`
`Alvaro Guillen , Ramin Najmabadi Kia, Bernard Sales.
`
`Brussels Universities-Service Télématique & Communication
`Bld. du Triomphe CP 230 B-1050 Brussels Belgium
`
`Abstract
`
`With the emergence of multimedia applications,
`networks have to provide more and more bandwidth and
`should guarantee the quality of the communication
`between users. To address these issues, high speed
`architectures based on Virtual Circuit model such as
`
`ATM have been designed and successfully implemented.
`The current
`trends in telecommunication deployment
`indicate that very large Internets based on such model
`will be available in the nearfuture. This paper addresses
`the routing problem in this context and analyzes the
`existing solutions. We conclude that the network has to
`provide pre-computed and on-demand routes using a
`variation of the Distance Vector algorithm. Among other
`things, source routing and complete route recording are
`used to prevent loopformation.
`Keywords: Routing, Distance Vector, Link State, VC
`Routing, 905 based routing, Network Interconnection.
`
`of multimedia
`emergence
`the
`However, with
`applications, the network has to provide more and more
`bandwidth and should guarantee that the communication
`between ESs satisfies certain characteristics in terms of
`performance. To address
`these issues, high speed
`architectures based on Virtual Circuit (VC) model such
`as ATM have
`been
`designed
`and
`successfully
`implemented. The current trends in telecommunication
`deployment indicate that a very large Internet based on
`such model will be available in the near future. In this
`
`environment, both public and private networks have to
`collaborate to support the communication between end
`users. This raises the question on how systems can
`calculate and maintain routes in a VC context having in
`mind that
`these routes have to match the user
`requirements in terms of Quality of Service (Q08).
`
`2.
`
`Environments
`
`1.
`
`Introduction
`
`2.1. VC/QoS model
`
`aspect of computer
`Roofing is a fundamental
`communication. In particular, efficient and operational
`routing procedures are a pre-requisite for the global
`networking environment. The routing procedures may
`be characterized by aspects related to distributed versus
`centralized computation of network paths and by their
`adaptability to environment changes.
`For the datagram case,
`the Internet community has
`pioneered the routing protocol development. They have
`defined routing protocols for different environments.
`Examples of such protocols are [RIP]
`[ARP]
`[OSPF]
`[BOP]. This situation is continuously evolving and new
`protocols are currently designed and implemented
`[BGP4]
`[IDPR]. 0n the other hand, ISO has taken
`benefit of the valuable experiences of the Internet
`community and proposed a solution for the datagram
`routing based on three protocols: ES-IS [1809542], IS-
`IS [18010589] and IDRP [15010747].
`
`0-8186-3670-X/93 $03.00 © 1993 IEEE
`
`‘
`
`80
`
`The following terms are used in this paper:
`- End System (ES): a system which is attached to the
`VC-network and hosting the application process,
`- Switching node (SN): systems performing a packet
`switching based on the VC model,
`- System: an ES or a Switching node,
`- Source node: the switching node to which the Source
`E8 is attached.
`the Switching node to which the
`- Destination node:
`Destination BS is attached,
`- Node-id: a linguistic construction, built on a given set
`of symbols and identifying unambiguously a system
`within the VC network.
`- Set-up packet:
`the packet used by the network to
`establish a VC.
`
`The environment we are considering may be viewed
`as composed of a set of ESs (i.e. systems hosting
`different application processes) and a set of Switching
`
`Petitioner Apple Inc. - EX. 1005, p. 1
`
`Petitioner Apple Inc. - Ex. 1005, p. 1
`
`
`
`nodes arbitrarily interconnected by means of links. Each
`System is identified by an address which is unique in the
`context of the network. In this context, a set of QoS
`values are associated with each link interconnecting two
`Systems which give an indication of the quality of this
`link. Among other things, these values depend on the
`Switching nodes‘ status, the basic capacity of the links
`and the traffic currently supported by the links and by
`the Switching nodes.
`The switching technique performed by the Switching
`nodes is based on the point—to-point Virtual Circuit
`model (e.g. X25. ATM. ST-lI, ...). For the purpose of
`this paper, a VG is a communication path between two
`ESs consisting of a set of Switching nodes and links.
`We do not place any constraints on the protocols used
`in our environment, but only mention some (minimal)
`characteristics of them. We consider that the system has
`to be capable of establishing a VC consisting of a
`concatenation of Switching nodes and links in the
`network. The data communication requires that first a
`VG has to be set up. Once the VC set-up phase is
`completed, the data is transferred on this VC.
`A complete route joining two ESs can not be obtained
`only by considering the ES addresses. The reason for this
`is that, in the context of a VC. the ES can ask for some
`characteristics
`to
`be met
`by
`the VC. These
`characteristics are known as the Quality of Service
`(QOS) over the VC. These QOS values result from the
`end
`users
`requirements who
`expect
`that
`the
`communication
`between
`them
`have
`specific
`characteristics corresponding to the needs of
`the
`application that will be used on this VC. The QoS is
`negotiated during the VC set-up phase. As a result of this
`negotiation, a VC set-up request may be rejected either
`because the QoS required by the ESs can not be satisfied
`by the network or because of an action initiated by the
`destination end user.
`
`On a given VC. the QoS is not re-negotiated. Instead,
`when the QoS provided for a given VC no longer
`matches the values negotiated during the set-up.
`the
`network may clear the VC.
`For the purpose of this paper, we consider that the
`links are symmetric (i.e. at each moment the same 008
`is available for both directions of data transfer).
`However.
`the method presented here can be easily
`adapted to be used with asymmetric links.
`
`2.2. Routing topology
`
`In a large network consisting of several thousands of
`syStems, it is likely that these systems are administered
`and managed by different organizations
`(telecoru
`operators. etc.) each of them having their own policy in
`
`terms of routing algorithms, supported protocols for
`routing
`information
`exchange,
`resource
`allocation
`scheme and management, tarification, etc.
`With that respect, a large network environment is
`generally partitioned in different autonomous entities
`which have to cooperate in order
`to support
`the
`communication between users belonging to different
`communities. The concept of routing domains1 (or any
`other equivalent term) has been introduced to cover
`these requirements in several architectures. A Routing
`Domain (RD) is defined as a group of E83 and Switching
`nodes sharing the same routing algorithm,
`the same
`protocols for routing information exchange, etc. Such a
`domain can define its own policy with respect to other
`RDs (the type of routes available through it,
`traffic
`constraints,...). Routes between Source and Destination
`ESs belonging to different domains are composed of two
`parts:
`intra-domain part (the route supported within a
`given domain) and inter-domain part (the different RDs
`supporting the route). These different
`routes
`are
`combined to support the end-to-end route.
`The
`interaction between E88 and the network
`(Switching nodes) is referred to as ES-IS routing.
`For the purpose of this paper, only the Intra RD case
`will be discussed. The Inter RD routing in a VC/QoS
`environment is addressed in [NajSal]
`
`2.3. More about QoS
`
`In a distributed environment, each application has its
`own set of requirements concerning the "quality of the
`transmission". For instance, a file transfer might be
`characterized by the transfer of an high volume of
`information without any error. High throughput
`and
`error free transfer would typically be the most critical
`parameters. It is on a par with the fact that, for different
`data units. the transit delay between the users may vary
`within a certain range of values. A real-time HDTV
`(High Definition TeleVision) transmission requires also
`a high throughput but, on the other hand, it imposes a
`constant transit delay and tolerates that the error rate
`oscillates within a given range of values. As far as
`applications are concerned, we can consider that each
`application has
`its own well defined requirements
`concerning the underlying data transmission service.
`
`1This structure was first proposed by the Internet community and was
`
`adopted later by ISO as a framework for routing architectures [T119575]
`
`[ETGl4].
`
`81
`
`Petitioner Apple Inc. - EX. 1005, p. 2
`
`Petitioner Apple Inc. - Ex. 1005, p. 2
`
`
`
`The term Type of Service (T08) or Quality-of—Service
`(QoS) is used to refer to the parameters allowing to
`control the characteristics of the data transmission.
`
`The semantics of each QoS have to be unique within
`the network. This means that all the Switching nodes of
`the network have to agree on the same semantics for
`each QoS parameter and treat each of them based on
`their semantics in the same way. Even though the
`definition of these semantics is important,
`it
`is not
`directly related to the routing decisions. Instead,
`the
`values associated to QoS parameters and their variation
`have to be considered by the routing mechanism.
`Another important aspect
`is the way ESs specify
`values for QoS parameters. Some applications require
`strict values
`for QoS parameters, but
`some other
`parameters could be viewed as topology dependent. An
`example of this is the Transit Delay parameter (e.g.
`international vs. national connections). A first solution is
`to consider that the transit delay is specified by relative
`values such as "high - medium -low". The significance
`of these relative values is entirely determined by the
`network
`(this
`could
`be
`unacceptable
`for
`some
`applications). Another solution is to consider that,
`in
`general, an application should consult some kind of
`directory services (DNS, X500,
`...)
`for application
`address resolution. The purpose of such a directory is to
`store
`information which is
`relatively static. This
`information can include the values of QoS which are
`"statically" known as possible to join the requested
`destination from the source (e.g. information related to
`geography,...).
`In this way,
`the user will have an
`indication of what is feasible.
`
`2.4. More about VC set-up model
`
`In a VC context, the QoS of a communication results
`from a negotiation between the user and the network. In
`the existing (point to point) VC model, the connection is
`first established from the Source ES to the Destination
`
`ES. For this purpose a route consisting of links and
`Switching nodes is selected on the basis of the QoS
`values requested by the Source ES. During this process,
`the intervening nodes may participate to the routing
`decision and diminish the original QoS values. When the
`incoming call is initiated at the destination, a set of QoS
`values will be proposed. These values match both the
`Source ES requirements and the network capabilities
`(the Destination ES does not intervene in the selection of
`the route).
`For each parameter. the: destination user selects one
`value within the range received from the network,
`indicates it in the call accepted packet and sends it
`across the established connection (e.g. see the model
`
`described in [1508348] and [18010028]....). The major
`drawback of this method is that
`the connection is
`
`the Source ES
`established only on the basis of
`requirements and the network capabilities without
`considering those of the Destination ES.
`Another model proposes to slap the negotiation
`process in the following way: the Source ES proposes
`the minimum service it agrees to use, and the network
`tries to establish a communication satisfying at
`least
`these requirements [RFC1363]. If the Source BS is
`always able and willing to handle these QoS,
`this
`method works well. However, if the destination can not
`or will not handle the call with these QoS, it may refuse
`the call.
`
`A drawback of this method is the fact that specifying
`only the minimal requirements is not satisfying in some
`circumstances. For example, an application trying to get
`the best QoS within a range might have to try several
`times to establish a connection before succeeding.
`starting with the highest QoS values, which are rejected.
`and going down to the minimum acceptable. The
`rejection of the call with high QoS values might be a
`result of
`the Destination ES‘s
`incapability
`(or
`unwillingness) to accept these values. A possible way to
`take into account the Destination ES‘s requirements is to
`include these requirements in the refusal message. In this
`way,
`the Source ES can retry to establish a new
`connection with these new QoS requirements matching
`both Source and Destination ES requirements. The
`drawback of this method is that it implies an additional
`round-trip for the VC set-up phase.
`As a result of this, a new model for VC set-up is
`needed. In this model, the communication is established
`on the basis of the Source and the Destination ES
`requirements. To do so.
`the network calculates an
`appropriate route and sets up a VC over this route. In
`this context, the ideal scheme would be:
`1) the Source ES sends a message indicating its intention
`to communicate with a given Destination ES with
`certain characteristics in terms of Q08,
`it can
`2) the Destination ES selects QoS values that
`support and that matches the request from the Source
`ES,
`3) an appropriate route matching the QoS selected by the
`Destination ES is calculated,
`4) the VC is set-up on the basis of this route,
`S) the number of round-trips for the VC set—up should be
`one, as with the current models.
`In general,
`this leads to a more accurate resource
`allocation scheme
`and management. The general
`problem is to design an architecture in terms of routing
`and VC set-up model which is as close as possible to the
`ideal case we mentioned above.
`
`82
`
`Petitioner Apple Inc. - EX. 1005, p. 3
`
`Petitioner Apple Inc. - Ex. 1005, p. 3
`
`
`
`3. A review of routing methods for
`QOS/VC routing
`
`3.1. SPF Distance Vector and Link State
`
`The SPF methods used for packet switching routing
`can be classified in two main classes:
`the Distance
`Vector (DV) algorithm based on a method developed by
`Bellman-Ford [BellFord] and the Link-State (LS) based
`on a method originally proposed by Dijkstra [Dijk].
`Examples of DV implementation are provided by RIP
`[RIP], DECNET Phase IV algorithms [SHW-86] and
`DR? [18010747] (the OSI protocol used to route the
`ISO-1P traffic in the inter-domain case). Examples of LS
`protocols are 18-13 protocol
`[18010589]
`[RFC1195]
`[Callon] (designed for the ISO-1P and adapted for IP),
`OSPF protocol (designed for 1? environment) [OSPF]
`and IDPR [IDPR]
`(a link state protocol
`for
`inter-
`administrative routing).
`In the basic DV method, the node keeps trace of its
`"distance" to each Destination. When a node detects that
`a distance to one of its neighbors has changed or when it
`receives an update indicating that the distance to one
`Destination node has
`changed,
`it
`advertises
`this
`information to its neighbors (except
`to the one from
`which it received this information). In this way, nodes
`can maintain routes corresponding to the shortest path
`based on particular metrics between any pair of
`Source/Destination ESs. With this method,
`the route
`calculation process is performed in a distributed way,
`each system keeping track of one or more routes for each
`Destination node it can reach.
`
`The storage complexity for basic DV is given by
`O(m.d.g) where m is the number of metrics, d is the
`number of ESs in the RD and g is the maximum degree
`of the graph edges in this RD.
`011 the other hand, with the basic LS technique, a
`node is aware of local topological changes (e.g. metrics
`changes, failure of nodes, ...), it constructs a link state
`PDU which will be broadcasted in the entire network. In
`
`this way, each node of the network can maintain the
`complete topological map of the network. The route
`calculation is performed in a centralized way by using
`this global topological map (by using for example, a
`Dijkstra-like algorithm). The storage complexity is in
`the order of O(m.n.g) for the graph and 0(m.d) for each
`SPF route (In is the number of Systems = ES+SN). Once
`a topological change is detected,
`this requires
`the
`execution of an SPF style algorithm for each Destination
`node m times (In is the number of metrics),
`i.e. a
`
`processing complexity in the order of O(m.d.n.g). In the
`basic LS algorithm, the SPF algorithm used should be
`the same in each node for the same Routing Domain in
`question otherwise long term loops may occur since two
`nodes may disagree on the "best" route to be used. A
`Routing Domain has a limited size in terms of nodes and
`ESs supported, both DV and LS methods scale in such
`environment.
`
`3.2. Pre-computed routes vs. on-demand routes
`
`In a datagram environment such as IP and CLNP, it is
`important to have pre-computed routes since each packet
`is to be routed and forwarded accordingly. This is
`possible since 1)
`these architectures use a limited
`number of metrics and 2)
`these metrics are never
`combined. This also means that routes are maintained
`on-line: for LS, each time the topology changes, the SPF
`algorithm is executed while in the basic DV, when in a
`node the distance to a Destination node changes,
`this
`node advertises the modification as soon as possible to
`its neighbors.
`As far as QoS/VC routing is concerned, it is simply
`not possible (or far too expensive) to pre-compute and
`store
`a
`route
`for
`each
`source/destination/QOS
`combination. For the DV method,
`this means that
`a
`Switching node should continuously maintain a routing
`base in the order of 0(v**q.d.g), where q is the number
`of different QoS parameters and v is the maximum
`number of different values that one QoS parameter can
`take.
`In the LS case, a LS PDU (Link Sate PDU - i.e. a
`message sent by a node indicating the state of its links)
`can carry the m (208 values
`(or
`for
`some 005
`parameters, a min and a max. value indicating that any
`value in the range are supported) characterizing a given
`link. In this way, the complete topology map including
`QoS characteristics associated to each link can be easily
`maintained. However, a pre—computed method requires
`to execute in the order of O(v**q.d) times the route
`calculation algorithm in each node.
`We can note that the request in terms of QoS required
`by the E53 in the network generally corresponds to few
`well—know pattern of QoS combination values. Examples
`of this are given in the previous section. This means that
`a number of requests are predictable in terms of QoS
`combination requirements for which routes can be
`computed by the network in advance. Nevertheless, we
`can not exclude that
`the users might
`require an
`unpredictable combination of QoS parameter values.
`This means that the number of different routes to a
`
`single Destination ES that the network should maintain
`may be lower than P, with P<<v"'*q. The value of P
`
`83
`
`Petitioner Apple Inc. - EX. 1005, p. 4
`
`Petitioner Apple Inc. - Ex. 1005, p. 4
`
`
`
`depends on the resources that the network is capable to
`allocate for route storage and calculation purposes.
`
`3.3. Link State vs. Distance Vector
`
`About the LS method in a VC context, we formulate
`two observations:
`the first one concerns the topology
`distribution process and the second one the route
`calculation itself.
`
`1) The process of updating the link state data base can be
`based on different strategies. At one extreme, an LSP
`PDU can be generated by a given node each time a
`new VC is established through it. At
`the other
`extreme, such LSP PDU is sent only in case of a
`topological change (i.e. node and/or link failure and
`node/link set-up). The second one is very poor in the
`sense that it induces a kind of distributed backtracking
`algorithm at the VC set-up phase because of lack of
`network load information. Surprisingly, the first one
`could be as poor as the second one since the network
`is always unstable. A solution for this problem is that
`the nodes send updates less frequently, e.g. when a
`Moular condition in
`the nodes holds
`(e.g. a
`particular threshold is reached...) However, in this
`case, the number of attempts to establish a VC before
`succeeding directly depends on the tuning of these
`threshold conditions. Other
`authors propose
`to
`geographically limit the advertisement of LS routing
`updates. However, these techniques can not always be
`applied to calculate routes to all the Destination nodes.
`In addition,
`the same problem mentioned above
`remains.
`
`the quality of a route
`2) As in our VC environment
`depends
`on
`a
`combination
`of
`different QoS
`requirements. The route computation induces either to
`a backtracking algorithm (i.e.
`time complexity =
`O(Exp(n)) or to a storage complexity in the order of
`OCEXPO'ID-
`The distributed nature of the DV class of algorithms
`makes the on-demand route calculation more difficult. In
`the DV method, routes are calculated in a distributed
`way and the route calculation is initiated from the
`Destination node. A possible solution to fix this problem
`is, on a user request, to ask the Destination node to
`initiate the DV algorithm in order
`to establish the
`required route. However,
`this implies that additional
`round-trip time is needed to complete the VC set-up
`phase.
`
`3.4. Loop problem
`
`inherent to the method used to establish routes [Cheng]
`and from "transient
`loops" during the convergence
`process of the algorithm due to the fact that inconsistent
`routing tables are present
`in systems. Solutions have
`been proposed to limit the long term loop problem (e.g.
`Split Horizon. Hold-count, complete route recording, ...).
`In contrast, LS protocols only induce transient loops.
`This only results
`from the dissemination of LS
`information because nodes have inconsistent copies of
`the complete map topology.
`In addition, LS has a
`convergence time that is better than DV (in the order of
`the "diameter" of RD). Furthermore a new class of DV
`algorithm has been designed to avoid loops during the
`DV route calculation process. In these circumstances,
`the DV and the LS algorithm have a similar behavior
`with respect to the convergence time [Shankar].
`To provide loop free routes which are essential in a
`VC context, either a loop detection or a loop avoidance
`mechanism is
`to be supported. Examples of loop
`detection techniques
`include the conveyance of a
`connection identifier with the call packet or the support
`of a complete route recording function. Loop avoidance
`can be achieved for example by the use of a completely
`loop free routing algorithm (both transient and long term
`loops)2 or by Source
`routing. These algorithms have
`been extensively discussed in the literature (permanent
`and transitory loops at distance-vector-like algorithms
`and transitory loops at
`link-state-like
`algorithms)
`[Cheng]
`[ZauGar]. However, usually the suggested
`solutions were related to SPF (Shortest-Path First) search
`which either prevents their adaptation to QoS based
`routing [Awerb] or this adaptation would imply a very
`poor performance [Garcia89].
`In addition to loop
`avoidance, Source routing has a number of advantages:
`1) For LS, one advantage is that only one system has to
`calculate the route. Another advantage is that different
`nodes can use different algorithms, this can be useful.
`e.g., for testing purposes.
`2) For DV method, this allows to design on-demand DV
`algorithm in which the nodes are "memory-less " with
`respect to the DV method on the condition that the
`complete route is recorded during the route calculation
`process.
`
`3.5. Our basic proposal
`
`The method that will be described in this section aims
`
`to achieve a trade off between user requirements in
`terms of QoS sensitivity of routes and the network
`
`It has been shown in the literature that the basic DV
`solution suffers from "long term loops" which are
`
`2DUAL [Garci389u] provides solution to both DV and LS loop
`
`problem. However, we don‘t address this possibility in this paper.
`
`84
`
`Petitioner Apple Inc. - EX. 1005, p. 5
`
`Petitioner Apple Inc. - Ex. 1005, p. 5
`
`
`
`capability to provide such routes. The routing protocol
`will provide means to calculate routes satisfying user
`requirements at different costs (for the network) and
`with different levels of reliability. The design of the
`protocol
`is
`principally
`based
`on
`the
`following
`observations:
`
`1. Considering the number of QoS parameters and the
`number of possible combinations of these parameters,
`it
`is too expensive (in some cases impossible)
`to
`calculate and maintain routes for each of them.
`2. Not all the QoS combinations will be used in the real
`world.
`
`3. Those combinations that are actually used can be
`classified depending on their usage frequency.
`4. The major part of the traffic on a network demands
`few well-known QoS characteristics. Nevertheless, on
`comparatively rare occasions, specific traffics will
`require an arbitrary QoS characteristic. The network
`should be able to provide for both of these situations.
`As a result of these observations, the routing protocol
`will provide two basic categories of
`routes: pre~
`computed and on—demand routes.
`
`3.6. Pre-computed routes
`
`Pre-computed routes constitute the basic service
`offered by the routing mechanism. These routes
`correspond to those matching the QoS profiles defined
`by the network management. The main characteristic of
`these routes is that they are always available and are
`maintained up to date by the network. The algorithm
`used for this purpose is a variation of the classic
`distance—vector algorithm called path-vector (similar to
`IDRP). Combined with Source
`routing for call
`forwarding,
`this method
`offers
`some
`interesting
`advantages:
`1. Routes are calculated in a distributed manner.
`
`2. In case of a modification on a path, updates will only
`be sent on the affected path.
`3. Loop formation at the connection set up time will be
`avoided by Source routing.
`4. Backup routes can be calculated and maintained.
`
`3.7. On-demand routes
`
`These are routes with special QoS requirements not
`covered by the pre—computed routes and are calculated
`only on the request of the user. In this case the user
`should formulate an explicit request for the route to be
`calculated. These routes can be of one of the following
`types:
`1. Routes which are memorized by Switching nodes but
`not updated in case of a modification on the route. In
`
`case of a modification on these routes, the only action
`taken by the network is to inform the user and purging
`the route.
`2. Routes which are calculated by the network and not
`updated at all. In case of modification on these routes.
`no action is taken by the network.
`The user can choose one of these types depending on
`the importance of the route. 0n the other hand. with the
`route calculation demand the user has the possibility to
`specify ifat the same time he wants a VC to be set up or
`not.
`
`In order to achieve this, at the reception of a route
`calculation demand, the network will be explored to find
`a suitable route to the Destination node. In the first
`phase, a special packet (Exploration-PDU) containing
`the user request will travel through the network from the
`Source node to the Destination node covering all the
`possible routes conforming to the user requirements. In
`this way the Destination node will be notified of the
`possible routes from the Source node to itself. In the
`second phase, the Destination node will choose one or
`several of these routes and send them back to the Source
`node. If the Source node had requested a VC to be set
`up, it will be done during the second phase.
`
`3.8. VC set-up
`
`As a result of QoS negotiation process, the VC is set
`up from the Destination ES to the Source ES. This means
`that
`
`1) for the art-demand routes, one exploration process is
`initiated from the Source node to search the routes
`
`2)
`
`that match the user requirements, and the Destination
`node selects one of them and establishes the VC on
`this route, and
`request
`a
`for
`routes which are pre-computed,
`(including requested (208 by the Source ES, Source
`and Destination ES address - this packet is referred to
`as
`the Set-up_1ntention packet)
`is
`sent
`to the
`Destination ES by using some available route. then the
`Destination
`node
`selects
`an
`appropriate
`route
`according to the requested QoS, its own capability and
`available routes and initiates a VG set-up for this
`route.
`
`in this way, we provide a three-parties
`Note that,
`negotiation process with only one round-trip time.
`
`3.9. Problem analysis
`
`In this context we can note the following:
`008
`Parameter
`interrelation:
`Considering
`parameters as a set of independent parameters could
`simplify their study but it is evident that some of them
`
`85
`
`Petitioner Apple Inc. - EX. 1005, p. 6
`
`Petitioner Apple Inc. - Ex. 1005, p. 6
`
`
`
`are closely related to each other. For instance, supplying
`a given security level over a link might imply that a part
`of available bandwidth will be consumed to provide this
`security level. Therefore, monitoring and updating the
`available resources by a node and the current status of a
`link is not a simple task and should take into account the
`interrelation between these parameters.
`the
`Resource reservation: as explained earlier, at
`first phase of the Exploration Switching nodes will
`choose (based on their available resources and the
`current status of their links) to accept an Exploration-
`PDU or not. But, the actual reservation of resources is
`made in the second phase of the Exploration, using a
`Set-up message. Thus if the status of an Switching node
`or its links changes during the period of time between
`the acceptance of an Exploration-PDU and the reception
`of the corresponding Set-up message,
`the VC set-up
`procedure may fail. We can propose two solutions: 1)
`make the resource reservation at the first phase, at the
`reception of a Exploration-PDU; or 2)
`include an
`estimation of VC set-up success probability in the
`Exploration-PDU. This estimation can be based on the
`number of Exploration-PDUs processed by a node
`during a certain period of time and the available
`resources of the node during this period. It will allow the
`Destination node to choose a route having the greatest
`success probability.
`The first solution guarantees the success of VC set-up
`on a route. But it is very inefficient since lot of resources
`will be blocked on Switching nodes which will not
`participate in the final VC set-up. The second solution
`improves the probability of VC set-up success.
`Exploration complexity: The basic
`exploration
`algorithm is exponential but
`the complexity of this
`method is
`reduced by limiting the propagation of
`Exploration-PDUs using the following mechanisms:
`1. Before forwarding an Exploration-PDU to an adjacent
`node,
`the accessibility map is consulted. So no
`Exploration-PDUs are sent to an adjacent node not
`leading to the Destination node.
`2. Before fonvarding an Exploration-PDU to an adjacent
`node. the requested QoSs will be compared with the
`Intermediate-Q05 field of the packet and the currently
`available QoS values for the link to the adjacent node.
`50, if this link does not satisfy the user requirements,
`no packets will be sent to the adjacent node.
`3. Packets will not be sent over a loop. The path field of
`Exploration—PDUs contains the complete list of nodes
`traversed by the packet. By using this field,
`loop
`formation is avoided.
`
`4.
`
`Conclusion
`
`In this paper, we investigated methods for routing in a
`VC/QoS context. When analyzing the different methods
`for packet
`switched routing.
`two solutions
`seem
`interesting enough to be pointed out. The first one is to
`design a method based on Link state for VC/QoS
`routing. However, LS has two important disadvantages:
`one is that the algorithm for route calculation (which is
`calculated in one node)
`is exponential. The other
`problem is related to the LS strategies for update
`distribution. Frequently distributed updates may result in
`continuously unstable situation which can lead to a very
`poor performance.
`The other method is based on DV-style algorithms.
`This solution has a number of advantages including the
`fact that updates are only sent over affected routes. In
`addition, the process of route calculation is distributed
`among network nodes. Furthermore.
`the DV-based
`algorithms save a significant amount of memory in
`comparison with LS.
`In the context. of VC/QoS roofing we noted that the
`number of QoS parameters and the number of possible
`combinations of these parameters make it unfeasible to
`continuously calculate and maintain routes for each of
`them. A subset of the possible QoS combinations is
`actually useful and they can be cl