Computer Communications 23 (2000) 418–424
Research note

A distance-vector routing protocol for networks with unidirectional links
F.C.M. Lau a,*, Guihai Chen a,b, Hao Huang b, Li Xie b
Department of Computer Science and Information Systems, The University of Hong Kong, Hong Kong, People’s Republic of China
National KeyLab of Novel Software Technology, Nanjing University, Nanjing, People’s Republic of China
Received 8 April 1999; accepted 13 October 1999

We propose a simple distance-vector protocol for routing in networks having unidirectional links. The protocol can be seen as an
adaptation for these networks of the strategy as used in the popular RIP protocol. The protocol comprises two main algorithms, one for
collecting “from” information, and the other one for generating and propagating “to” information. Like the RIP protocol, this one can handle
dynamic changes and tolerate node and link failures in the network. q 2000 Elsevier Science B.V. All rights reserved.
Keywords: Routing algorithm; Distance vector; Directed graph; Undirectional link; Routing information protocol

1. Introduction or improve their own tables. We adopt the RIP strategy, but
not the RIP protocol itself because it does not work for
Current Internet routing protocols stand on the assump- UDLs. Although newer protocols for interior gateway
tion that any links between two neighboring nodes are routing (IGPs) have been introduced, such as the open
bi-directional. Unidirectional links (UDLs), however, are shortest path first (OSPF) protocol [6,7], RIP still maintains
emerging and could be as ubiquitous as bi-directional its edge in certain aspects. In small networks, RIP has very
links in future networks. Examples of systems having little overhead in terms of bandwidth used and configuration
UDLs include direct broadcast systems (DBSs) using and management time, and is amenable to distributed,
satellite [1] and mobile radio networks [2]. Impacts of asynchronous operation. RIP is also very easy to implement,
unidirectional links on routing protocols are discussed in especially in relation to the new IGPs. RIP has also been
Refs. [2,3]. According to Ernst and Dabbous [3], adapting widely adopted by many personal computer manufacturers
current protocols for bi-directional networks so that they for use in their networking products, including Xerox,
also work for networks with UDLs (NUDLs) does not Apple, Novel, 3Com, Ungermann-Bass, Banyan, and
seem to lead to good results. They discussed ways to Cisco. It is expected that RIP will continue to enjoy its
adapt existing protocols for unidirectional networks. For popularity for some period of time.
distance-vector protocols, they gave an example of handling In adopting the RIP strategy, we realize that in an NUDL,
a simple unidirectional network. But a complete protocol for exchanges of routing tables between neighboring nodes
the general case is not given or discussed. New protocols become difficult because information can only flow in one
need to be designed for networks having UDLs. We propose direction along a UDL. We also realize, however, that if the
such a protocol in this paper. network is strongly connected, there must exist a path—a
The protocol we propose is new in the sense that it is not a series of UDLs—that goes the other way connecting exactly
direct modification of any existing protocol. Nevertheless, it the same two nodes. This latter path is sometimes referred to
uses a strategy similar to that of the distance-vector-based as a “back channel” between two adjacent nodes in the
routing information protocol (RIP) [4,5], the well-known literature on NUDLs. Our protocol finds this back channel
internet protocol that is widely deployed in internetworks for a pair of adjacent nodes dynamically during runtime. In
as well as single networks. The RIP strategy is to have every contrast, the modified RIP discussed in Refs. [8,9] depends
node periodically send its routing table to all its immediate on the existence of a pre-arranged back channel between a
neighbors who would then use the received table to update pair of adjacent nodes.
Clearly, in a strongly connected NUDL, between every
* Corresponding author. Tel.: 1852-2859-2170; fax: 1852-2559-8447. pair of nodes there must exist at least one circuit-i.e. a cycle
E-mail address: (F.C.M. Lau). of nodes. Each of the two nodes should be able to reach the
0140-3664/00/$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.
PII: S0140-366 4(99)00196-6

Any protocol that is that represent a good approximation of the current situation going to work for NUDLs can very much be viewed as a of the network. and its TO table the node’s the best routes from the feeding node to all the other nodes current view of the shortest paths from this node to various on the circuit. and fields Given a UDN of n nodes. Using their sented by a capital letter. to node Q by (P. Only the TT is used in actual routing. every node packs its FT in a packet. other node by following this circuit. also entry. (a) A FROM entry. a certain event (to be discussed in the next two to time. The distance On the other hand.B.Q). and Q is P’s to-neighbor or t-neighbor. P. it is important that the protocol would select of (P. The protocol must strive to produce routing tables sections) in a node would trigger the node to send its TT . and the second node following ND sion in these two sections assumes that the networks have in this path is NX. unidirectional networks or UDNs. the last node of the path.NX is the each node a routing table containing an entry for each of the NX field of the entry e. when created or when “renewed. / Computer Communications 23 (2000) 418–424 419 Fig.TT is P’s TT. are like “islands” in a large space. Section 5 then shows that the protocol works also for When TTL drops to zero. Time To only UDLs-that is. We table. we use the object-oriented “dot” notation to represent a node’s tables. the distance attempts only to discover one circuit for every pair of between A and B with respect to a certain path is the sum of adjacent nodes connected by a UDL. to this We present the protocol in Sections 3 and 4. 1(a). node. respectively. Note that it is possible that none of these is a The protocol recently proposed by Prakash and Singhal true shortest path because knowledge takes time to travel in [2] in the context of mobile ad-hoc networks requires a distributed system. the entry would be deleted. a networks where nodes and links may go down occasionally. P is Q’s from-neighbor or f-neighbor multiple circuits could exist for the same destination node. requirements. the entries in both tables capture in terms of its correctness. (A. the protocol described in this paper between two adjacent nodes P and Q is d(P.” is given a TTL known as archipelagos where clusters of bi-directional links value of T. The discus.Q). e.X). TTL}: we propose requires O(n) space in a node. the protocol {ND. Similarly. It is Each node maintains two tables. Ernst and Dabbous have Every node is assumed to have a unique identity. We consider dynamic Periodically. and bandwidth and space the distances between this node and the other nodes. The circuit represents also various nodes to this node. 2. F. repre- proposed such a circuit discovery protocol [3]. 1(b). (b) A TO entry. For example. Live (TTL) is the timer value associated with this entry. other n 2 1 nodes in the network. circuit discovery protocol. Lau et al. that ends at X is written as necessary routing table from the circuits. They gave. DT. the problem is to generate in within an entry. As this is a find it difficult therefore to give an evaluation of the protocol distance-vector protocol. Preliminaries In the following. The link cost many cycles. and NX is the next node following this node. only the best or appropriate circuits to keep in a node. A node’s FT receiving node a back channel for propagating its routing represents the node’s current view of the shortest paths from information to the feeding node. however. a path A simple procedure can then be used to generate the from A. and the messages the nodes send out are of size O(n). is the number of nodes in the network. We denote a UDL that joins node P protocol. 1. and the table needs to This is the format of an FT entry: be sent around from time to time. In contrast.….Q) is denoted by d(P. as shown in Fig. and so on. table entries. This entry represents a path from some node ND. only a high-level overview of the protocol. Distance here actually refers to link cost. shown graphically in Fig. maintaining in every node a table of size O(n 2).M. Every so the topology may change. abbreviated FT and TT. and sends it to every t-neighbor. and link costs may vary from time often. A field value “–” means that the field is not important in the discussion. The circuit gives the the costs of those links making up the path. i. each node maintains and manages a set of circuits. except where ND is the destination node.e. NX. An networks having both UDLs and bi-directional links. other nodes. a FROM table and a TO not clear how this is exactly done using their protocol. to B. where n the FROM table is for a node’s internal use. the distance is DT.Q). Each link has a link cost which we assume is known at all Considering the fact that even a small graph could have times to the two nodes connected by this link.C. FROM packet. The format of a TO (table) entry is similar.

IF …e:NX ˆ e 0 :NX† or …e:DT # e 0 :DT† replace e 0 The tables given above represent a stable state of the by e.C. nodes have entries that are as shown in Table 1. using the above algorithm. p and q.8. Initially.2.2.7.D. stationary one. We say that two entries. Let’s look at how the algorithm responds to the following possible situations.FT would be equal to P.5.– D.A. if the algorithm E.A.5. It can also be easily seen that all the entries in a 2. suppose that node D first learned of node A (a) for every entry f [ F. in Table 1 particular the timers. the e above) that can be derived from the FROM packet: (1) entries representing alternative paths but whose DT is longer than the DT of existing matching entries. where the number on a link is the link’s paths that join other nodes in the network to this node. it can handle also node . 2.– D.1. f:DT 1 d…P. Construction and maintenance of FROM tables Instead of a formal proof.FT. But if the network is a dynamic one so that and executes the above procedure for this entry.B.C. all its incident D.A– E.– E. Each node periodically sends to all its t-neighbors a node’s FT represent shortest distances from all the other FROM packet containing its FROM table.B.4.– B.– • Node and link failures: When a node fails.– } is replaced by {A.C. 2(a).C. If in fact the network is a add e to Q. it is not deleted.ND. the entry is channel.– – D. d…P. system which will not change as time goes by as long as ELIF e:ND ± Q the network remains stationary. One can easily verify that these tables are in fact what Algorithm FROM: comes out of executing the algorithm several times after 1.– E.7. then the condition applies.7.E. T} periodically. After several initial rounds of message exchanges is essential in the construction of the node’s routing table.C.FT.– } in its {f:ND. The example is as A node’s FROM table collects information about the shown in Fig. f:NX. e ˆ through B an C.5. Q generates an entry. When a node Q receives a FROM packet F from the “procedure” that prunes away the longer paths. and the entry IF there exists a matching entry e 0 [ Q:FT {A. Q.420 F. Q†.– } in a FROM procedure: packet from C. links and nodes might be deleted or added and link costs might change over time.9.ND ˆ a. Lau et al. and (2) the entry that represents a self-cycle (form Q back to Q).D. (b) The tree at E.– C.– – can handle (single) link failures.3.E.E.FT in contents in Note that the above algorithm ignores two kinds of entries the next moment. (a) A simple UDN.D. the FROM tables of the various the TT. For f-neighbor P. 2(a) as the example.– – B.T} in D.C.A. instance. When the times of a FROM entry expires.– A. we use an example to demon- strate the correctness of the algorithm.2. Note that if P sends out a FROM packet F.C.– – C.e.E. in a TO packet to a certain f-neighbor through some back 4.2.C.7.5.– A. always true that F.C.– C.6. Q†.– B. all FROM tables are empty.2.– links become inoperational. containing P’s FROM table.M. Therefore.5. 3. in the algorithm to make sure that the FROM tables at stable state table entries reflect or are a close approximation to the A B C D E current situation at all times.B. / Computer Communications 23 (2000) 418–424 Fig.6. it is not necessary that entries be kept alive (b) Q then generates the entry e ˆ {P. B.FT. then we need those little devices.– A.– Fig. if p. This is due to the “ # ” condition in 3.4. (i.C.3.6. match.D. It cost.– C. and so it has the entry {A. nodes to this node. startup. T} and executes the table. later on D receives the entry {A. using again – A.

T}.2.5. clearly.– } expires and gets deleted. the resulting paths We assume that the nodes in the network are capable of would form a tree. 1.3. from C to E. when the entry{A.C) is one instead of two.A. 1(a)).– }. • Changes of link costs: A decrease in a link’s cost presents 2(b) shows E’s FROM tree at stable state. Fig.– } to {C. if (A. entry {A. Assume that node Q has just received a FROM packet F from its f-neighbor P and F. X ± P—that is.5. according to Step 4 of the algorithm. there cannot exist two or more matching entries {A.5.C) does not come back in an FT. by the procedure in step 3(a) of Algorithm will soon be replaced by {A. The entries of a node’s FT at any time form a tree.C). it example.5. why it is necessary to send the entire FROM table to every Note that the path (from A to E) just mentioned represents t-neighbor from time to time.7. i.– } and {E.D) in nodes (C. We can see that the algorithm ignores all incoming say X. performing source routing.7.FT contains an entry with ND ˆ Q.C.B. although the entry with ND ˆ Q is A) in the path to any subsequent node in the path.B. Subse. entries for the other nodes in former. A could take the trouble of passing is used to trigger the execution of the other algorithm.5.5. As we will see in Section its entry {C. F. if the cost of (C. up soon enough. What A receives is in fact the other hand.– } in D. and {A.C. Suppose for a and X must have sent along all of its own entries includ- moment that the cost of (B.C) is broken when its entry {A. For this to be possible. This is not necessary.D.– } from C shows up in A D.FT. But this knowledge will soon be proven not lead to P. Interestingly.FT are deleted. D. {A. For example.FT with its entries that are worse in terms of distance than their ND ˆ X. Given any node P. for ignored as far as updating Q’s FROM table is concerned. We should also point out the reason for the the branch must have been propagated to P through X.D. but the latter is obviously better than the FROM and by (a) above. this information to all these nodes in the path for incorpor- Algorithm TO. We refer to this tree we just discussed as a FROM tree. there is no entry in P. To perform source routing from Q to P.M. A non- C. A can use the It should now be clear from the examples just presented that source routing method to send packets to E via this back timers are indeed essential in a dynamic environment. Lau et al. but deleted entries.– } that comes to D via C would not stop the with ND ˆ X must not have been propagated to P before.C) in Fig. while all the other nodes in this branch have matching entries in the local FT.FT with ND ˆ X. These two This could not have happened because by Algorithm entries match.E. if we A. / Computer Communications 23 (2000) 418–424 421 failures.e.– } will also be deleted. By Proposition 1. E. (a) A node. D will update mation that A should make use of.6. ating in their TT. . to newly discovered cases and will be appropriately handled its t-neighbor A. (b) Consider an orphan branch that terminates at a node.6.C. as every Every entry in a node’s FROM table contains a link one of these nodes would have received or will receive a (that highlighted in Fig.5.C. also E’s view of the shortest path from any node (other than In Algorithm FROM.T} and FROM. equality part of the “ # ” in Step 3(a). however. we have a cycle (a closed circuit) that joins by Step 3 of the algorithm. Algorithm SOURCE_ROUTE: Proposition 1. P. The have more than one entry in P. On the above condition.B. the entry their entries in P. ing an entry with ND ˆ X. the tree in Fig.– }. All the tree edges (links) are pointing backwards towards After a while.– } because of the 4. as well as {D. A would incorporate this information in its TT. Q encodes this path in the packet (s). or both. or (b) an “orphan” branch which does through (A. no problem as the link will be treated like any other When E sends its FROM table.C. tree has either (a) a cycle in which there is a node having D and E. and the entry the root. F. {A. say X. connect all these links together.FT contains a non-tree. respectively.– }. 2(a) is changed from three to four. This is useful routing infor- Fig.– }.4. still think that A can reach them more than one parent. and then back to A.2.C. D. the branches are the different paths embedded in the table. which is the first link in similar packet from its f-neighbor like what has happened to the path represented by the entry. which will be discussed in Section 4. E’s entry {A.6. 2(b). however.FT contains a (shortest) Proof. D will learn of the fact that (A.D. Suppose the link (A.C.– } instead of having to wait for {A.– }. in Assume the contrary the P. Therefore.3.4. {A.T}. as its “virtual link” or “back channel” to E. E) in the circuit.C.e.C. The following algorithm will be useful in the construction and maintenance of TT. and channel. {A. some of C’s timers expire. When this new entry is propagated to other hand. A could treat the path to E as found in the circuit E later on-i.FT from expiring.B.C. it would immediately replace D’s entry {A.– }. an entry {A. the tree in question is rooted at path from Q to P. Either case is not possible: wrong.D. is handled by the condition …e:NX ˆ E’s latest view of the shortest paths from A to all the other e 0 :NX† in step 3.– }—it replaces E’s {C. having more than one parent would quently. 2(a) goes down.6.B. e ˆ {C. An increase in a link’s cost. Then.– } to expire. on A to C.8.

The latter information is to be found in the TT. it generates an entry. which takes place when A sends its FROM Algorithm SOURCE_ROUTE. Q.– .M.FT its path to B. and try to We need to show that the algorithms work for bi- install that in the TT. and e3 ˆ {D. it knows that Q considers (P. and executes the procedure: other hand. Imagine if there were another path from P the RIP strategy. N1 .– }. Q†. P. in fact. any node would be able {Ni . C. (b) Another view. Therefore. …. (a) for each entry t in T. –}: the circuit we discussed above coincides with the physical (b) For i ˆ 1 to m. C. to follow path from P to Q. When the timer of a TO entry expires. We need to argue for the necessity of sending TTs around 2. / Computer Communications 23 (2000) 418–424 E’s FT. T}. but not how this node may do so to that node. 1.TT. A can easily trace the path to be …N0 ˆ A. di . There is no need for N0 . Note that when a circuit is found in a node.C. Q. this is done in step 2(c) in the algorithm Q would have sent its TT to a different f-neighbor instead of below. table to B and B finds a circuit. but that entry is not part of the traced path. m $ 1: Let e1 . So P takes the TT from Q both as a confirmation of Q’s Algorithm TO: receipt of P’s FROM packet as well as an indication that (P. the FT coming from IF there exists a matching entry e 0 [ Q:TT f-neighbors might not contain all the necessary TO informa- IF …e:NX ˆ e 0 :NX† or …e:DT # e 0 :DT† replace e 0 tion. 2}.FT contains an entry whose ND ˆ Q. but the process will take a much longer time because A node’s FROM table is not a routing table because an entries like {X. d2 .Q). d…P. to P.X. Nm ˆ P†. The FROM table A receives from C ELIF e:ND ± Q does not contain the entry {A. We can treat a bi-directional link as a be the Q here. and therefore e1 ˆ {A. C. d3 ˆ 5 1 …2 2 0† ˆ 7: The above two steps. The FT a node Q receives from its (b) P then generates the entry e ˆ {Q.Q) (in Q’s view) is a shortest path from P to Q. T}. In fact. t:DT 1 d…P. such as that then embedded in the circuit.C. 5. T} and executes the proce- dure given above. deduce from C. Initially. {C. The following proposition 3. the node would send out its routing table to to Q (via other nodes) which has a smaller cost than d(P. the entry is deleted. Q generates the entry e ˆ ring. A receives from E a packet containing pair of UDLs. if the network is not a ring. d3 . On the and d0 ˆ dm11 ˆ 0. 5. C. Q†. Archipelagos Steps 2(a) and (b) of the above take the circuit as the new shortest path from Q to any other in the circuit. The following constructs and maintains receives Q’s TT.422 F. A then generates the following entries for e. Refer to the example in Fig. E ˆ N3 †. As a result. if the network is a ring. each bi-directional link is a . Lau et al. 2 }: Note that there is another entry in the table with NX ˆ N1 ˆ C. …. if F.– . By Step 2(b). When a node Q receives a FROM packet F from an because FTs that are being sent around seem to already f-neighbor P. T}. T}.Q) the shortest the TT. –. d1 ˆ 0 1 …7 2 5† ˆ 2 {D. all TT are empty. em be the sending any TT around in a ring. where the network by e.-}. does not need a proof. 2. Let’s refer to Fig.FT where ei ˆ {–. 2 again. E. Proposition 2. {E. e2 ˆ {C. T} f-neighbor P contains at most one path that goes from Q and executes the procedure given above. 2 }. Suppose E’s table is the one as shown in Table 1. N2 . The TT is Step 3(b) is there in the algorithm because when P the routing table. d1 . contain a fair amount of TO information. B must send (c) Q sends a TO packet containing its TT to P using its TO table to A. 3. {B. When a node P receives a TO packet T from its child Q. and let the A there directional links. Ni . and therefore.TT. (a) An archipelago. step 3(b) of the algorithm in order to finally come to Q. D.7. (a) Q traces the path from Q to P in F. That is. 4. C. D.– } and so A cannot add e to Q. d2 ˆ 2 1 …5 2 2† ˆ 5 Fig.FT: …Q ˆ Algorithm TO can stop after Step 2(b). where X is Q’s t-neighbor in the entry there indicates how some other node may send packets circuit would have to come a full circle one link at a time via to this node. This is easy to see because corresponding entries in F. 7. where di ˆ di21 1 ‰ei :DT 2 ei11 :DTŠ to generate all the necessary TO entries for itself. Construction and maintenance of TO tables algorithm would still work in coming up with the necessary TT. e ˆ {t:ND. N1 .B. consists of two rings. 2. its f-neighbor. given a circuit. can be omitted and the 4.

in an archipelago.D. in fact.11].– . 3(a). The simplest UDN. The reason is clear in Fig.P. For Fig.C. and would be • the protocol for the boundary nodes.C. because TT can be all the edges. F. of the need to keep entries alive even after the network has Therefore. boundary node.A) but also the path (A. With such a message-thrifty variant. and to generate the as a regular node. An alternative would be to send out apply only the FROM algorithm to the islands (thus saving messages only when there is a change to the topology or some execution time and communication bandwidth). perimeter of an island. In Section Furthermore. the problem can be overcome using techniques It seems that in the above we are sending too many tables such as split horizons and poison reverse updates [10. case (b) has to go through some non-island edges two views of an archipelago. (# 69803005).B) for computing new TO entries. We give an example below to routing path to enter.Q. which is also the island. those at the independent of the network size. The next exchange would therefore “close a The RIP protocol is widespread because of its simplicity.– } and {P. would not do anything to its own TT understood if we view the two connections (one UDL plus a except renewing the entry {Q. . however. appears as a simple cycle in the one on the right. Lau et al. networks. we see that routing cannot be easily broken into an intra-island component and an extra-island component. P and Q would insert {Q.2.– }. also called an n is the number of nodes in the network. 5 better entry. 7. A would have more than just the circuit (A. over a single bi-directional edge. Initially. a simple cycle involving only two nodes. For an all-bi-directional network. as shown in Fig. For this simple example of an archipelago having just one island. This can be easily receiving Q’s TT. upon RIP. As our protocol is modeled after these tables out to their respective t-neighbors. 4. Refer to Fig. similarly for Q. Like in traditional RIP-based order to keep their table entries alive. We are now in the process of designing an algorithm incorporating both of the above possible improvements. respectively.– . instance: case (a) in the figure has to go through the island even though both the source and destination are outside of cycle. we have shown that the algorithm works for cycles having we cannot treat an island as one abstract node and allow a more than two nodes.– . or pass through the island at any illustrate that the algorithms also work for simple cycles. 2.P. P back channel) between every two nodes in a UDN as a and Q will continue to exchange their tables periodically in single bi-directional path. because of the shortest-path requirements.– } It suffers however from convergence problems such as the and {P. Fig.Q. TO table from the FROM table.M. the FROM table. it could run into similar problems. and when a link changes cost and only those affected entries apply both the FROM and the TO algorithm to the rest of the need to be sent around. come and go. D. C must have cannot get away with passing TTs and FTs around just like dropped the entry {A. Our approach has been a “busy” one because derived from FT. one approach would be to reached a stable state. 5. B. and finally A. The TO algorithm seems superfluous. form a bi-directional subnetwork. would be of a much smaller size to keep track of the islands. than those messages carrying the entire table. where The FROM and TO tables have size equal to O(n).– } because {A. and therefore it is not easy to discover and These messages. then E. exit. This requires maintaining extra edges in a FROM table other than those of the FROM Even if we can treat islands separately. Routing through an island. 6. the island nodes tree. Consider the simplest UDN. however. The difficulty with this approach. At some point in time. both FT are empty. / Computer Communications 23 (2000) 418–424 423 where several instances of shortest-path routing are shown.C. where each bi-directional link even though the source and destination are island nodes. After the first exchange of FT. and then send count-to-infinity problem. both as an island node and to send only one table. circuit” for both P and Q.– .E. P and Q each has an entry in their FT: {Q.B. 4.– . 4. 3 shows the island. Consider Fig.– } was a their non-island counterparts. If somehow C had kept this entry (marked special) which was then propagated to D. could be complicated because Another space and bandwidth saving possibility would be they have to play two roles. it suffices size of messages that are being sent around periodically over to execute only the FROM algorithm.– }.Q. into their TT. Acknowledgement Guihai Chen is supported by a China NSF grant Fig. links as well as nodes might control messages periodically and using watchdog timers.3. P. where A.– }. is that: nodes still need to keep an eye on each other by exchanging • in a dynamic environment. C. Concluding remarks respectively. network.

-G. M. Reading. Hedrick. Internet Draft. Huitema. Zhang. Routing in the Internet.C. Routing in Communictions Networks. Internet Request for Comments 1723. 1995.424 F. Dabbous. Lau et al. NJ. in: M. W. [3] T. www. Moy. wireless local access. S.M. March 1996 [http:// March 1999. Proceedings of the First International Workshop [8] W. Routing Information Protocol. C. INRIA Sohpia-Antipolis. INRIA Research Report No. Hungary. [5] G. Prakash. Duros. Dynamic routing in networks with on Satellite-based Information Steenstrup [4] C. .txt]. June 1988. A circuit-based approach for routing in [10] C. Internet Request for (Ed. Englewood Cliffs. Version 2—Carrying Additional Information. Prentice-Hall. shop on Satellite-based Information Services. 2. Dao. Huitema. Prentice-Hall. 1994. pp. OSPF: Anatomy of an Internet Routing Protocol. 1998. Impact of unidirectional links in wireless October 1997. [2] R. 1995. November 1997. Addison- [1] Y. Distance-vector routing. Englewood Cliffs. Moy. Ernst. Mobile Networks and Computing. ad-hoc networks. Rutgers University. November 1996. E. 1991. Steenstrup. 3292. OSPF Version.T. Malken. Budapest.). [11] G. T. / Computer Communications 23 (2000) 418–424 References [6] J. Ernst. 24–29. Singhal. Comments 1058. Proceedings of DIMACS Workshop on [9] E. Malken. Internet Request for Comments 1247. Proceedings of the Second International Work- pp. [7] J. MA. Handling of unidirectional links with RIP. NJ. unidirectional links. M. Integrating direct broadcast satellite with Wesley. New York. NJ. Duros. Dabbous.inria. 83–157. unidirectional links networks.