 2 Redirect for TOS and network 3 Redirect for TOS and host

- Only routers may use redirect (other routers are assumed to be informed by full-fledged routing protocol, and not by occasional redirects!! - redirect must be addressed to hosts (not routers) - network redirection hard to be used (without netmask info!)

G.Bianchi, G.Neglia, V.Mancuso HostHost routingrouting tabletable creationcreation ÎManual creation Îvia router solicitation ICMP message

ICMP router solicitation (type 10)

ICMP router advertisement (type 9)

Router solicitation: asks who are the routers connected Router advertisement: return router list and preference preference: when multiple routers are connected to the same network preference values configured by administrator

G.Bianchi, G.Neglia, V.Mancuso StaticStatic versusversus dynamicdynamic routingrouting ÎStatic routing Öbased on static routing table entries Æentered manually Æchanged via ICMP redirects ÎFine when Önetwork small Ösingle connection point to other networks Öno alternative paths toward destinations ÎNot fine when one of above conditions fails

G.Bianchi, G.Neglia, V.Mancuso DynamicDynamic (adaptive)(adaptive) routingrouting AllAll IPIP routingrouting protocolsprotocols areare dynamicdynamic ÎRouting table entries change in time, depending on Ölink failures Æwhen a link is down, you need to avoid it! Önetwork topology changes Æwhen a new backbone added, use it! ÖTraffic load and congestion Æwhen a link is less loaded, use it 2.failure 1.old path

3.new path

G.Bianchi, G.Neglia, V.Mancuso DynamicDynamic routingrouting ÎRequirement: ÖInformation exchange among routers is required, to dynamically update routing table Æextra load Æneed for inter-routing message formats ÎRisks Öoscillation Ætoo fast adaptation procedures Öinefficiency Ætoo slow adaptation to changed situation Öloops

G.Bianchi, G.Neglia, V.Mancuso AutonomousAutonomous SystemsSystems aa keykey conceptconcept forfor TheThe InternetInternet Î Internet organized as a collection of Autonomous Systems (ASs) AS 1 Î each AS normally administered by a single entity Î each AS selects its own routing protocol to AS 2 allow inter-router communication within the AS ÎInterior Gateway Protocol (IGP) ÖIntra-Domain routing protocol Öwithin an AS ÎExterior Gateway Protocol (EGP) ÖInter-Domain routing protocol AS 3 Öamong different ASs

G.Bianchi, G.Neglia, V.Mancuso BorderBorder routersrouters

ÎRouting within AS is AS 4 arbitrary chosen by AS administrator IGP Îbut there must be one or AS 1 more border routers in charge of communicating to IGP the external world its AS 2 internal routing IGP information (data collected EGP by the IGP used) IGP ÎBorder routers are the only entitled to exchange EGP information AS 3

G.Bianchi, G.Neglia, V.Mancuso IPIP routingrouting protocolsprotocols

ÎIGPs AS 1 ÖHELLO ÖRIP1 ÖRIP2 RIP ÖOSPF (1 & 2) AS 2 ÖIS-IS ÖIGRP, EIGRP BGP-4 proprietary OSPF (CISCO) BGP-4 Ö… RIP-2 ÎEGPs BGP-4 ÖEGP Æyes: Same name of entire protocol class! AS 3 ÖBGP-4

G.Bianchi, G.Neglia, V.Mancuso RIP Routing Information Protocol

and distance vector protocols in general

G.Bianchi, G.Neglia, V.Mancuso RoutingRouting InformationInformation ProtocolProtocol ÎMost widely used Öand most criticized… ÎOfficial specification: RFC 1058 (1988) Öbut used from several years before ÎUses UDP to exchange messages Öwell known UDP port = 520 IP datagram UDP datagram UDP IP header header RIP message

G.Bianchi, G.Neglia, V.Mancuso RIPRIP messagemessage 0 7 8 15 16 31 Command (1-6) Version (1) 0 Address family (2) 0 IP address 0 20 bytes 0 Metric Up to 24 more routes with same 20 bytes format Command: 1=request; 2=reply (3-6 obsolete or non documented) Address family: 2=IP addresses metric: distance of emitting router from the specified IP address in number of hops (valid from 1 to 15; 16=infinite)

G.Bianchi, G.Neglia, V.Mancuso MessageMessage sizesize

Ö8 UDP header Ö4 bytes RIP header Ö20 bytes x up to 25 entries Îtotal: maximum of 512 bytes UDP datagram

Î25 entries: too little to transfer an entire routing table Ömore than 1 UDP datagram generally needed

G.Bianchi, G.Neglia, V.Mancuso InitializationInitialization

ÎWhen routing daemon started, send special RIP request on every interface Öcommand = 1 (request) Öaddress family = 0 (instead of 2) Ömetric set to 16 (infinite) ÎThis asks for complete routing table from all connected routers Öallows to discover adjacent routers!

G.Bianchi, G.Neglia, V.Mancuso OperationOperation afterafter initializationinitialization ÎRequest: Öasks for response relative to specific IP addresses listed in the request message ÎResponse: Öreturn list of IP addresses with associated metric Öif router does not have a route to the specified destination, returns 16 ÎRegular update: Örouters send part (or all) of their table every 30s to adjacent routers Öa router deletes (set metric to 16) an entry from its routing table if not refreshed within 6 cycles (180s) Ædeletion after additional 60s to ensure propagation of entry invalidation Îtriggered update: Öupon change of metric for a route (transmits only entries changed)

G.Bianchi, G.Neglia, V.Mancuso RIPRIP 22

ÎDoes not change the protocol operation Îsimply adds information in the all 0s fields of the RIP message ÎIt is designed to maintain full compatibility with RIP routers Öal least if they don’t get confused from the non 0 entries

G.Bianchi, G.Neglia, V.Mancuso RIPRIP 22 messagemessage formatformat 0 7 8 15 16 31 Command (1-6) Version (2) routing domain Address family (2) Route tag IP address Subnet mask 20 bytes Next hop IP address Metric Up to 24 more routes with same 20 bytes format

Most important modification: subnet mask (allows use with VLSM and CIDR) Next hop address: specifies where packet should be sent when addressed to Ipaddr details in RFC 1388

G.Bianchi, G.Neglia, V.Mancuso RIPRIP logiclogic ÎDistance Vector routing protocol ÖBellman-Ford algorithm ÆMETRIC = distance ÆSTATE INFO = vector Îeach router maintains a table with: Öbest known distance (in hop count) to each destination Ölink to use to reach the destination Îfully distributed protocol Övector (table) updates via communication with neighbors

G.Bianchi, G.Neglia, V.Mancuso RIPRIP operationoperation exampleexample (1)(1)

b a c

A Link 1 B Link 2 C

Link 3 Link 4 Link 5

D Link 6 E All routers start from empty tables e d

Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc b 0 loc c 0 loc d 0 loc e 0 loc Direct forwarding

G.Bianchi, G.Neglia, V.Mancuso a RIPRIP operationoperation exampleexample (2)(2)

A Link 1 B Link 2 C

Link 3 Link 4 Link 5 All routers start from empty tables D Link 6 E Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc b 0 loc c 0 loc d 0 loc e 0 loc

Router A emits message (A,0) to adjacent routers (B,D), which update table as: Router B Router D dst hop lnk dst hop lnk b 0 loc d 0 loc a 1 1 a 1 3

G.Bianchi, G.Neglia, V.Mancuso RIPRIP operationoperation exampleexample (3)(3)

A Link 1 B Link 2 C New step: B propagates its updated routing table to Link 3 Link 4 Link 5 neighbohrs A, C, E D Link 6 E Router B dst hop lnk b 0 loc a 1 1

Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc b 0 loc c 0 loc d 0 loc e 0 loc b 1 1 a 1 1 b 1 2 a 1 3 b 1 4 a 2 2 a 2 4

G.Bianchi, G.Neglia, V.Mancuso RIPRIP operationoperation exampleexample (4)(4)

A Link 1 B Link 2 C Step 3: D propagates its routing table to A, E Link 3 Link 4 Link 5

D Link 6 E Router D dst hop lnk d 0 loc a 1 3

Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc b 0 loc c 0 loc d 0 loc e 0 loc b 1 1 a 1 1 b 1 2 a 1 3 b 1 4 d 1 3 a 2 2 a 2 4 Already updated! d 1 6

G.Bianchi, G.Neglia, V.Mancuso RIPRIP operationoperation exampleexample (5)(5)

A Link 1 B Link 2 C Step 4: A propagates its routing table to B,D Link 3 Link 4 Link 5 Router A D Link 6 E dst hop lnk a 0 loc b 1 1 d 1 3

Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc b 0 loc c 0 loc d 0 loc e 0 loc b 1 1 a 1 1 b 1 2 a 1 3 b 1 4 d 1 3 d 2 1 a 2 2 b 2 3 a 2 4 d 1 6 ...ETC ETC ETC...

G.Bianchi, G.Neglia, V.Mancuso RIPRIP operationoperation exampleexample final routing tables

A Link 1 B Link 2 C Step 5: C -> B, E Step 6: E-> B, C, D Link 3 Link 4 Link 5 Step 7: B-> A, C, E

D Link 6 E Link 6 under-utilized!!

Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc a 1 1 a 2 2 a 1 3 a 2 4 b 1 1 b 0 loc b 1 2 b 2 3 b 1 4 c 2 1 c 1 2 c 0 loc c 2 6 c 1 5 d 1 3 d 2 1 d 2 5 d 0 loc d 1 6 e 2 1 e 1 4 e 1 5 e 1 6 e 0 loc

G.Bianchi, G.Neglia, V.Mancuso ApparentApparent limitslimits ofof RIPRIP ÎHop count is a too simple metric! ÖBut Bellman-ford algorithm does not require to operate with hop count! Can be trivially extended to different distance metric: the core of the algorithm does not change! Æqueue length on considered hop Ætime delay in ms Æpacket loss ratio measured Æetc… ÎSlow convergence Örouters distant N hops need N steps to update their tables ÎLimited to small network sizes Öas infinite=16, nodes cannot be more than 15 hops far away Æbut just raise infinite to 32...

G.Bianchi, G.Neglia, V.Mancuso RealReal limitlimit ofof RIPRIP ““countcount toto infinityinfinity”” problemproblem ÎInsane transient reaction to node/link failures! ÖConvergence still remains, but very slow ÖLoops may occur while routing tables stabilize Öthe slower, the higher value infinite is chosen!! ÆValues higher than 16 are terrible ÖAn intrinsic and unavoidable drawback for all Distance Vector schemes

G.Bianchi, G.Neglia, V.Mancuso CountCount toto infinityinfinity exampleexample A Link 1 B Link 2 C Link 3 D Link 4 E

Router A Router B Router C Router D Router E dst hop lnk dst hop lnk dst hop lnk dst hop lnk dst hop lnk a 0 loc a 1 1 a 2 2 a 3 3 a 4 4

LINK 1 breaks A B Link 2 C Link 3 D Link 4 E B does not get Router B Router C Router D Router E refreshes from a. Then uses refreshes dst hop lnk dst hop lnk dst hop lnk dst hop lnk from C, that tell a=3! a 3 2 a 2 2 a 3 3 a 4 4 a 3 2 a 4 2 a 3 3 a 4 4 a 5 2 a 4 2 a 5 3 a 4 4 Next steps: a 5 2 a 6 2 a 5 3 a 6 4 a 7 2 a 6 2 a 7 3 a 6 4

G.Bianchi, G.Neglia, V.Mancuso TheThe countcount toto infinityinfinity problemproblem

ÎThe problem is that NO ROUTERS have a value more than 1 + the minimum of adjacent routers

Îsituation stabilizes only when count gets to infinity

Îit is more critical the higher infinity is set!

G.Bianchi, G.Neglia, V.Mancuso SplitSplit horizonhorizon ““solutionsolution””

A Link 1 B Link 2 C Link 3 D Link 4 E

ÎThe distance is NOT reported on the line from which information comes ÖC tells correct distance to D, but lies (says infinity) to B Îdiscovers link failure in 1 hop

G.Bianchi, G.Neglia, V.Mancuso SplitSplit horizonhorizon failurefailure

A B Line CD goes down…

1) because of split horizon rule, A and B tell C that dist(D)=inf 2) C concludes that D is unreachable C and reports this to A and B 3) but A knows from B that dist(D)=2, and sets its dist=3 4) similarly, B knows from A distance from D … etc until distance = infinite D

Regardless the hack used, there is always a network topology that makes the trick fail!

G.Bianchi, G.Neglia, V.Mancuso OSPF Open Shortest Path First

and Link state protocols in general

G.Bianchi, G.Neglia, V.Mancuso LinkLink statestate routingrouting protocolprotocol

ÎEach router must: Ödiscover its neighbors Ömeasure a “cost” of the line connecting to the neighbor Ægenerally delay, e.g. via ICMP echo Æbut may be link bandwidth, etc Öconstruct a packet containing the information about all the connected links Ösend the packet to all the other routers

G.Bianchi, G.Neglia, V.Mancuso FloodingFlooding approachapproach When a router receives a packets:

ÎChecks if the packet is new or a copy Öa packet is new either if Æif it is first addressed to the node Æor if it was already received before, but THIS packet contains updated information Îif old, destroys the packet; Æto avoid duplicates and unuseful network load Îif new, forwards the packets on all links except the one it came from, Îby evaluating each packets, the router dynamically reconstructs the network topology

G.Bianchi, G.Neglia, V.Mancuso ExampleExample (1)(1) Packets received: Network discovery 2 B 4 B C a 4 6 c 2 A f 6 F

2 Router A C 4 B C 3 b 2 6 1 d 3 A D e 1 E F 2 D 4 B C 3 c 3 6 1 f 7 A D E F 7

G.Bianchi, G.Neglia, V.Mancuso ExampleExample (2)(2) Packets received: Network discovery

E 2 a 5 4 B C 3 6 1 c 1 A D f 8 5 E F 7 8 F Router A b 6 d 7 Confirm knowledge e 8

G.Bianchi, G.Neglia, V.Mancuso DiverseDiverse costcost perper linklink

ÎActually, the realistic case is a diverse cost per link Öin case of delay, queuing on one link is different from queuing on the reverse link!

No problem. The algorithm still works perfectly: every link will be represented twice: one packet per each direction!

G.Bianchi, G.Neglia, V.Mancuso ShortestShortest PathPath ComputationComputation

ÎOnce discovered topology and link cost, shortest path trivially computed ÖDijkstra algorithm

ÎAnd routing tables built accordingly

G.Bianchi, G.Neglia, V.Mancuso AdvantagesAdvantages ofof linklink statestate protocolsprotocols

ÎMuch faster convergence ÎDo not need periodical update Öbut event update: packet transmission is triggered by link state changes ÎNo count to infinity ÎDo not need to transmit routing tables Öonly state of local links (but flooded)

G.Bianchi, G.Neglia, V.Mancuso RealReal--lifelife protocolsprotocols ÎMuch more detailed ÎNeed to cope with all possible types of failures Ölink crashes and loss of flooding packets Æuse sequence number, ages Örouter that “forget” to signal link state ÖInconsistent packet reports Ædue to route changes while network picture being worked out ÎLink state Internet protocols: ÖOSPF, OSPF-2, IS-IS (Intermediate System - Intermediate System)

G.Bianchi, G.Neglia, V.Mancuso OSPFOSPF featuresfeatures ÎOSPFv2: RFC 1247 (1991) ÎUses IP packets directly! ÖUses own value in the protocol field of the IP header to allow demultiplexing at node ÎCan compute multiple routing tables for different TOS Ömin delay, max thr, max reliability, min cost Öno problem: based on cost associated to link! ÎSupport subnets (uses netmasks) ÎAllows authentication (via cleartext passwd transmission) ÎWhen equal cost links found, OSPF uses load balancing

G.Bianchi, G.Neglia, V.Mancuso HowHow OSPFOSPF WorksWorks

ÎLink failure detection ÖNot receiving HELLO message for long time ÆDefault, 40 seconds or 4 HELLO Intervals ÎIf neighboring routers discover each other for the first time ÖExchange their link-state databases ÎSynchronizing two neighbor’s link- state databases ÖDefault refresh information every 30 minutes

G.Bianchi, G.Neglia, V.Mancuso HowHow OSPFOSPF WorksWorks

ÎThe direct connected routers detect state change of the link ÖTrigger the Link State Update to neighbors ÖCompute the shortest path ÎOther routers flood the updates to whole network ÖUse sequence number to detect redundant updates ÖConfirm the updates (Link State Acknowledge) ÖCompute the shortest path

G.Bianchi, G.Neglia, V.Mancuso OSPFOSPF ------MessageMessage TypesTypes

ÎHELLO: Type 1 ÖIdentify neighbors ÖElect a designated route for a multi-access network ÖTo find out about an existing designated router Ö“I am alive” signal

G.Bianchi, G.Neglia, V.Mancuso OSPFOSPF ------MessageMessage TypeType

ÎDatabase Description: Type 2 ÖExchange information during initialization ÆSo that a router can find out what data is missing from its topology database ÎLink State Request: Type 3 ÖAsk for data that a router has discovered is missing from its topology databases or to replace data that is out of date

G.Bianchi, G.Neglia, V.Mancuso OSPFOSPF ------MessageMessage TypeType

ÎLink State Update: Type 4 ÖUsed to reply to a link state request and also to dynamically report changes in network topology ÎLink State ACK: Type 5 ÖUsed to confirm receipt of a link state update ÖSender retransmit until an update is ACKed

G.Bianchi, G.Neglia, V.Mancuso SomeSome remarks:remarks: MultipleMultiple shortestshortest pathpath Single Shortest path link cost path cost = 6 2 1 4 1 1 2 3 1 3

Multiple Shortest paths path cost = 6 2 1 4 1 1 2 3 1 3

G.Bianchi, G.Neglia, V.Mancuso SomeSome remarks:remarks: LoadLoad BalancingBalancing drawbacksdrawbacks

4 2 equal 5 3 4 1 2 50% cost paths 5 3 1 50% 5 4 3 2 1 out-of-order delivery

G.Bianchi, G.Neglia, V.Mancuso SomeSome remarks:remarks: risksrisks ofof traffictraffic dependentdependent costscosts

ÎRoute oscillation

1 A A A A 1+e 2+e 0 0 2+e 2+e 0 D B 0 0 D 1+e 1 B D 0 0 B D 1+e 1 B 0 e 0 0 1 1+e 0 C C C C 1 1 e …recompute …recompute … recompute initially

G.Bianchi, G.Neglia, V.Mancuso Exterior Gateway routing Protocols (EGPs)

G.Bianchi, G.Neglia, V.Mancuso ExteriorExterior GatewayGateway ProtocolsProtocols ÎHave several “non technical” problems Öpolicies: avoid routes which may be strategically critical Æe.g. allowing IP packets to transit over corporate links Æe.g. avoiding crossing critical states (midwest, etc) Öpolicies: manually configured Îefficiency is secondary Ömost concern is reachability of networks Öa further efficiency problem: different ASs may use different metrics that cannot be compared… ÎFirst protocol used: EGP (!) Öworked only for star topologies!! ÎNow replaced by BGP (v4)

G.Bianchi, G.Neglia, V.Mancuso BGPBGP viewview Routers connected only when there is a network interconnecting them

France B www.madrid.es

Italy switzerland D C Germany A

How to compare path costs? A,C may use RIP (hops), D may use OSPF delay, B may use IS-IS

G.Bianchi, G.Neglia, V.Mancuso NetworkNetwork classificationclassification byby BGPBGP

ÎStub Networks Öhave only one BGP router Öcannot be used for transit ÎMulticonnected Networks Öhave more BGP routers ÖMight be used for transit, but refuse ÎTransit Networks Öbackbones willing to handle packets of third parties (eventually with restrictions)

G.Bianchi, G.Neglia, V.Mancuso InterInter BGPBGP RouterRouter connectionsconnections ÎVia TCP Öto provide reliable information exchange Öand hide all the features of the underlying network(s) ÎKeepalive message exchange Öperiodically (about 30s), to check if peer router is up ΓDistance” Vector approach, but exchange FULL path information Önot only next hop information Önot vulnerable to count to infinity

Refer to RFC 1654 (BGP-4) for further information

G.Bianchi, G.Neglia, V.Mancuso