A Backward-Compatible Inter-domain Multipath Routing Framework

1y ago
23 Views
3 Downloads
597.50 KB
6 Pages
Last View : 15d ago
Last Download : 4m ago
Upload by : Genevieve Webb
Transcription

This paper was presented as part of the High-Speed Networks 2011 (HSN 2011) Workshop at IEEE INFOCOM 2011A Backward-Compatible Inter-domain MultipathRouting FrameworkXiaomin Chen, Mohit Chamania, Admela JukanTechnische Universität Carolo-Wilhelmina zu BraunschweigEmail: {chen, chamania, jukan }@ida.ing.tu-bs.deAbstract—We present a framework to facilitate the interdomain multipath routing for carrier networks that are basedon circuit switching technologies, without significant changes tothe existing inter-domain protocols. We first introduce a simplemechanism, namely, Single Routing Plane Multipath (SRPM)which can be easily implemented in the existing PCE-based interdomain architectures. The main idea behind the SRPM methodis to represent multiple link disjoint paths as a single virtualedge in the PCE and thus use the existing single path routingmechanisms to enable multipath routing. We further presenta more generic mechanism, referred to as Multiple RoutingPlane Multipath (MRPM) method, with the goal to facilitate theprovisioning of multipath connections over dynamically selecteddomain chains. In this method, we propose to virtualize thenetwork into multiple slices and represent as multiple routingtopologies by the PCEs. In this way, multiple routing planes areconstructed to facilitate inter-domain routing. For both methods,we consider the buffer size as constraint for multipath routingdue to the resulting differential delay. The results show thatthe proposed methods can significantly improve the blockingperformance and are backward compatible, while only slightlyimpacting the intra-domain traffic.I. I NTRODUCTIONIn the past few years, the carriers have witnessed the evergrowing commercial and scientific applications demandingQuality-of-Service (QoS) guaranteed paths with enormousbandwidth requirements. For example, image processing inastronomy requires transmission of 100 GBs for a ten squaredegree sky image [1], and very high definition visualization [2]requires bandwidths in excess of 10s of Gbps. Given that thenumber of high-bandwidth services is only going to increasein the future, the current paradigm of dedicating single path toprovision these services may not be feasible in the future, evenin the optical WDM networks. Whereas it is not rare to see thatconnections are required to traverse multiple domains, interdomain provisioning frameworks in carrier networks havenot yet considered the possibility that one service maybeprovisioned over multiple paths across multiple domains.A large number of algorithms, and also commercial standards such as VCAT in SONET/SDH consider the use ofmultipath routing in carrier networks. However, multipath provisioning in multi-domain networks carry significantly largerchallenges comparing to the single-domain scenario. First, tofacilitate inter-domain multipath provisioning, current multipath approaches would require extensions or modifications tothe existing protocols, such as BGP-TE [3]. Since operatorsare usually reluctant to migrate to new routing protocols, itis unlikely that a multipath solution based on a new inter-domain routing protocol would be adopted on a large scale.Although few vendors have developed the switches that canenable transmission of individual flows over Equal Cost Multipaths (ECMP) [4] for the purpose of the inter-domain loadbalancing, the requirement for implementing the ECMP onspecific hardwares furthermore limits the deployment of theinter-domain multipath routing on a larger scale. Second, theuse of multipath routing leads to packet re-ordering at thereceiver. While some applications such as bulk data transfermay not require packet re-sequencing, applications such as thereal-time streaming have to restore the order of packets, whichconsequently requires buffering. In high bandwidth networks,buffering has to be properly dimensioned as it can amount toa significant size for Gb/s scale traffic flows.In this paper, we attempt to work around these challengesand present a backward compatible framework to facilitatethe inter-domain multipath routing in the carrier networks. Wepropose to use single path routing on individual virtual routingplanes to enable multipath routing while avoiding modificationof the widely deployed BGP protocol. Our framework isbased on the renowned Path Computation Element (PCE) [5]architecture, where dedicated PCE servers in each domaincooperatively compute end-to-end multi-domain paths basedon the abstracted information advertised by each domain. Wefirst present a simple mechanism, referred to as Single RoutingPlane Multipath (SRPM), in which the domain information isabstracted to a single virtual topology stored in the TrafficEngineering Database (TED) of the domain PCE. The mainidea behind this mechanism is to utilize the inter-domain singlepath routing protocol for multipath routing and thus induceminimal changes to the network system. To make use of thisidea, we represent the multiple link disjoint paths betweenthe border nodes to be a single virtual edge in the virtualtopology while a single domain chain is obtained by runningBGP between PCEs. Each domain decides the transit segmentsin its virtual topology for the connection while the end-to-endpath is computed by extending the Backward Recursive PathComputation (BRPC) algorithm [6] on the virtual topologies.While the SRPM is a simple step forward the inter-domainpath computation schemes based on PCE, it can not providepaths from alternative domain chains which is important forload balancing in the multi-domain networks. To this end, wepresent a mechanism, referred to as Multiple Routing PlaneMultipath (MRPM), to enable multipath routing over multipledomain chains. PCEs virtualize its domain to multiple virtualslices which are represented as multiple virtual topologies.133

Multiple virtual routing planes are composed using a singlevirtual slice from each domain, on which a BGP instance is runto facilitate the inter-domain routing. The connections can beserved by utilizing the paths on a single virtual routing plane,or by utilizing multiple paths, with different paths computedon different virtual routing planes. To facilitate multipathprovisioning in both schemes, we consider the traffic splittingof one individual flow over these computed paths with allsynchronization operations e.g. re-sequencing taking place atthe connection source/destination. Both methods proposed arebackwards compatible with the PCE architecture as well asthe inter-domain routing protocols and our results show thatthey can significantly improve the blocking performance in amulti-domain provisioning scenario.The rest of the paper is organized as follows. In Section II,we briefly present the related work and our contribution. Section III presents the proposed inter-domain multipath routingmechanisms. Section IV presents the performance evaluationand the conclusions are presented in section V.II. R ELATED W ORK AND O UR C ONTRIBUTIONMechanisms proposed to enable inter-domain multipathrouting in the Internet have primarily focused on achievingflexible packet forwarding over diverse Autonomous System(AS) level paths towards the destination domain. Xu et. al. [7]presented a mechanism where domains can negotiate with eachother in order to forward packets to an alternative down-streamdomain. Kaur et.al. analyzed the the challenges associatedwith inter-domain multipath routing and presented a newmechanisms for the same. All proposed mechanisms to daterequired extensions or modifications of the existing interdomain routing protocol.Inter-domain multipath routing in connection-orientedcarrier-grade networks is a new topic and previous workhas primarily addressed single domain scenarios. In our pastwork [8], we have explored the usage of multipath routingwith multi-domain reach in carrier-grade Ethernet. In orderto facilitate the inter-domain multipath routing, we proposeda special virtual topology design mechanism, in which theshared segments have to be identified and represented toresolve the bandwidth reservation conflicts. However, thespecialized virtual topology design may not be flexible inhighly dynamic networks.In this work, our goal is to propose a backward compatible approach to facilitate the inter-domain multipath routing.Therefore, we propose and analyze an approach which doesnot require significant changes in the existing inter-domainrouting protocols, and is easy to implement, such as SRPM.Our approach is based on a standardized single path PCEbased inter-domain routing architecture [5]. Furthermore, wepropose to utilize multiple routing planes, by means of networkvirtualization, which is novel. In addition, our mechanism,such as MRPM, can use alternative domain chains, in contrastto the current approaches where the domain chain is predefined in the inter-domain path computation. It should benoted that multipath solutions in our framework are onlyprovisioned to the high bandwidth requests that single pathcan not serve, while directing the multiple individual flowswith same source and destination over multiple paths for loadbalancing such as the applications using ECMP in MPLSnetworks is not considered.III. M ECHANISMS TO E NABLE I NTER - DOMAINM ULTIPATH ROUTINGA. Preliminaries and AssumptionsThe multi-domain network, denoted as G(V, E), is composed of M inter-connected domains and each domain isdenoted as Gm (V m , E m ), m 1, 2, ., M , where onlyborder nodes and the connectivity between border nodes areincluded, i.e., vim V m are the set of border nodes of themdomain Gm and emis the virtual edge between nodeij Emmvi and vj . The available capacity and delay of emij is given ph,ijijwe also define the set of border nodes which are connectedby inter-domain links, i.e., the set BN (Gn , Gm ) includes theborder nodes of domain Gn which are connected to bordernodes in domain Gm . We compute up to K distinct pathsbetween a pair of border nodes vim and vjm in domain Gmmmmrepresented as P(emij ) {pij 1 , pij 2 , .pij K }. The value ofK is decided by the domain and restricted by the maximumavailable link-disjoint paths between the nodes. The availablembandwidth of path pmij k is denoted as bij k and the delaymis denoted as dij k . The connection request is denoted asr(s, d, Br , Dr ), where s, d are the source and destinationnodes and Br , Dr are the required bandwidth and end-to-enddelay constraint respectively. The available buffer size at nodevim V m is denoted as Mim .The differential delay caused by the usage of multiple pathsfor provisioning a connection is defined by the difference inthe delay between the paths used to provision a request. Thedelay of an end-to-end path P is given by dP , the differentialdelay between two paths P and P 0 is defined as:dd(P, P 0 ) dP dP 0 (1)The re-sequencing buffer size requirement is decided bythe differential delay between the path with highest delay andother paths and the traffic routed on each path [9]. Assume theset of paths for a connection request r is the path set P(r),with P̃ as the path with highest delay, and traffic on path Pis denoted by tP , the re-sequencing buffer size required iscalculated as:XMr tP · (dP̃ dP )(2)P PThe two mechanisms proposed in this paper, i.e., SRPMand MRPM, both rely on the PCEs to abstract and exchangedomain information for the inter-domain routing. The virtuallinks represent the connectivity information between bordernodes and include information about TE parameters such ascapacity, delay, etc. We now present both methods.B. Single Routing Plane Multipath (SRPM)SRPM is a simple extension to today’s PCE-based interdomain path computation schemes. The domain chain is134

obtained by the BGP running between PCEs while the end-toend path computation is done based on the BRPC. However,in order to facilitate the multipath routing, the multiple linkdisjoint paths are first calculated between border nodes andrepresented as a single edge in the virtual topology of thedomain. The capacity of the virtual link given by the sumof the available bandwidth of thebetween the pairP pathsmof the border nodes, i.e., bm, bijk ij k and the delay ofvirtual link emissetasthemaximumdelay in the P(emijij ),mmi.e., dij max{dij k }. The steps involved in computing thepath/paths for a given request is shown in Alg. 1.Fig. 1. The multiple virtual topology representation of a two-domain networkAlgorithm 1: Single Routing Plane Multipath MechanismInput: r(s, d, Br , Dr )Output: Provision connection for r(s, d, Br , Dr )Step 1: Compute the domain chain from s Gs to d Gd asG(r) Gs , G1 , ., GM , GdStep 2: Initiate PCE Signaling using the modified BRPCalgorithm (Alg. 4)Step 3: If path computation in PCE is successful, initiateinter-domain path setupAlg. 3. The set of transit paths is ordered by the path delay,and the algorithm iteratively chooses a path from this orderedlist as the path with the maximum delay, and then goes onto compute the required path set while ensuring that the resequencing buffer availability constraints are met. Note that aseach individual domain can decide upon using a single pathor a multipath solution to provision a request, i.e., it is thedomains’ responsibility to compensate for differential delayand therefore the use of a multipath solution is constrained bythe availability of the buffer at the domain border nodes.The complexity of SRPM is decided by the number of linkdisjoint paths per virtual link and the number of border nodespair. Assume each domain has M pair of border nodes andK paths between each pair of border nodes, the complexityof SRPM in a network with N domains is O(N · M 2 · K 2 ),while the complexity of using BRPC in the same network isO(N ·M 2 ·V ), where V is the set of all nodes in the networks.Algorithm 2: Multiple Routing Plane Multipath MechanismInput: r(s, d, Br , Dr )Output: Provision connection for r(s, d, Br , Dr )computedP athArray[ K ] null;for each virtual routing Plane doStep 1: Compute the domain Chain from s Gs tod Gd as G(r) Gs , G1 , ., GM , GdStep 2: Initiate PCE Signaling using the standard BRPCalgorithm [6](Compute Max Available Bandwidth path)if path computation in PCE is successful theninitiate inter-domain path setup;endStore computed path information incomputedP athArray[k];endif no Single Path Found thenUse algorithm described in computeTransit (Alg. 3) tocompute min-delay min-buffer requirement multi-path;if multi path found theninitiate inter-domain path setup over the multiplecomputed paths;endendC. Multiple Routing Plane Multipath (MRPM)In this mechanism, we assume the use of conventional interdomain routing protocols to compute the domain chain, whichin this case use the shortest path algorithm on the virtualtopology. Once the domain chain is computed, the modifiedBRPC proposed in Alg. 4 is initiated by the PCE in thedestination domain. In the modified BRPC, instead of usingonly a single path between a pair of ingress and egress routersinside a domain, the PCE is allowed to use multiple pathsbetween the given pair to facilitate the request.Each domain attempts to compute a set of path segmentswith the shortest delay which simultaneously minimize thedifferential delay, utilizing the computeT ransit function inThe MRPM mechanism composes the multipath solutionsby running single path routing on different virtual routingplanes. Each virtual routing plane is in turn generated usingdifferent virtual topologies advertised by all domains. InMRPM, each domain advertises multiple virtual topologiesin order to construct multiple routing planes. To facilitate theinter-domain multipath routing, each domain calculates K linkdisjoint paths between each pair of border nodes and picksone from each to compose a virtual topology. The variouscombinations of the paths between the border nodes lead to themultiple virtual topologies of the domain. A simple illustrativeexample is shown in Fig. 1. Each domain advertises two virtualtopologies to construct two virtual planes. The Alg. 2 providesa multipath solution to a connection request. It should be notedthat connection request will only be served with multipathrouting when it can not be served by single path routing.In this mechanism, domain chain is obtained by the BGPon each virtual routing plane and the standard BRPC isinitiated by the PCE in the destination domain on eachrouting plane to compute an end-to-end path with maximumavailable bandwidth. The MRPM first checks if there existsan single path solution for the connection request before themultipath routing. The paths used for the multipath routingare decided by the PCE in the source domain by selecting135

from the computed paths from each routing plane, using thecomputeT ransit function in Alg. 3. The available buffer sizeand bandwidth requirement are considered as constraints.It should be noted that the MRPM is not exclusivelydesigned for inter-domain multipath routing. Instead, it alsoprovides multiple options for the inter-domain single path routing. The final solution for the connection demand should havethe total bandwidth no less than the bandwidth requirement,and the available buffer is sufficient to support resequencingof the flow. Note that the total number of routing planes usedin the multi-domain scenario depends on the number of virtualtopologies advertised by individual domains. For example, ina two-domain network, if we assume that Ki and Kj virtualtopologies are advertised by the domains, then the number ofvirtual routing planes that can be constructed is max{Ki , Kj }.Assume that each domain has M border nodes, the complexityof the MRPM mechanism in a network with N domains andK virtual routing planes is O(K · N · M 2 ).Algorithm 3: Function computeTransit: compute the transit path with minimal buffer requirementsFunction computeTransit(ingress vi , egress vj )Output: Optimal Intra-domain Multi-Path Segments(P ath, Bw) from vi to vjSort the K paths in P(emij ) in the increasing order of the pathdelay;for k 1 to K doSegment s null; Bw bmij ;m ks.add(Path pmij k , Bw bij k );The required buffer Mr 0;for l k 1 to 1 doB w bmij l ;mmM r bmij l · [dij k dij l ]if Mr {Mir Mdr } thenbreak;endif Bw Br thenreqBw Br s.bandwidth;s.add(Path pmij l , Bw reqBw);return s;endms.add(Path pmij l , Bw bij l );endendD. Common FunctionsIn order to compute the multi-path transit segment insidea domain, we make a minor modification to the BRPCalgorithm, which only checks the distinct paths between bordernodes instead of searching all paths towards the upstreamingborder nodes. As shown in Alg. 4, the modified BRPC initiatesthe virtual shortest path tree (VSPT) in the destination domain(Step 1), and then in every subsequent transit domain (Step2b), it uses the function computeTransit (Alg. 3) to computemultipath transit segments. The computeTransit function orders the path in increasing order of delay, and then sequentiallychooses a path segment to be set as the highest delay segment.The algorithm then iterates over all paths with delay lessthan or equal to the delay of the selected segment to see ifsufficient bandwidth is available and that the re-sequencingbuffer constraints are not violated.IV. P ERFORMANCE E VALUATIONIn this section, we evaluate the performance of the SRPMand MRPM on a multi-domain network composed by fourdomains as shown in Fig. 2. The link delay of all topologiesis proportional to the geographic distance of the physicalnetwork. The representative topologies used for the domainsare scaled to comparable sizes, i.e., the link delay of theJanos U S Ca and Cost266 topologies are scaled downto be in the same order as the delays in the France andGermany50 topologies. The delay of the inter-domain linksis assigned proportionally to the physical distances depictedin Fig. 2. The border nodes of each domain are marked inblack and the capacity of internal links in each domain isassumed to be 40Gb and the capacity of the inter-domain linksis assumed to be 100Gb. The bandwidth required by all intradomain connections are assumed to be 1Gb and the bandwidthrequired by the inter-domain connections randomly variesfrom 5Gb to 10Gb. The connection demands are assumed toarrive in a Poisson process and are uniformly distributed inthe networks. The buffer constraint is set to 100M B whenmultipath is used for the connections. All the results shown inthe following are based on the scenario that the inter-domaintraffic load is constantly set to be 30Erlang, while the intradomain traffic load changes dynamically. It is based on thefact that the inter-domain connections usually less dynamiccompared to the intra-domain traffic. Three link-disjoint pathsare computed between each pair of border nodes for the virtualtopology construction. In order to achieve a fair comparison, itis assumed that the traditional inter-domain single path routingcan use all these paths between the border nodes too, whileonly one path can be used at one time.In the results that follow, we first evaluate the performanceof the SRPM and MRPM on inter-domain traffic, and thenstudy their impact on the intra-domain traffic. Finally, we showthe cost of using the SRPM and MRPM with regard to theinter-domain signaling load. 95% confidence interval is usedin all the results for the blocking probability. It should be notedthat the network load shown in all the results that follows isthe intra-domain traffic load which does not include networkload caused by the transit traffic.A. Impact on Inter-domain trafficThe impact of SRPM and MRPM on the inter-domain trafficis shown in Fig. 3. It can be seen that both SRPM and MRPMcan reduce the blocking of the inter-domain connections, ascompared to the inter-domain single path routing, especiallywhen the networks are heavily loaded. As both SRPM andsingle path routing use the same set of paths when computingan inter-domain path, the advantage of using multipath routingin SRPM only becomes clear when network load is high.However, significant reduction is obtained by the MRPM,about 3% at 130Erlang, due to the fact that the MRPM allowsfor alternative domain chains in the inter-domain routing.136

Algorithm 4: Modified BRPC Algorithm to support Multipath computation Inside Single DomainInput: Domain Chain G(r) G1 , G2 , ., GM , r(s, d, Br , Dr )Output: Optimal Inter-domain Path from s to d//Step 1: Initialize BRPC Tree in Destination DomainTree T(r) null;for all BNs vx BN (GM 1 , GM ) doCompute Min Delay Single Path from d to vxif Path found thenAdd path to T (r)endend//Step 2: Recursively Extend the BRPC Tree through allintermediate Domainsfor index M 1 to 1 do//Step 2 (a): Extend tree to ingress nodes of current domainfor all BNs vx BN (Gindex 1 , Gindex ) doUse inter-domain links between domains Gindex 1 andGindex to extend T (r) to vx ;end//Step 2 (b): Compute Intra-domain Path Segmentsfor all BNs vx BN (Gindex 1 , Gindex ) dobestPath null;for all leaf nodes vy T (r) doObtain multi-path intra-domain segmentinformation using function computeTransit(vx , vy )Alg. 3if obtained segment better than bestPath thenbestPath obtainedSegment;endendif bestPath! null thenInsert bestPath in T(r)endendend//Step 3: Choose optimal path in Source Domainfor all leaf nodes vx T (r) doCompute Min Delay Single Path from s to vxif Path found thenAdd path to T (r) by checking if better than anexisting path to s in T (r)endendif path found to s T (r) thenreturn path;endFig. 2.The multi-domain network used for simulation6.5Single PathSRPMMRPMInter domain Traffic Blocking Probability (%)65.554.543.532.521.5Fig. 3.5060708090100110Intra domain Traffic Load (Erlang)120130The average blocking probability of inter-domain connectionssome domains, while leading to the decrease in some domains.For instance about 6% increase in the intra-domain connectionblocking is observed in the Cost266 network, and about 0.7%reduction is observed in Janos network.B. Impact on Intra-domain TrafficC. Signaling CostIn order to show the impact of enabling multipath routing onthe inter-domain connection requests, we study the blockingprobability of the intra-domain traffic in all domains. The intradomain blocking probability in case of inter-domain singlepath routing is also studied as the benchmark. We show tworepresentative figures here. It can be seen in Fig. 4 and Fig. 5,the SRPM has a slightly negative impact on the intra-domaintraffic due to the fact that it can accept more inter-domaintraffic. However, the MRPM obtains a slight reduction inperformance for Janos U S Ca network as shown in Fig. 4.It is due to the fact that the MRPM can utilize alternativedomain chains and balance the network load over the domains.The inter-domain traffic can be routed over the preferabledomain chain more often which increases the network load inFinally, we study the cost of enabling inter-domain multipath routing with regard to the signaling cost. Table I showsthe number of inter-domain updates circulated in the networkobserved per second. In our simulation, an update is triggeredwhen the available capacity on any of the advertised virtuallinks is affected. It is clear here that the inter-domain singlepath routing has the lowest signaling frequency, which can beattributed to the fact that this technique exhibits higher interdomain blocking, and therefore in cases when an inter-domainconnection is blocked, an advertisement is not triggered. Thehighest update frequency is observed by the SRPM due toits lower inter-domain blocking. In case of MRPM, we showthe number of updates triggered for each routing planesinitiated in the simulation, here K 3, and see that while137

Intra domain Traffic Blocking Probability in Janos (%)5Single pathSRPMMRPM4.543.532.521.515060708090100110Intra domain Traffic Load (Erlang)120130Fig. 4.The average blocking probability of the intra-domain traffic inJanos U S Ca networkIntra domain Traffic Blocking Probability in Cost266 (%)12Single pathSRPMMRPM10864205060708090100110Intra domain Traffic Load (Erlang)120130Fig. 5.The average blocking probability of the intra-domain traffic inCost266 networkfor each individual routing plane, the update frequency isnot much higher than the inter-domain single path routing.Although the update frequency is comparably rather high whencombined together, the results indicate that the total interdomain update rate increases only linearly with the numberof virtual topologies K.TABLE II NTER - DOMAIN T OPOLOGY U PDATES PER SECONDSingle 5.6537165.714527D. Discussion: Impact of TE on inter-domain linksIn practice, inter-domain links usually are deployed withsignificant high bandwidth comparing with the internal linksof the domains. Therefore, slicing the domain into multiplevirtual routing plane may not lead to the congestion onthe inter-domain links. However, the policies applied for theTraffic Engineering (TE) on the inter-domain links can leadto the different domain chains, which is not considered in thispaper but as the future work of this study.V. C ONCLUSIONIn this paper, we presented a backward compatible framework to facilitate the deployment of multipath routing in theinter-domain service provisioning without significant changesto the existing inter-domain routing protocols. Our frameworkis based on the existing PCE-based inter-domain service provisioning architecture, which compute end-to-end path based onthe abstract domain information advertised by each domain.We first presented a simple approach (SRPM) which representsmultiple transit paths between border nodes as a single virtuallink in the virtual topology to facilitate the utilization of thetraditional inter-domain single path computation. We furtherpresented an approach which can provide multiple paths overmultiple domain chains, which requires the PCEs to virtualizeits domains to multiple virtual topologies, where we considered the most challenging multipath routing scenario wheretraffic of an individual flow is split into multiple paths andconsiders constrained on the available re-sequencing buffer.The results showed that our framework can enable interdomain multipath routing with superior blocking performanceand acceptable increase in signaling with respect to single pathrouting. Both mechanisms can reduce the inter-domain connection blocking, especially, the MRPM method. While bothmechanisms have slight negative impact on the connectionrequest inside domains, the MRPM can balance inter-domaintraffic load among multiple domains which leads to the lessblocking of the intra-domain connections in some domains. Interms of signaling overhead, the method MRPM needs furtheroptimizations, as it seems to exhibit a high signaling load,especially the number of routing planes is high.VI. ACKNOWLEDGMENTThis work has been partially supported by GEYSERS (FP7ICT-248657) project funded by the European Commissionthrough the 7th ICT Framework Program.R EFERENCES[1] E. Deelman et al., “Pegasus: A framework for mapping complex scientificworkflows onto distributed systems,” Scientific Programming Journal,vol. 13, no. 3, pp. 219–237, 2005.[2] D. Simeonidou et al., “Optical Network Services for Ultra High DefinitionDigital Media Distribution,” IEEE GOSP 2008, London, UK.[3] BGP Traffic Engineering Attribute. [Online]. Available: -te-attribute-04[4] BGP Multipath Load Sharing for Both eBGP and iBGP in an MPLSVPN. [Online]. Available: http://www.cisco.com[5] A. Farrel et al., “A Path Computation Element (PCE)-BasedArchitecture.” [Online]. Available: http://www.ietf.org/rfc/rfc4655.txt[6] A. Farrel et al., “A Backward-Recursive PCE-Based Computation(BRPC) Procedure to Compute Shortest Constrained Inter-DomainTraffic Engineering Label Switched Paths.” [Online]. Available:http://tools.ietf.org/html/rfc5441[7] W. Xu a

Inter-domain multipath routing in connection-oriented carrier-grade networks is a new topic and previous work has primarily addressed single domain scenarios. In our past work [8], we have explored the usage of multipath routing with multi-domain reach in carrier-grade Ethernet. In order to facilitate the inter-domain multipath routing, we proposed

Related Documents:

Basic 3 Forward outside and inside edges on a circle, forward crossovers, backward one-foot glide, backward snowplow stop, backward half swizzle pumps on a circle, moving forward to backward and backward to forward two-foot turn, beginning two-foot spin Basic 4 Basic forward outside and forward inside consecutive edges, backward edges on a circle,

Domain Cheat sheet Domain 1: Security and Risk Management Domain 2: Asset Security Domain 3: Security Architecture and Engineering Domain 4: Communication and Network Security Domain 5: Identity and Access Management (IAM) Domain 6: Security Assessment and Testing Domain 7: Security Operations Domain 8: Software Development Security About the exam:

an interconnected network architecture. There are inter-domain and intra-domain routing protocols. The inter-domain routing protocol has experienced increasingly frequent anomalies, such as IP prefix hijackings, route leaks, or impact from large-scale disruptive routing events. The intra-domain routing also suffers from various attacks

An Active Directory domain contains all the data for the domain which is stored in the domain database (NTDS.dit) on all Domain Controllers in the domain. Compromise of one Domain Controller and/or the AD database file compromises the domain. The Active Directory forest is the security boundary, not the domain.

30,000 Business Cards IBM PC Compatible 30,000 Games IBM PC Compatible 300 Arcade Games IBM PC Compatible 300 Arcade Games [Jewel Case] IBM PC Compatible 300 Great Games for Windows '98: Version 2.0 IBM PC Compatible 303 Game Collection IBM PC Compatible 333,000 Games IBM PC Compatible 333

Introduction 1.1 Introduction to Backward Stochastic Differential Equa-tions What is Backward Stochastic Differential Equations? The most classical form of backward stochastic differential equation (BSDE) is Y t Z T t f(s;Y s;Z s)ds Z T t Z sdW s (1.1.1) where F FW, the terminal condition is a Rd-valued FW T-measurable random variable .

9 CSI Domain 1 – Food and Nutrition 13 CSI Domain 2 – Shelter and Care 17 CSI Domain 3 – Protection 21 CSI Domain 4 – Health 25 CSI Domain 5 – Psychosocial 29 CSI Domain 6 – Education and Skills Training 33 Important Events – CSI

ISO 14001:2004 and ISO 9001:2000 15 Annex B (informative) Correspondence between OHSAS 18001, OHSAS . Standard vi List of tables Table A.1 – Correspondence between OHSAS 18001:2007, ISO 14001:2004 and ISO 9001:2000 15 Table B.1 – Correspondence between the clauses of the OHSAS documents and the clauses of the ILO-OSH Guidelines 20 Summary of pages This document comprises a front cover .