1. Technical Field
The present invention generally relates to mobile networks and, more particularly, to assigning gateways for heterogeneous wireless mobile networks.
2. Description of the Related Art
Inter-domain networking across mobile ad hoc networks is an important capability to enable practical applications, such as search and rescue operations by multi-agencies, disaster recovery efforts by multi-national organizations (such as the RED CROSS, the MEDECINS SANS FRONTIERES, law enforcement), and coalition military operations by multiple forces in a region with little infrastructure support. Inter-domain networking allows different organizations with potentially heterogeneous networking technologies to communicate with each other while preserving the organizational boundaries and their own networking policy. In recent years, the research community started to pay attention to this important yet relatively unexplored problem, and several proposals have been made to address technology gaps. The proposals involve the following: architecture and framework design; inter-domain routing and policy support; and deployment and control of helper nodes to connect multiple domains.
One of the key components to enable inter-domain networking (in both wired and wireless networks) is the gateway. Gateway nodes act as control points to collect and distribute inter-domain routing information, and also enforce inter-domain routing policy enacted by each domain. In addition, gateways play the important role of isolating the intra-domain routing mechanism of one domain from that of other domains. More importantly, in mobile ad hoc networks (MANETs), gateways may need to perform protocol translation since different domains may employ different routing schemes (e.g., reactive, proactive, geo-routing, and so forth). General issues in designing an inter-domain routing protocol in MANETs and building gateways have been presented.
Previous work assumed gateway functionalities are statically assigned to a subset of nodes. While this approach will work well in a static scenario (e.g., wireless mesh), it may be problematic in MANETs due to node mobility.
According to an aspect of the present principles, a method is provided. The method includes exchanging routing and connectivity information between a plurality of nodes. Each of the plurality of nodes is respectively included in a corresponding one of a plurality of mobile ad hoc networks. The routing and connectivity information excludes global positioning satellite information. The method further includes determining, for a given node from among the plurality of nodes, whether a particular set of gateway functionalities of the given node are redundant with respect to one or more other nodes from among the plurality of nodes, based on topology information derived from the routing and connectivity information. The method also includes dynamically assigning the given node as a gateway or a non-gateway by respectively turning on or turning off the particular set of gateway functionalities of the given node when the particular set of gateway functionalities of the given node are respectively determined to be non-redundant or redundant with respect to the one or more other nodes.
According to another aspect of the present principles, a system is provided. The system includes a first multi-domain mobile ad hoc network including a first set of nodes, and a second multi-domain mobile ad hoc network including a second set of nodes. The system further includes a centralized server having a dynamic gateway assigner configured to receive routing and connectivity information from a plurality of nodes formed from the first set of nodes and the second set of nodes, and to determine, for a given node from among the plurality of nodes, whether a particular set of gateway functionalities of the given node are redundant with respect to one or more other nodes from among the plurality of nodes, based on topology information derived from the routing and connectivity information. The given node is configured to dynamically assign itself as a gateway or a non-gateway by respectively turning on or turning off the particular set of gateway functionalities of the given node when the particular set of gateway functionalities of the given node are respectively determined to be non-redundant or redundant with respect to the one or more other nodes.
According to another aspect of the present principles, a computer readable storage medium comprising a computer readable program is provided. The computer readable program when executed on a computer causes the computer to perform the respective steps of the aforementioned method.
According to yet another aspect of the present principles, a method is provided. The method includes exchanging routing and connectivity information between a plurality of nodes. Each of the plurality of nodes is respectively included in a corresponding one of a plurality of mobile ad hoc networks. The routing and connectivity information excludes global positioning satellite information. The plurality of mobile ad hoc networks includes multiple intra-domains and multiple inter-domains. The method further includes deriving a real-time intra-domain topology of the multiple intra-domains and a real-time inter-domain topology of the multiple inter-domains from the routing and connectivity information. The method also includes determining, for a given node from among the plurality of nodes, whether a particular set of gateway functionalities of the given node are redundant with respect to one or more other nodes from among the plurality of nodes, based on the real-time intra-domain topology of the multiple intra-domains and the real-time inter-domain topology of the multiple inter-domains. The method additionally includes dynamically assigning the given node as a gateway or a non-gateway by respectively turning on or turning off the particular set of gateway functionalities of the given node when the particular set of gateway functionalities of the given node are respectively determined to be non-redundant or redundant with respect to the one or more other nodes.
These and other features and advantages will become apparent from the following detailed description of illustrative embodiments thereof, which is to be read in connection with the accompanying drawings.
The disclosure will provide details in the following description of preferred embodiments with reference to the following figures wherein:
,
, Δ) for a given formula F, according to an embodiment of the present principles;
As noted above, the present principles are directed to assigning gateways for heterogeneous wireless mobile networks.
In general, any static gateway assignment is bound to suffer from the above described connectivity problem in dynamic MANETs. At a high level, there are two options to overcome this problem as follows: (1) make all nodes as gateways so that every node can handle inter-partition traffic; and (2) dynamically assign the role of gateway to suitable nodes that can make inter-partition connections as topology changes. The first approach is not very attractive because gateway nodes typically consume more power for multi-protocol processing and will generate more control traffic for inter-domain route and policy update. Thus, the present principles are directed to the second option.
In particular, the present principles aim to provide distributed mechanisms to elect gateways adaptively and optimally. By optimal, we mean to elect a minimal number of nodes to become gateways while all network partitions (i.e., represented as sub-graphs) are connected. To solve this problem, we formulated a novel graph optimization problem which called Minimal Gateway Assignment Problem, and proved its NP-completeness. We then designed efficient algorithms to solve this problem. We first designed centralized algorithms assuming full topology information and show that it has a theoretical approximation bound. We then designed distributed algorithms with various degrees of assumptions on the level of cooperation between domains and the availability of full topology information, and prove the correctness of the proposed distributed algorithms.
In one or more embodiments, we assume that routing information is sent to an external computation unit through a separate control channel (e.g., 3G, WiMax, etc.), or a subset (or all) of the nodes in the network can obtain the entire network topology information. Then we can apply the centralized algorithm to compute gateway assignment. We provide two centralized mechanisms with different computation complexity and performance bound in terms of number of gateways.
(i) SimpCen: In each step a pair of inter-partition neighbors is put into gateway assignment when such assignment can improve the connectivity at the partition-level topology.
(ii) Cen: To improve the performance of SimpCen, a heuristic is used to check thoroughly to make sure that the selected nodes as gateways will always provide the largest decrease in the number of disjoint components with the smallest number of gateway assignments in each step.
Distributed Mechanisms
In this case, we do not assume global network topology knowledge. Nodes exchange routing information with their neighbors, and progressively propagate the information throughout the network. When the nodes receive the routing information, the nodes determine their roles in the network locally, and switch on/off their gateway functionalities.
We disclose three mechanisms, with different level of cooperation and shared information. Regarding the levels of cooperation, we provide two illustrative schemes of cooperation, namely a tightly cooperative scheme and a loosely cooperative scheme. In the “tightly cooperative” scheme, nodes in one domain can enforce the decisions of nodes in other domains in order to achieve a better decision. On the other hand, the loosely cooperative algorithm will only use the other domain decisions as reference. Regarding the level of shared information, we have two levels of shared information, namely full information and partial information. Full information means a node can make decisions only if it has the full topology information. Partial information means a node can make decisions even it only has partial information.
(i) DIS-Tight: Tightly Cooperative with Full Topology Information. At each step, a node is selected randomly (e.g., by using a back-off timer) to make a decision. The node will decide to activate its gateway functionality base on the current topology information. If the node decides to activate its gateway functionality, then its inter-partition neighbors will also activate their gateway functionality
(ii) DIS-Loose: Loosely Cooperative with Full Topology Information. A node can decide if it wants to become an active gateway based on the number of gateways in its neighborhood. The idea behind this scheme is the following: if a node has a greater number of inter-partition neighbors, the chance to reduce the number of disconnected components in the partition-level graph will be higher if the node is a gateway. That is, for example, if a Node A has more neighbors than a Node B, then the chance that Node A can build a communication link with other gateways is higher than for Node B. Nodes do not enforce their inter-partition neighbors to activate their gateway functionality.
(iii) DIS-Local: Tightly Cooperative Algorithms with Partial Information. In some scenarios, the network can be a relatively large graph such that it may take a long time to propagate the assignment decision throughout the network. To address this issue, a node is allowed to make decisions even when the node has partial information. In an embodiment, nodes only collect information from their 1-hop neighboring partitions. Of course, other hop distances can also be used, while maintaining the spirit of the present principles.
TABLE 1 shows some common notations used herein.
Moreover, as used herein, the word “domain” refers to a logical concept determined by the organization that the network nodes belong to, whereas “partition” is a physical concept determined by the connectivity. For example a single domain owned by the RED CROSS can be divided into multiple partitions. In this regard, it will be more precise to call our problem an inter-partition connectivity problem since a MANET domain can be partitioned into multiple sub-networks. However, we use “domain” and “partition” interchangeably when there is no ambiguity.
Inter-domain refers to items (e.g., topology, networking, interactions, information, etc.) that relate to different domains. Intra-domain refers to items (e.g., topology, networking, interactions, information, etc.) that relate to the same (single) domain. Multi-domain refers to having and/or otherwise involving more than one domain. Inter-partition neighbors refers to neighbors which do not belong to the domain having a node under current consideration with respect to assigning that node as a gateway. Real-time intra-domain topology refers to an essentially real-time representation of the topology of one or more intra-domains. Real-time inter-domain topology refers to an essentially real-time representation of the topology of one or more inter-domains.
Thus, the present principles provide multiple mechanisms to elect gateways dynamically and optimally. By dynamic, we mean to elect gateways according to runtime network topology that changes over time. By optimal, we mean to elect a minimal number of nodes to become gateways while all network partitions are connected. Multiple mechanisms with different level of coordination, cooperation and shared information are disclosed.
As will be appreciated by one skilled in the art, aspects of the present invention may be embodied as a system, method or computer program product. Accordingly, aspects of the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment (including firmware, resident software, micro-code, etc.) or an embodiment combining software and hardware aspects that may all generally be referred to herein as a “circuit,” “module” or “system.” Furthermore, aspects of the present invention may take the form of a computer program product embodied in one or more computer readable medium(s) having computer readable program code embodied thereon.
Any combination of one or more computer readable medium(s) may be utilized. The computer readable medium may be a computer readable signal medium or a computer readable storage medium. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the context of this document, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device.
A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.
Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for aspects of the present invention may be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider).
Aspects of the present invention are described below with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other devices to function in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.
The computer program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks.
The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
A display device 516 is operatively coupled to system bus 504 by display adapter 510. A disk storage device (e.g., a magnetic or optical disk storage device) 518 is operatively coupled to system bus 504 by I/O adapter 512.
A mouse 520 and keyboard 522 are operatively coupled to system bus 504 by user interface adapter 514. The mouse 520 and keyboard 522 are used to input and output information to and from system 500.
A (digital and/or analog) modem 596 is operatively coupled to system bus 504 by network adapter 598.
Of course, the processing system 500 may also include other elements (not shown), including, but not limited to, a sound adapter and corresponding speaker(s), and so forth, as readily contemplated by one of skill in the art.
Minimal Gateway Assignment Problem
We now formally formulate the gateway assignment to support interoperation subject to connectivity constraint.
First, consider a set of potential gateways that belong to different partitions. We suppose that a gateway can belong to only one partition. A pair of neighboring gateways can act for a bridge for the nodes in their respective partitions.
Given a connected graph =(
, ε) as the topology of gateways in multi-partition MANETs, where each node n∈
is a potential gateway. Then we partition the set
as
, which represents the collection of disjoint connected subgraphs of
(i.e., the set of partitions), such that
satisfies the following constraints:
(1) (Connectedness): For each subgraph where i⊂
i, is a subgraph of
and
i is a connected graph.
(2) (Disjointness): For any pair of subgraphs 1=
i, ε),
(
j, εj), we have
i∩
=Ø, and
i=
Namely, we assume that the gateways in each domain are connected (i.e., by direct connections or indirect connections through other non-gateway nodes). We note that an administrative domain that is partitioned into multiple sub-networks without intra-domain connectivity will be regarded as multiple partitions.
Given a subset ⊂
, we define the partition-level graph with each node as a partition in
as
dm[
]=(
dm(
)), where the set of links
dm(
) is defined as follows:
(Inter-partition Links): For a pair of distinct partitions ,
∈
, (
,
) ∈
dm(
) if there exist n, n′∈
, such that n∈
n′∈
, and (n, n′)∈ε.
⊂
, such that the partition-level graph
dm[
]=(
dm(
)) is connected. That is, a node is assigned as an (active) gateway, if it is in the assignment
.
Definition 1: (Minimal Gateway Assignment Optimization Problem, MGA()). Given a connected graph
=(
, ε), and a collection of disjoint connected subgraphs
of
, we find a gateway assignment
with the smallest size |
|.
Definition 2: (Minimal Gateway Assignment Decision Problem, MGA(, Δ)). Given a connected graph
=(
, ε), and a collection of disjoint connected subgraphs
of
, we decide if there exists a gateway assignment
, such that |
|≦Δ.
Theorem 1: Minimal gateway assignment decision problem MGA(, Δ) is NP-complete.
Proof: It is easy to show that the gateway assignment decision problem is in NP, by checking the connectivity of graph gw[
] and |
|Δ, for a given gateway assignment
.
To show MGA(, Δ) is NP-hard, we rely on a polynomial time reduction from the 3SAT problem.
Definition 3: (3SAT Problem) Consider a 3-CNF formula F that includes m clauses and h variables, i.e. F=c1c2
. . . cm, where each ci=yj
yj
yj
Given a 3-CNF formula F, we assume each clause does not include a literal and its complement (as this is trivially satisfiable). We construct a corresponding MGA(, Δ), such that F is satisfiable, if and only if MGA(
, Δ) is satisfiable.
First, we set =Ø, ε=Ø and
=Ø. For each literal xj,
, and create a subgraph
x
x
, such that
x
Next, for each clause ci=yjyj
yj
, and create a subgraph
c
. If yj
It is easy to see that the construction of () is polynomial in time.
, Δ) for a given formula F, according to an embodiment of the present principles. In the example of
(
x2
x3)
(x1
x2
x3). Moreover, the dashed circles indicate the nodes belonging to the same partition. The truth assignment is x1=0, x2=0, x3=1, which is depicted as a gateway assignment with the selected nodes indicated by a √ (check mark).
(If Part): We show if F is satisfiable, then MGA(, Δ) is satisfiable with a gateway assignment
such that |
|=Δ. First, we set
=Ø. Then for each clause ci, we add ci∈
. Next, for each variable
. Otherwise, if
. It is easy to see that the partition-level graph
dm[
]=(
dm (
)) is connected, and |
|=h+m.
(Only-if Part): We show if MGA(, A) is satisfiable, then F is satisfiable. Suppose
is the satisfiable gateway assignment. Since each ci is a partition with one single node, ci∈
. Hence, this takes up m nodes. Next, each pair of xj,
. This gives rise to a consistent assignment for each variable. Also, since the partition-level graph
dm[
]=(
dm(
)) is connected, each ci is connected to at least one xi or
Therefore, we show that MGA (, Δ) is NP-hard, because 3SAT problem is NP-complete.
Since MGA(, Δ) is NP-complete, the optimization problem MGA(
) is unlikely to be solvable in polynomial time. However, we next present a simple greedy algorithm that yields a constant approximation bound.
Centralized Gateway Assignment Algorithms
We first present SimpCen, a simple centralized algorithm for MGA() based on a greedy heuristic. In SimpCen, in each step, a pair of inter-partition neighbors is put into gateway assignment
when such assignment can improve the connectivity at the partition-level topology. We let Comp(
) be the number of disjoint components in a graph
.
,
), Output
← Ø
dm[
]) > Comp(
dm[
∪ {n,n′}]) then
←
∪ {n,n′}
It is easy to see that the runtime of SimpCen is O(|ε|·(||+|ε|v)), because the runtime to find out the number of components in the partition-level graph takes O(|
+|ε|) by using breadth-first search with hash tables to store the visited nodes and partitions.
Theorem 2: Given minimal gateway assignment problem MGA(), let
opt be the gateway assignment such that |
opt| is the smallest. Let
SimpCen be the gateway assignment output by SimpCen. Then, we have the following:
|SimpCen|≦|
opt|−2
That is, SimpCen is at least as good as the 2-approximation polynomial-time algorithm for the MGA() problem.
Proof: The proof is given in the Appendix. Now we show that the analysis in Theorem 2 is tight.
Theorem 3: For any constant ∈>0, there exists an instance on which SimpCen outputs SimpCen which |
SimpCen|>(2|
opt|−2)−∈.
Proof: The proof is given in the Appendix.
To improve the performance of SimpCen, a heuristic is to check thoroughly to make sure that the selected nodes as gateways will always give the largest decrease in the number of disjoint components with the smallest number of gateway assignment in each step. This gives the algorithm Cen.
For each node n∈, we denote the Nbita(n) as the intra-partition neighbors of n, and n is always connected to Nbita(n). The inter-partition neighbors of n is denoted as Nbitd(n):
Nbitd(n){n′|n′∈
\Nbita(n) and (n,n′)∈ε}
,
), Output
← Ø
\
do
dm[
]) > Comp(
dm[
∪ X(n)]) then
dm[
]) - Comp(
dm[
∪ X(n)])
* ← arg
W(n)
*|X(n)|
←
∪ X (n*)
In each iteration, Cen selects a node that needs a minimal number of new gateways that help to connect the maximum number of disjoint components. The value X(n) defined in line 4 and updated in line 7 refers to the set of new nodes needed to assign into (i.e., to become gateways) in order to reduce the number of disjoint components in
dm[
] and the value of W(n) defined in line 10 refers to the number of disjoint components that can be connected if X(n) are assigned into
. Random tie-breaking is applied when there are multiple n*.
Following a similar proof as the one in Theorem 2, it is easy to see that Cen is at least as good as 2-approximation, because Cen only assigns a gateway if the assignment reduces the number of disjoint components in dm. The runtime of Cen is O(|
|·|ε|·(|
|+|ε|)), since for each n we only calculate Comp
dm[
∪{n, n′}], for {n, n′}∈ε which is bounded by |E|.
is empty n6 is added into
because n6 makes the most number of new connected partitions among all ni∈
(i.e., W(n6)=3). After n6 is added, n1, n4 and n7 are also assigned into
because n1, n4, n7∈Nbitd(n6) (i.e., they are the inter-partition neighbors of n6). Notice that although n2∈Nbitd(n6), n2 is not assigned into
because Partition-A and Partition-B are already connected with n1, n6∈
. Finally, the algorithm terminates with |
cen|=4.
Distributed Gateway Assignment Algorithms
In the distributed algorithms, a node decides that if it becomes active gateway or not according to the connectivity information collected from its intra- and inter-partition neighbors. For simplicity, we assume the following:
(A1) Network topology changes at a slower rate compared to the convergence speed of the gateway assignment algorithm (i.e., the topology is stable during the execution of algorithm).
(A2) Each node can learn Nbita from the intra-domain routing protocols (e.g., Destination-Sequenced Distance-Vector Routing Protocol (DSDV), Optimized Link State Routing Protocol (OLSR)), and Nbitd by proper neighbor discovery mechanisms.
(A3) At any point in time, only one node makes the decision. This can be approximated by various mechanisms, e.g., exponential backoff timer. More discussion on this issue is set forth with respect to Remarks 1 and 4.
(A4) When node n assigns itself as a gateway, it will propagate the assignment decision throughout the entire network. Other nodes will progressively learn the t (i.e.,
at step t) by receiving such notifications. We will show how to relax this assumption hereinafter.
We present two classes of distributed algorithms herein, with full and partial topology information. In both cases, we assume that nodes agree to help each other and share their connectivity information. For the algorithms with full information, we further design two algorithms with different levels of cooperation among the nodes in different partitions, namely, tightly cooperative and loosely cooperative algorithms. We now describe them in detail.
Tightly Cooperative Algorithms with Full Topology Information
First, we describe a simple procedure to construct the weights W(n) and X(n) of Cen in a distributed fashion without the knowledge of the whole topology G. One main factor is the number of disjoint components in the partition-level graph, Comp(dm[
t]). Suppose each node only knows the inter-domain neighbors Nbitd and intra-domain neighbors Nbita. At each step, each node maintains a partial graph
dm[
t] by exchanging the list of connected partitions from the inter-domain neighbors and intra-domain neighbors. Progressively, a complete topology
dm[
t] can be obtained that requires the message complexity at most O(|ε|·|
|). Now we consider two algorithm designs.
(1) Naive Algorithm: One simple way to design a distributed algorithm is to randomly select a node n at one time instant and assign both n and the set of its inter-partition neighbors Nbitd(n) to be the set of gateways . One benefit of this simple approach is that the design of the distributed protocol will be simpler and the computation overhead is limited. This naive approach is used as a baseline for comparison hereinafter.
(2) Distributed Tightly Cooperative Algorithm: In this algorithm DIS-Tight, we assume that a suitable timeout mechanism is in place to coordinate the decision process among the nodes. Assuming that timer granularity is larger than the propagation time within a domain, only one node will make decision at a time. When the backoff timer expires at node n, node n then collects the W(n′) and X (n′) from all n′∈Nbita(n) (i.e., the intra-partition neighbors of n), and find out the optimal node to become the gateway in (n), where
(n) is defined as the partition that node n belongs to. If node n is not the optimal node, it starts the timer again and waits for the next turn. Otherwise, node n assigns itself as a gateway and requests X (n) to become gateways as well, and broadcast the decision. Note that although DIS-Tight is triggered by an exponential backoff timer, the process of collecting information for partition-level graph
dm[
t] is assumed to run in the background triggered by the broadcast events.
Theorem 4: The number of gateways assigned by DIS-Tight, |DIS-Tight|, bounded by 2|
opt| where
opt is the gateway assignment such that |
opt| is the smallest. That is, DIS-Tight is a 2-approximation.
* ← arg maxn∈Nbita(self) W(n)
|X(n)|
t ← t−1 ∪ {n*} ∪ X(n*)
t and
dm[
t]
t and
dm[
t] then
t and
dm[
t]
dm[
t]) > Comp(
dm[
t ∪ X(n)]) then
dm[
t]) - Comp(
dm[
t ∪ {n}∪ X(n)])
Proof: The proof is given in the Appendix.
Theorem 5: The distributed process of DIS-Tight will terminate at a correct state where all partitions are connected in finite steps.
Proof: The proof is given in the Appendix.
DIS-Tight|=5.
Remark 1: (Exclusive execution of DIS-Tight). In general, there is no efficient mechanism to ensure only one node to execute a distributed procedure in the large MANETs environment because that requires either complete locking of all nodes or synchronization of timers. One practical way to “roughly” achieve this condition is to utilize an exponential backoff timer, which has been used extensively in the IEEE 802.11 and the Ethernet protocols, to reduce the probability of simultaneously triggering a distributed procedure at different nodes. We thus apply an exponential backoff mechanism in DIS-Tight.
Even if the timers expire simultaneously in DIS-Tight, it does not affect the correctness of DIS-Tight because DIS-Tight only assigns nodes to become gateways but not disables gateways. When more nodes become gateways, the partition-level graph will eventually reduce to one single connected component. However, the result in Theorem 4 may not hold because executes DIS-Tight simultaneously may lead to a suboptimal solution. We leave the users of DIS-Tight to decide whether (i) the number of gateways in the network, or (ii) communication overhead for locking or synchronization, is more important in operation.
Remark 2: (When to execute DIS-Tight). Ideally, DIS-Tight should be executed whenever the topology is changed (i.e., resetting the gateway assignment). On the other hand, executing DIS-Tight will incur communication overhead and extensive changes to the active gateway set. With practical models of link characteristics such as the expected lifetime and change rate, one can further optimize when to execute DIS-Tight (e.g. aggregate multiple events of topology changes). In practice, each partition can execute DIS-Tight (i) periodically at certain time interval, or (ii) when partition-level topology changes have been detected.
Loosely Cooperative Algorithms with Full Topology Information
The assumption of DIS-Tight that a node in other partition will always follow the request from a node making a decision may not be valid. In general, nodes in each partition should make its own decision based on its energy level and other considerations such as its mobility and mission objectives. Therefore we consider a case of “loosely” cooperative model where the nodes in each partition are willing to optimize the global objective, but without enforcement from the outside partition. In this case, a node n can decide if it wants to become an active gateway based on the number of gateways in Nbitd(n) and the size of Nbitd(n). The main idea behind this algorithm is fairly simple: if node n has more number of inter-partition neighbors and gateways, the chance to reduce the number disconnected components in the partition-level graph will be higher if node n is a gateway. We consider two versions of loosely cooperative algorithms.
(1) Naive Algorithm: Similar to the naive algorithm herein, when the timer of node n is expired, it assigns itself as an active gateway. The only difference between the loosely cooperative and tightly cooperative version is that in the loosely cooperative version, node n does not assign its inter-partition neighbors as gateways. This naive algorithm is used as a baseline for comparison hereinafter.
* ← arg maxn∈Nbita(self) WS (n)
NG(n))
t ← t−1 ∪ {n*}
t and
dm[
t]
Nbitd(NGita(self)) then
t and
dm[
t] then
t and
dm[
t]
dm[
t]) - Comp(
dm[
t ∪ {n}])
(2) Distributed Loosely Cooperative Algorithm: Similar to the tightly cooperative version, when node n timer expires, it collects the WS (n′), NG (n′) from all n′∈Nbita(n), and finds out the optimal node to become the gateway in (n). Different from Cen and DIS-Tight, WS(n) is defined as follows:
WS(n) Comp(
dm[
t])−Comp(
dm[
t∪{n}])
which means the reduction in the number of disconnected components in dm(n) when only node n is assigned to
t. NGita( ), and NGitd(n) are defined as follows:
NGita(n){n′|n′∈
∩Nbita(n)}
NGitd(n){n′|n′∈
∩Nbitd(n)}
which means the set of gateways in the intra-partition and inter-partition neighbors of node n respectively.
Theorem 6: The distributed process of DIS-Loose will terminate at a correct state where all partitions are connected in finite steps.
Step 1: n1 decides that it is a gateway because Nbitd=2>0, which is the best among the unassigned nodes in Partition-A
Step 2: n2 decides that it is a gateway because Nbitd(n2)=2>0, which is the best among the unassigned nodes in Partition-A
Step 3: n3 decides that it is a gateway because WS(n3)=1>0, which is the best among the unassigned nodes in Partition-B
Step 4: n4 decides that it is a gateway because Nbitd(n4)=2>0, which is the best among the unassigned nodes in Partition-B
Step 5: n5 decides that it is not a gateway because Nbitd(n5)⊂Nbitd(n6) and the optimal node in Partition-C is n6. After n6 is selected as a gateway, n5 stops its timer
Step 6: n6 decides that it is a gateway because WS(n6)=2>0, which is the best among the unassigned nodes in Partition-C
Step 7: n7 decides that it is a gateway because WS(n7)=2>0
Step 8: n8 decides that it is not a gateway because Nbitd(n8)⊂Nbitd(n7)
Finally, the process terminates with |DIS-Loose|=6.
Remark 3: (Leader selection in each partition). In DIS-Tight and DIS-Loose, when node n discovers that it is not the optimal node for (n), it will sleep and wait for the optimal node to wakeup. We argue that node n should not notify the optimal node in its partition for faster convergence, because if that happens a partition with more nodes will assign gateways at a faster rate than a partition with less nodes. One way to speed up the convergence is to elect a leader for each partition. When the timer at a leader expires, the leader collects the partition connectivity information and assigns the optimal node in its partition to become a gateway. We note that electing a leader for each partition only improves the convergence time but does not reduce the number of gateway assigned by our proposed algorithms.
Tightly Cooperative Algorithms with Partial Information
In some scenarios, can be a relatively large graph such that it may take long time to propagate the assignment decision throughout the network. To address this issue, we relax the assumption (A4) such that a node n can make decisions even n only has partial information. In the relaxed assumption (A4), we assume that nodes only collect information from
(n)'s 1-hop neighboring partitions.
(self)) changed then
(n)
(NGitd(NGita(self))) then
(NGitd(n))|
(Nitd(self))
(NGitd(n)) then
(NGitd(self)) already covered by others
(Nitd(self)
(NGitd(NGita(self))) then
(self)
(self)
(self)
Algorithm 5 shows the pseudocode of the gateway assignment algorithm, DIS-Local. The basic idea of DIS-Local is that each partition tries to establish a gateway pair to its 1-hop neighboring partitions. Different from DIS-Tight, a partition, (n), in DIS-Local tries to form a 1-level tree root at
(n) in the partition-level graph. The resulting mesh then will have a richer set of edges in
dm[
], compared to the one generated by DIS-Tight.
Similar to the previous distributed algorithms, DIS-Local relies on the exponential backoff timer to achieve exclusive execution. For a node n, when the 1-hop neighborhood of n is changed, n will broadcast such changes to its intra-partition neighbors (lines 29-31). Once the nodes in (n) receive that information, they decide which nodes will become gateways. Node n will assign itself as a gateway if one of n's neighboring partitions is not connected by the other gateways in
(n) (lines 5-8). On the other hand, n will not assign itself as a gateway if all n's neighboring partitions is connected by the other gateways in
(n) (lines 17-20). A simple heuristic to keep nodes with larger number of connected partitions is applied to reduce the number of gateways in
(n) (lines 11-16).
Theorem 7: The number of gateways assigned by DIS-Local, |DIS-Local|, is bounded by |
|(|
|−1).
Lemma 1: The worse case performance ratio of DIS-Local over Cen is
Theorem 8: DIS-Local guarantees a partition is connected to its 1-hop neighboring partitions if there is no simultaneous execution.
The proofs of Theorem 7, Lemma 1, and Theorem 8 are given in the Appendix.
Remark 4: (Exclusive execution of DIS-Local). Similar to Remark 1 herein, the exponential backoff cannot completely guarantee exclusive execution. When multiple nodes are executing DIS-Local simultaneously, a partition can be disconnected from its 1-hop neighboring partitions. We argue that this will only slow down the convergence speed of DIS-Local because the disconnection will be detected eventually as far as the exponential backoff timer generates a exclusive execution sequence. To guarantee connectivity, when node n wants to make a decision, n needs to apply appropriate locks on (i) the nodes in other partitions (line 7), and (ii) the nodes in (n).
Discussion
Weighted Gateway Assignment
In some scenarios, we may want to optimize for other performance metrics than the number of gateways. For example, it is reasonable to assume that gateway election can be based on node capability, remaining energy level, security level, and stationarity (that the node will not disappear as easily as other nodes). The problem then becomes a node-weighted gateway assignment problem. We believe that the proposed centralized algorithms can be easily extended to support the weighted version of the problem. For example, in each iteration of Cen, instead of selecting a node which requires the minimal number of new gateways (i.e., |X(n)|), we can select a node which optimizes the total weight of the new gateways. Similar extension can be made to the distributed algorithms.
Resilient Gateway Assignment
In general, there is a trade-off between the number of gateways and the resilience of inter-partition connectivity. If there are only a few inter-partition links, then they may be easy to disconnect due to mobility. We can address the problem by provision more redundant links, say, by guaranteeing connectivity across partitions. Developing practical distributed algorithms to guarantee such condition is an interesting research topic, which will be pursued in the future.
Thus, herein we addressed the gateway assignment problem for enabling interoperations among heterogeneous MANETs. We formulated the problem as an optimization problem which assigns minimal number of gateways to support full connectivity among different network partitions, and proved that this problem is NP-complete. We also designed centralized algorithm that guarantees a tight 2-approximation bound and have shown that empirically it performs very close to the optimal solution (within 4% compared to the optimal). We studied two important design choices to develop the distributed versions: (i) level of cooperation (i.e., tightly or loosely cooperative), and (ii) level of shared information (i.e., full or partial topology information). Cooperation across domains is a very important factor in order to obtain good active gateway assignment. The proposed algorithms have been shown to converge to a correct state in finite time and have been shown to work well in medium size inter-domain MANET scenarios.
Appendix
Proof of Theorem 2
Proof: First, note that |opt|≧|
|. Thus, it is sufficient to show that |
SimpCen|≦2|
|−2.
Assume that SimpCen terminates in k steps. At step l, we denote blas the number of new nodes that is assigned into . Also, we denote the number of disjoint components at step l as Comp(
dm[
l]). Note that Comp(
dm[
0])=|
| and Comp(
dm[
k])=1. Also note that k≦|
|−1 because SimpCen assigned {n, n′} into
only if the assignment reduces the number of disjoint components, and there are at most |
| disjoint components. Then we have the following:
Proof of Theorem 3
Proof: Consider the following example. There are 3 partitions, Partition A contains n1, Partition B contains n2, n3, Partition C contains n4, n5, and {n1, n2}, {n3, n5}, {n1, n4}∈ε. The optimal solution is opt={n1, n2, n4}. However, depends on the ordering of edges in ε, SimpCen can first select {n1, n2} then {n3, n5} since both edges will reduce the number of connected components.
Proof of Theorem 4
Proof: For DIS-Tight to assign node n to become a gateway, n must have W(n)>0. As only one node is allowed to make a decision at each step, assigning node n as gateway implies the assignment will reduce the number of disconnected components in dm[
t] by at least one. The rest of the proof then becomes similar to the one in Theorem 2 and we omit the details of the proof for brevity.
Proof of Theorem 5
Proof: We first show that DIS-Tight will terminate in finite steps. When node n makes a decision, n executes one of the following actions: (a) selects itself as a gateway in (n) and stop the timer, (b) stop the timer because of W(n*)=0 or (c) do nothing (e.g., let the optimal node in
(n) to assign itself) and wait for next timer expiration. Obviously action (a) and (b) will progressively lead the distributed process to terminate. Node n keeps repeating action (c) until it becomes the best unassigned node in
(n). Assuming node n ranks kth in
(n), node n will stop its timer after the first k nodes in
(n) are selected. As |
(n)| is finite, the distributed process will terminate in finite steps.
Next we show that the distributed process will terminate at a correct state by contradiction. Assume the process stop at a state where some partitions are not connected. This implies that all timers stop (i.e., the process is terminated) and W(n)>0 for some n∈\
. However, this cannot happen because n stops its timer only if W(n)=0 or n∈
.
Proof of Theorem 7
Proof: It is easy to see that |DIS-Local|≦|
|(|
|−1). Since each partition establishes a gateway pair with its 1-hop neighboring partitions, there will be at most
such gateway pairs. We omit the details for brevity.
Proof of Lemma 1
Proof: We proved that the performance of Cen is bounded by 2(||−1). The remaining details will be easy to construct and we skip that for brevity.
Proof of Theorem 8
Proof: We show this by contradiction. Assume a partition is not connected to its 1-hop neighboring partitions. This implies that DIS-Local disables some gateways, say wrong, incorrectly such that there are some 1-hop neighboring partitions can only be reached by
wrong. However, this cannot be true because DIS-Local only disables gateways when the 1-hop neighboring partitions are already connected by other gateways in
(n) (lines 13 and 18).
Having described preferred embodiments of a system and method (which are intended to be illustrative and not limiting), it is noted that modifications and variations can be made by persons skilled in the art in light of the above teachings. It is therefore to be understood that changes may be made in the particular embodiments disclosed which are within the scope of the invention as outlined by the appended claims. Having thus described aspects of the invention, with the details and particularity required by the patent laws, what is claimed and desired protected by Letters Patent is set forth in the appended claims.
This invention was made with Government support under Contract No.: W911NF-06-3-0001 awarded by the U.S. Army. The Government has certain rights in this invention.
Number | Name | Date | Kind |
---|---|---|---|
6744740 | Chen | Jun 2004 | B2 |
6980524 | Lu et al. | Dec 2005 | B1 |
20020181443 | Coffey et al. | Dec 2002 | A1 |
20030096576 | Salonidis et al. | May 2003 | A1 |
20050169238 | Yang et al. | Aug 2005 | A1 |
20060094426 | Donaho et al. | May 2006 | A1 |
20070047510 | Cho et al. | Mar 2007 | A1 |
20080200168 | Jiang | Aug 2008 | A1 |
20080253386 | Barum | Oct 2008 | A1 |
20090141653 | McNeill et al. | Jun 2009 | A1 |
20090147702 | Buddhikot et al. | Jun 2009 | A1 |
20090219834 | Babbar et al. | Sep 2009 | A1 |
20090238099 | Ahmavaara | Sep 2009 | A1 |
20090285126 | Lu et al. | Nov 2009 | A1 |
Entry |
---|
Baker, D.J.; Wieselthier, J.E.; Ephremides, Anthony; McGregor, D.N., “Distributed Network Reconfiguration in Response to Jamming at HF,” Military Communications Conference—Progress in Spread Spectrum Communications, 1982. MILCOM 1982. IEEE , vol. 1, No., pp. 23.2-1,23.2-7, Oct. 17-20, 1982. |
Christian Frank and Kay Romer. 2005. Algorithms for generic role assignment in wireless sensor networks. In Proceedings of the 3rd international conference on Embedded networked sensor systems (SenSys '05). ACM, New York, NY, USA, 230-242. DOI=10.1145/1098918.1098944 http://doi.acm.org/10.1145/1098918.1098944. |
Baker, D.J.; Ephremides, Anthony, “The Architectural Organization of a Mobile Radio Network via a Distributed Algorithm,” Communications, IEEE Transactions on , vol. 29, No. 11, pp. 1694,1701, Nov. 1981. |
Chau, C., et al. “Inter-Domain Routing for Mobile Ad Hoc Networks” MobiArch '08. Aug. 2008. pp. 61-66. |
Chen, Y., et al. “Clustering Algorithms for Ad Hoc Wireless Networks” Ad Hoc and Sensor Networks. 2004. pp. 1-16. |
Crowcroft, J., et al. “Plutarch: An Argument for Network Pluralism” ACM SIGCOMM 2003. Aug. 2003. (9 pages). |
Ma, W., et al. “Comparisons of Inter-Domain Routing Schemes for Heterogeneous Ad Hoc Networks” 2005 International Conference on a World of Wireless, Mobile and Multimedia Networks (WOWMOM 2005). Jun. 2005. (10 pages). |
Ramasubramanian, V., et al. “Sharp: A Hybrid Adaptive Routing Protocol for Mobile Ad Hoc Networks” MobiHoc '03. Jun. 2003. pp. 303-314. |
Schmid, S., et al. “Turfnet: An Architecture for Dynamically Composable Networks” Proceedings of 1st IFIP International Workshop on Autonomic Communication (WAC 2004). Oct. 2004. (21 pages). |
Number | Date | Country | |
---|---|---|---|
20120294187 A1 | Nov 2012 | US |