Patent Grant
7420952
The present invention describes an adaptive software layer for a distributed set of wireless communication devices that communicate with each other in a wireless network. The software control layer addresses low latency requirements (for applications such as voice) and high throughput requirements (for applications involving data transfer). One embodiment of the present invention provides the software control for wireless (devices, such as, but not limited to Access Points, employed in a convergent enterprise network supporting voice, video and data. A topical application of the software control layer is a home or personal networking environment (PAN) using Ultra Wide Band or Wi-Fi as the communications medium. Another topical application of the adaptive software control layer is extending wireless communication range using mesh networks for Metropolitan Area Networks (MAN). Lastly, the software control layer is also relevant to both home and enterprise Wireless Local Area Networks (WLANS),
There is increasing demand within the enterprise, the home and within cities to employ one wireless network to support both voice, video and data traffic. Currently, the “voice” network, e.g. the telephone system, is separate from the “data” network e.g. Internet connectivity and access to enterprise data over a Local Area Network (LAN). Convergence is, as the name implies, the ability to converge these two networks into one network, centrally managed by one access server servicing a network of relay and leaf nodes.
The challenge lies in providing—within the same wireless network—the ability to address potentially conflicting latency and throughput needs of diverse applications. For example, voice needs to be transmitted with low delay (latency). Occasionally lost voice packets, while undesirable, is not fatal for voice transmissions. Conversely, data transmissions mandate delivery of all packets and while low latency is desirable it is not essential. In essence transmission across the wireless network should ideally be driven by the needs of the application. The table below lists some types of applications and their latency requirements.
A wireless network provides service to a diverse set of applications, with varied latency requirements. One approach to make dumb wireless devices, that are nodes of the network, more application aware by implementing QoS (Quality of Service) reservation schemes dictated by the application server.
Changing the rate the queue is serviced can also be accomplished by specialized communications between wireless communication devices such as Access Point (AP) nodes and the access server to ensure that voice and data, for example, are serviced at different time intervals. Unfortunately, this adversely affects scalability and redundancy of the system: the access server is now micromanaging the network and has become a single point of failure. A paramount concern of any network is distributed control, especially a network handling voice, video and data.
Another shortcoming of a centralized approach—central control and central execution—is the cost of maintaining a central control point with all intelligence and control at one location and dumb communication devices distributed in the enterprise. The cost of the central control point is high, and the dumb access points are not any less expensive than smart access points—since the smarts is in the software. Thus a distributed approach is far less expensive—. In addition to being more cost effective a distributed approach is more fault tolerant and has built in fail-safe redundancy. The only way to get redundancy out of centralized approaches is to buy multiple central control points—an expensive approach.
Building a reliable wireless network comes with other constraints specific to wireless. Some routing paths may be best for voice and video, others for data. In Ethernet applications separate routing paths is easily accomplished. But in a wireless network, operating over radio, the cost of associating and disassociating with a relay node—to switch to new routing paths—is prohibitive. Multiple radios, supporting separate voice and data channels is possible but expensive. It is preferable, therefore, if each AP node can support both voice and data transmissions with a one “channel”.
Furthermore, mesh networks have been around for years now, the Internet being an excellent example. Routers exchange information about each other and build up their routing tables, and use the entries in the routing table to make routing decisions. Although they work, these algorithms are sub-optimal at best and are more oriented towards wired or wire-like interfaces, which are exclusive “non-shared” communication mediums. Wireless Personal Area Networks (WPANs) pose an entirely different set of challenges for AD-HOC networks because of the following reasons: shared non exclusive medium with finite communication channels; dynamically changing environment; shorter distances; used by resource constrained low power devices. There is therefore a need for an approach to solving these sets of challenges, using a low footprint distributed adaptive control layer that is aware of the above set of problems.
Accordingly, there is a need for, and an objective of the present invention, to develop an adaptive wireless network, based on “smart” communication devices such as Access Points (AP) that provide embedded intelligence at the edge of the network, are application aware and provide cost effective distributed sensing and control of the network. An additional objective of this invention is to allow the characteristics of the network to be set by a centralized access server, which can thus “tune” the character of the network to be anything between the two extremes of low latency to high throughput, based on the needs of applications running in the enterprise. The invention also supports the possibility of running multiple types of networks supporting anything between the two extremes of low latency and high throughput, using multiple radios at each node for each distinct type of network.
As an illustration of central control but distributed intelligence, consider
Since the signal strength varies inversely with the distance, it may be advantageous, from the perspective of better signal strength and overall better throughput for some AP nodes to connect to an intermediate AP rather than to the root, as shown in
The number 44.1 is a measure of the throughput computed based on the look up table shown in
While the throughput increased to 62.2, the tradeoff is more hops, resulting in a loss of latency for higher throughput. In
The objective of this invention is to allow the Access Server to set some latency/throughput constraints that causes each AP node to change their relationships to each other and consequently the character of the network. Control parameters, set by an access server can then tune the wireless network to provide a mix between the two extremes of max throughput and low latency. As shown in
The approach taken to modify the network is completely decentralized—the changes in the network take place with algorithms running in each AP node. The Access Server does not change the characteristics of each node, it simply sets the parameters governing the characteristic of the network—and let the AP nodes reconfigure their relationships to meet the objectives set by the Access Server. Thus the Access Server can control the behavior of the network without necessarily controlling the behavior of each node of the network. Benefits of this approach include a highly scaleable, redundant wireless network. Some other benefits include:
Uses of a self configuring application aware wireless network range from providing voice/data access to warehouses, factory floors, communications with process control equipment to home networking applications involving voice/data/video streaming. Some applications under consideration include:
Furthermore, another object of the embodiment of the present invention, is to solve the problems associated with ad-hoc wireless personal area networks by using a low footprint distributed adaptive control layer with algorithm that is aware of such problems (e.g. shared non exclusive medium with finite communication channels; dynamically changing environment; shorter distances; used by resource constrained low power devices).
These and other embodiments of the present invention are further made apparent, in the remainder of the present document, to those of ordinary skill in the art.
In order to more fully describe embodiments of the present invention, reference is made to the accompanying drawings. These drawings are not to be considered limitations in the scope of the invention, but are merely illustrative.
The description above and below and the drawings of the present document focus on one or more currently preferred embodiments of the present invention and also describe some exemplary optional features and/or alternative embodiments. The description and drawings are for the purpose of illustration and not limitation. Those of ordinary skill in the art would recognize variations, modifications, and alternatives. Such variations, modifications, and alternatives are also within the scope of the present invention. Section titles are terse and are for convenience only.
The object of this invention is a new type of wireless AP nodes that:
Each AP Node is implemented as a self-contained embedded system, with all algorithms resident in its operating system. The normal day-to-day functioning of the AP node is based entirely on resident control algorithms. Upgrades are possible through a communications interface described later.
Description of System Components
There are three typical components of the system proposed. In
In one implementation of the invention, the root node (20) acts as the interface between the wireless communication devices (30) and the Ethernet. All wireless devices (30) communicate to the Ethernet through a root node (20), which is has a radio interface and an Ethernet link.
In that implementation of the invention, other wireless communications devices or AP nodes (30) have two radios: one to communicate with its clients which includes wireless devices such as laptops, VoIP wireless phones etc. Clients to one AP node (30) also include other AP nodes (30) connecting to it. In
In an alternate implementation of the invention, all Wireless AP Nodes are roots—they are all wired. In this case the network is not a mesh network and redundancy is dependant on having a large number of nearby AP nodes. However there is still communication between the nodes so load balancing and wireless switching, as described later, is supported.
Description of Variables Used by Algorithm
CO: Cost to connect. When a node is congested it raises the cost of connectivity for a prospective child, thereby dissuading new connections till the congestion is alleviated. The cost to connect is a number transmitted to the child node. It is related to the level of congestion
LC Load from Children: This is the sum total of all traffic being passed on from children to their parent for retransmission to the parent's parent. In
LL Local Load: In addition to servicing children, each node may also be servicing individual clients that comprise their local load. For example, Node 002 services three child nodes. But in addition, there could be laptops in its vicinity that connect to it directly. They constitute a local load.
LS Local Signal Strength: is related to signal strength that a node sees from its parent. It is actually the throughput expected based on the actual signal strength seen by the radio and is computed based on a look up table as shown in
Since there is no central point of control in a distributed system, the same algorithms, running in every node, must determine what is best, based on the information it received from the Access Server and other nearby nodes. Much of this relates to selecting correct “route” or path to the root node. As an illustration, in
Assuming for the present, that the Access Server wishes the network to have the maximum throughput. Then, if each node independently makes the best selection of its parent—to maximize throughput—then a question arises of whether one can be assured that the network as a whole is running as “best” as possible.
To answer this, consider the network in
Since GT is product function of the LS times the GT of the potential node. Node 005 would have examined all potential parent nodes before selecting node 002. Similarly Node 002 has chosen Node 000. Other nodes would yield a lower GT. Thus, since each node is making a parent selection based on the “best” throughput, the throughput of the network as a whole is also maximized.
Thus, each node, starting from those closest to the root and spreading outwards maximizes its GT based on products related to the GT at each previous node, the overall throughput of the system is the sum of all individual throughputs, which have been maximized by the selection algorithm.
The implementation steps taken by the selection algorithm are:
Based on these two pieces of information, collected for all nearby AP nodes, the node selects one parent node that satisfies the requirements of low latency or high throughput. If low latency is the issue then a short routing path is desirable: In
At the end of step 4, the global throughput—computed as a product of the local signal strength to that potential parent node and the GT of the potential parent node—is computed and compared for each potential parent in the list of nearby nodes. The one with the highest throughput wins.
Controlling Network Latency/Throughput
The section on the selection of the parent assumed that only maximizing throughput was the sole objective. Often there is a tradeoff between low latency and high throughput as evidenced in
The aforementioned describes the algorithm for maximized throughput. For lowest latency, the choice of parent is restricted to the parent with the highest throughput with an upper bound on the number of hops the parent is away from the root. There are two ways in which the Access Server can control the latency of the network:
If the latency loss threshold is set high, the choices a node in selecting its parent is restricted, to nodes closer to the root, with shorter route paths. In contrast to the cutoff approach this approach is more forgiving: Selecting a longer path is allowed if throughput gains in choosing a longer routing path offset increased latency. In terms of the routing algorithm, this translates to computing the throughput for all nearby nodes.
Reference is now made to re-examining the parent selection process with latency restrictions in place. With reference to
Combinations of both restrictions, based on the parameters set, result in networks that address both latency and throughput requirements. This was shown in
Automatic Load Balancing
Described thus far is how the parent selection process takes into account latency/throughput criteria set by the access server. However, one must also take into account how the system behaves under load, when the load increases at one node, causing congestion.
Since this is a distributed system, each node is responsible for selecting a parent that can service it satisfactorily—it is not part of a congested route. During the selection process, the connect cost associated with selecting a new parent is supplied by the parent. Thus a congested route will have a higher connect cost than a less congested route —and a lower throughput. The routing algorithm selects the parent with the highest throughput. As its connect cost increases a congested parent is increasing less attractive.
In
Increasing the cost of connectivity acts as an incentive for nodes to find other routes, but does not prevent a child node from continuing its association. This ensures that all child nodes are serviced always. Additionally, the cost of connectivity is increased only until all child nodes that have the option to leave have left—for example, in
As the load is balanced, the congestion is reduced and the cost of connectivity is gradually reduced, enabling the child nodes that left to return. If this is not done, then nodes leaving one parent would contribute to congestion elsewhere, resulting in increased cost of connectivity elsewhere and system instability.
One characteristic of a mesh network is the ability for nodes to select alternate routes, in case one node fails or becomes congested. As shown in
In
It is desirable to configure the network to ensure all nodes have alternate paths. This is achieved by increasing the number of nodes connecting to the root. The access server can force this by increasing the latency cost factor, resulting in nodes that can connect to the root directly to do so rather than through another node closer to them to the root. This was described earlier as depicted in
By controlling the latency cost factor, or the upper bound of the max hops, the access server can change the configuration of the network resulting in a higher redundancy of the system and less likelihood of load congestion hot spots.
Real World Constraints to Load Balancing Algorithm
Implementation of the load balancing algorithm on the wireless devices, required modifications to the connect cost algorithms based on real world constraints. Wireless devices communicating with devices running the load balancing software may not all be running the same software. As an example, consider the case where the load balancing software is loaded on wireless Access Points but not on laptops communicating with the access points. Clearly, the laptop has no way of knowing that the connect cost has increased and therefore will continue to “stick” to the access point.
The load balancing algorithm has therefore been modified to work where there is no communication regarding connect cost by the following approach: When the load exceeds a cut off threshold, the Access Point (or other wireless device performing load balancing) will drop its signal strength to the lowest possible—thereby dissuading clients such as laptops from associating with it and encouraging them to seek another association with a higher signal strength access point.
Since the laptops seek the access point with the highest signal strength, this is a necessary but not sufficient cause for a re-association: some laptops may continue to “stick” to the access point, due to proximity or a sticky algorithm. The Access Point must therefore forcibly disassociate the laptop.
After disassociating all the stations that it needed to, in order to shed load, the access point can gradually increase its signal strength to attract back some the stations that it disassociated. Those that did not find other associations, will return, almost immediately after association, and the access point takes them back because despite the lowered signal strength, these devices have no place to go.
This load balancing algorithm has been implemented and demonstrated to shed load by moving one laptop from one root node to another when overloaded by two laptops on the same root node.
Monitoring the Health of the Network
By controlling the latency cost factor, or the upper bound of the max hops, the access server can change the configuration of the network resulting in a higher redundancy of the system and less likelihood of load congestion hot spots. In
Congestion in
Algorithms have been implemented that reside in the device and periodically check to see what potential associations are possible. If the number is reduced to one then a warning (shown in red) is forwarded to the Network Management System. The system administrator is then advised to add another root node to preserve the fail safe nature of the network. Note that this type of warning is coming from the edge device to the management system and without the algorithm in place, the management system would not know that a problem existed
Dynamic Channel Allocation and RF Interference Control
Managing the throughput of voice, video and data traffic in a wireless network is complicated by the nature of the wireless medium. Since wireless is a shared medium, all traffic can potentially interfere with other traffic on the same radio channel—at any point in time only one device can be active on any one given channel. This limits the bandwidth in cases where high bandwidth traffic (e.g. video) needs to be transported or when there are many devices on the network.
One solution is to allocate different channels to devices communicating on different portions of the network. For example, in
An algorithm to define what the best channel allocations should be between devices and their parents has been devised and shown in
Dynamic Channel Allocation and Seamless Roaming Requirements
A situation can occur when, as shown in
The algorithm implemented addresses the case where siblings of a multi layered wireless network are to be assigned the same channels. In
Protocols for sharing this information have been implemented and tested. Appendix A hereto describes the 802.11 Infrastructure Control Layer Protocol version 2.0. In addition, Appendix B hereto describes in another embodiment of the present invention, a distributed adaptive control algorithm for ad-hoc wireless personal area networks.
Note that seamless roaming requires that the Wireless AP shown in
Asynchronous Application Data Flow
Algorithms that show how data from nodes will flow to the root node have been modeled for both high throughput and for low latency requirements. High throughput data flow requirements are discussed first.
To service asynchronous applications each node services its children in a round robin manner, ensuring that all children are serviced in sequence. But to ensure that all children receive at least one service request, each recently serviced child must wait for at least another child to be serviced before it can be serviced again. Additionally, some child nodes servicing applications with higher priority will be serviced before others.
In one implementation of this algorithm, related to this invention, the priorities may be stored in the Access server and different applications fall into different priority buckets. By changing the priorities in the Access Server, applications with higher priority are serviced before other competing applications with a lower priority. Also with a priority bucket, applications with more data to transfer are serviced first.
In another implementation, the determination regarding which child to service next is based on which child needs servicing the most. This is determined by examining the load of each child, each time; the node services its children. It is done each time because:
The node then makes the decision to service the child with the highest need, in a priority bucket, provided it has not serviced that same child most recently. This is simply to avoid any one child from “hogging” all the attention.
This proprietary PCF (Point Control Function) implementation worked well for asynchronous applications, when compared to 802.11a standard DCF (Distributed Control Function) approach that was also implemented for benchmarking reasons as shown in
Isochronous Application Data Flow
Isochronous applications require more deterministic service intervals that are less sensitive to variations of load. The algorithm described above is not suitable when:
The algorithm to service Isochronous Application has also been implemented. In
Thus if the parent of the red service cycle (70) spends 10 ms with each child, it will revisit each child every 3*10=30 ms. Data from each child cannot then be retrieved at a rate faster than once every 30 ms.
Having retrieved the data, it will sit at the buffer of the parent, until the parent is serviced (the black (80) circle). Since there are 5 children in that service cycle, the service period is 5*10=50 ms.
Since both service cycles are running independently of each other with no synchronization, it is impossible to predict when the parent in either service cycle will service its children. It can be stated, however that each child in service cycle marked red (70) (005, 001, 008) will have data transferred to the root at best every 30 ms and at worst every 50 ms. In other words, in the isochronous network, the worst time interval is the maximum time period of all service cycles.
If it is assumed that, to ensure multiple routing paths, there are more nodes connected to the root, then the service cycle will be driven by the number of 1 hop nodes, in this case 5. Note that the network configuration is set for high throughput. In this configuration the worst service cycle is 5T. Ironically, the network configuration for a “low latency”. Isochronous network would have been 9T, will all nodes connected to the root. In other words, in the case of isochronous networks, the algorithm proposed provides a better service cycle and a better throughput. In general, splitting the number of nodes into two or more service cycles will improve the service cycle. The high throughput mode setting for the routing algorithm makes that happen naturally.
Internal Traffic Flow/Wireless equivalent of Switching
Referring again to
Traffic from Node 005 to Node 001 would logically travel to Parent 002 and then from 002 to 001. This affects the throughput of the entire system because the traffic is buffered in 002 and then retransmitted. If node 005 was aware of its siblings, within its wireless range, then Node 005 and Node 001 could communicate directly over wireless. In effect this would be a wireless equivalent of Ethernet switches.
This direct communication link between siblings (within range) increases throughput 100%. This is so because 2 transfers of data (Source node to parent and then Parent to destination node) are now reduced to 1 transfer (Source node to destination node).
In this embodiment, the algorithm has been implemented whereby traffic intended for a destination node is automatically sent to the destination node if it is within the family/BSS and within range. When the routing algorithm runs, the routing paths of each nearby node is recorded to determine the number of hops it is away from the root. From the routing paths, the list of siblings within range can be inferred—they all share the same parent in their routing paths. If data intended for these siblings is received, it will automatically be sent to the sibling, without involving the parent.
If the destination node is not in the list of nearby siblings, then the data has to be sent onwards to the parent node. At that point the parent node, which “knows” its siblings, can route traffic to one if its siblings. Thus the switching algorithm ensures that traffic is routed only as far as a parent whose child is a destination node.
Extensions with Multiple Radios
As shown in
Adding more radios to the outward interface increases throughput but also enables more freedom in making choices related to latency/throughput tradeoffs. For example, suppose that some traffic requires high throughput and other traffic low latency. If the compromise approach described in this invention is unacceptable because the range of requirements are too high, then two radios for the outward interface can reduce the range of requirements: One radio will address more of the low latency traffic with a low latency traffic route while the other will address the high throughput needs with a different high throughput traffic route. The wireless node now begins to resemble a wireless equivalent of network routers.
The algorithms described in this invention are still applicable: only the range of applicability has changed. The embodiment of the present invention is also relevant for wireless routers.
Security of the Network
Wireless transmissions are inherently insecure. While the technology to encrypt/decrypt secure data exists, the problem is communication of the keys over wireless to the nodes, from the access server. This common key distribution problem is addressed by the following embodiment of the system.
The wireless communication devices will have, as part of the algorithms resident in their operating system, the ability to generate a public and private key based on the RSA algorithm. These keys will be based on some unique identifier in the AP node the processor Chip Serial Number as an example.
When the Wireless device is first deployed, it will be connected via Ethernet cable to the access server and the node's public key will be transmitted to the Access Server. This public key will be used by the Access Server to transmit a common private key (using the symmetric AES encryption algorithm) to all nodes. Since only the Access Server knows the public key for each node, only the access server will be able to transmit this private key. Further, since the common private key for all nodes was transmitted in encrypted form to all nodes, it is impossible to decipher the common private key without knowing the public key for the node it was intended for. In addition, even if that public key for that node is known, it is useless since the private key for that node was never exchanged. The transmission of the common Private Key is thus secure.
Secure data transmitted by one AP node will then be encrypted with the common private key and be decrypted only at a destination AP node. By the same token, all data from the Ethernet to an AP node will be encrypted with the same private key for onward transmission.
The enterprise Access Server can be used to generate a new private key at regular intervals and transmit it to all Wireless AP Nodes in the system. The system is thus doubly secure.
Implementation of algorithm in Firmware
The control algorithms described above require significant resources—CPU. Memory—resulting in large footprint applications. Wireless devices and other network aware communication devices are typically embedded systems with low foot print requirements. A challenge that must be addressed—if this technology is to have practical applications is how to reduce the footprint of the software control layer to fit into embedded devices with 64 KB or 128 KB RAM.
A self-standing executable containing some of the algorithms and support functions has been produced within a footprint of less than 100 KB running on an Intel PXA250 Processor. Additionally in an embodiment of the present invention, the mesh and load balancing algorithms have been successfully ported to run on the hardware depicted in
The reason for the small footprint is that the approach of the embodiment of the present invention to building software is to include only the portions of an operating system needed by the programs. Thus, only functional blocks needed by the algorithms are added to the make file needed by the compiler to create the executable. If string functions are not required by any procedures in the program, then the string function library is not included when the executable is made. In contrast, a typical embedded operating system is more general purpose, requiring a larger footprint and possibly more CPU resources.
The language in which the algorithms are written is currently Java, and will may also include Microsoft™ .NET languages. In the embodiment of the present invention, a Java Class file converter has been built that takes Java Byte Code and disassembles it to produce what is referred to internally as R (for Real) classes. R classes are the C code equivalent of the Java Op codes, used by a Java Virtual Machine (JVM) to run the Java program. The R classes map to C code which is produced after examining the functional libraries needed and adding them to the list of support libraries needed to make an executable self standing. Once that is completed a self-standing executable is made for the processor.
An extensible service library of software components needed to build a complete Operating system (OS) is implemented through the embodiment of the present invention. This component based approach to building an Operating system from scratch enables one to select only the essential services needed by an application when it is ported from high level languages to small footprint embedded devices. Since only the essential services and not an entire general purpose OS is included, the footprint is compact. Additionally, there is a significant (3x-6x) performance improvement because layers of software needed to run code written in high level languages like Java or NET languages are no longer needed.
Thus there is a clear migration strategy in place from high level code generation to low level object code that includes all the functionality provided by an operating system to ensure that the object code is self contained.
There is implemented one version of this migration strategy where one begins with high level code written and tested in a development environment and can swiftly migrate it to a low footprint executable.
Since there is no OS, there is no easy way to tamper with the system. This approach—internally referred to as Application Specific Embedded OS software—thereby protects the security of the algorithms and enables the algorithms to run on low power (and less expensive) processors and with lower (and less expensive) memory requirements.
The simulations depicted are running the same code in each node shown in the figures. The code running in each node has been compiled to object code for the Intel™ PX250 processor as a proof of concept (
Upgrade Path for New Algorithms
A distributed network poses problems related to upgrading the software at each node. If the software is cast in concrete—as in an ASIC implementation—then there is no upgrade path available. Since the wireless standards are evolving, this is not a practical approach.
In the description of the embodiment of the modular approach to generating a self standing executable, it becomes apparent that there is no migration path available to the system to upgrade the executable easy.
This is resolved by providing a simple communication protocol for uploading new object code into the system. This has also been implemented as is internally called Simple Upgrade Protocol (SUP).
When the executable is made, a simple communication protocol is added, which, with proper authentication, using the public key of the node, can be used upload object code into the Flash memory of the device. A very thin boot kernel with the AP Node and the rest of the code is remotely installed. The boot kernel contains the simple communications protocol to upload object code and the security to ensure that only the Access Server can access the device. By building security at the boot level, one ensures that all code loaded into the system has to be authorized—since the security code cannot be overwritten.
Throughout the description and drawings, example embodiments are given with reference to specific configurations. It will be appreciated by those of ordinary skill in the art that the present invention can be embodied in other specific forms. Those of ordinary skill in the art would be able to practice such other embodiments without undue experimentation. The scope of the present invention, for the purpose of the present patent document, is not limited merely to the specific example embodiments of the foregoing description, but rather is indicated by the appended claims. All changes that come within the meaning and range of equivalents within the claims are intended to be considered as being embraced within the spirit and scope of the claims.
Appendix A
Exemplary 802.11 Infrastructure Control Layer Protocol Version 2.0
Design Considerations
Size Limitation
Since an Ethernet packet can only be a maximum of 1500 bytes the maximum length of a UDP datagram so as to not cause IP fragmentation is 1472 (1500−20 byte IP header−8 byte UDP header). Every SNIP packet consists of a 8 byte SNIP header followed by parameters. Since all IMCP functions use only 1 SNIP parameter the maximum length of the parameter value is therefore 1462 (1472−8 byte SNIP header−2 byte param length).
Packet Destination
All SNIP packets (either from Access Manager or from A/P) are sent via UDP to either the limited broadcast (FF:FF:FF:FF:FF:FF) or a chosen multicast address (01:00:5E:XX:XX:XX).
This method ensures that all A/P's get the message and also obviates the need for assigning I/P addresses to every A/P.
Definitions
Network Health Monitoring
Health Index
The number of unique Root paths determines the Health Index of an Access Point. In an “All-Root” network the Health Index determines the degree of redundancy of the Access Point's location. In an infrastructure mesh network for Relay nodes the Health Index determines the number of unique paths to the “wire”.
Heartbeat Interval
This attribute specifies the interval at which Access Points send in a “heartbeat” signal, that specifies the conditions of their operations.
Location Awareness Scan Interval
This attribute specifies the interval, at which Access Points scan the “airspace” to know about other Access Points in their vicinity, to determine their Health Index and for Dynamic Load Balancing purposes.
Dynamic Load Balancing
This Boolean attribute, determines whether an Access Point enables the adaptive dynamic load balancing functionality during its operation. If enabled the Kick-in Threshold and Delta T attributes determines the way this functionality is put to use.
Connect Cost
The Connect Cost is a value in the range [0,3] where 0 being the minimum Connect Cost. The Access Point increases its cost of connectivity during Dynamic Load Balancing, after it determines that the load has stayed beyond the Kick-in Threshold for a period of Delta T units. The Connect Cost is adaptively reduced according to the load.
Delta T
This attribute determines the duration of sampling for the Access Point to determine that the load was higher or lower than the Kick-in threshold for Dynamic Load Balancing purposes.
Kick-In Threshold
This attribute is a percentage of the total capacity of the Access Point. This is used in conjunction with Delta T for Dynamic Load Balancing purposes.
Dynamic Channel Allocation
Channel Number
When Dynamic Channel Allocation is disabled, this attribute specifies the channel of operation for the Access Point.
Topology
Latency Throughput Control
When zero, this instructs the Access Point to always choose a Parent, whose hop count is lowest. When non-zero, the Access Point may choose to “tradeoff” latency for throughput provided the difference in throughput is more than Latency/Throughput Trade-Off attribute.
Max Allowable Hops
This attribute controls the topology of the wireless network so that, any Access Point is never more than Max Allowable Hops away from the “wire”.
Latency/Throughput Trade-Off
This attribute determines, the difference threshold for an Access Point to choose a parent with more number of hops in exchange for lower latency.
Packet Information
Conventional Routing Protocols
Distance-Vector Routing (DV)
Each node maintains a table of approximate distances of nodes. Nodes send the table out to nearest neighbors. The receiving nodes update their tables and recalculate routes using a shortest path algorithm. Thus routing decisions are made using sub-global approximations and are not optimal especially in a dynamically changing environment like WPANs.
Link State Routing (LS)
Each node maintains a view of the entire network and broadcasts the state of the link to its nearest neighbors. The receiving nodes update their tables and recalculate routes using a shortest path algorithm. Pros include: widely used and commercialized; well tested; well documented. Cons include: well suited for static environments, not for dynamic environments; infrastructure oriented (dedicated hosts are organized for routing); not suited for resource constrained, low power devices
Standard AD-HOC Routing Algorithms
The IETF Mobile Ad-Hoc Networking Standard (MANET) has proposed the following AD-HOC routing algorithms.
Destination Sequenced Distance Vector Routing (DSDV)
AD-HOC on Demand Distance Vector Routing (AODV)
No Central Control
AD-HOC WPAN's typically work in environments where there cannot be any level of central intelligence as far as routing and parametric decisions are concerned. This requires the algorithm to be truly distributed and every device must be able to make decisions by itself.
Self-Configuring
AD-HOC WPAN's by definition need to be self-configuring without having the need for any network plan.
Self-Healing/Fault Tolerant
AD-HOC WPAN's need to be self-corrective and fault tolerant. Devices must be able to change their routing decisions in real-time as soon as a path gets clogged or closes down.
Dynamic Adaptive Load Balancing
The load on the network must be balanced fairly across all possible paths, and this decision must happen dynamically in an adaptive manner.
Pro-Active Routing
The routing decisions need to be made on a pro-active as opposed to an on-demand basis. This ensures that the task of routing does not interfere with the device's primary responsibility, which is to accomplish its own functionality.
Varied Bandwidth/QOS/Power Requirements
Devices have varied bandwidth requirements, some need isochronous performance (fixed latency), and some need bounded latency, and some might be power constrained and must use very low power for their transmissions.
Low Memory Footprint
The design of the algorithm must consider the fact that WPAN would typically consist of low footprint resource constrained devices.
Multi-Zone/Multi-Channel Support
The design of the algorithm must consider the support for routing between multiple Pico-cells or multiple network zones.
Network Layer Independent
The algorithm must not depend on the existence of a network layer protocol like IP. The algorithm must directly use the services provided by the MAC sub-layer.
Efficient Topology Lookup and Modification
Every device in an AD-HOC WPAN plays dual roles:
The design of the algorithm must consider the fact that, the primary role of every device on the network is to accomplish its own functionality, and routing of packets of other devices is secondary (unless the device is a special node that is just present for forwarding) and must not affect the performance of its primary role.
For the primary role, this means whenever the device has to send out its own data packet, to another device the routing decision must be very fast.
For the secondary role, this means whenever a device receives from another source destined to another device, it must be able to readily reference the routing decisions made by the source according to its Bandwidth/QOS/Power requirements.
When a device goes down, the topology and the routing decisions need to be modified in real-time so that network performance levels are maintained.
This application claims the benefit of priority from U.S. Provisional Application No. 60/421,930, filed Oct. 28, 2002, which is herein incorporated by reference. This application also incorporates by reference, applicant's invention disclosure document no. 527186.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5166929 | Lo | Nov 1992 | A |
| 6026303 | Minamisawa | Feb 2000 | A |
| 6046992 | Meier et al. | Apr 2000 | A |
| 6298053 | Flammer, III et al. | Oct 2001 | B1 |
| 6363062 | Aaronson et al. | Mar 2002 | B1 |
| 6426814 | Berger et al. | Jul 2002 | B1 |
| 6597919 | Kumar et al. | Jul 2003 | B1 |
| 6665536 | Mahany | Dec 2003 | B1 |
| 6683865 | Garcia-Luna-Aceves et al. | Jan 2004 | B1 |
| 6704301 | Chari et al. | Mar 2004 | B2 |
| 6831921 | Higgins | Dec 2004 | B2 |
| 6850502 | Kagan et al. | Feb 2005 | B1 |
| 6859652 | Karabinis et al. | Feb 2005 | B2 |
| 6865170 | Zendle | Mar 2005 | B1 |
| 6876643 | Aggarwal et al. | Apr 2005 | B1 |
| 6892068 | Karabinis et al. | May 2005 | B2 |
| 6904275 | Stanforth | Jun 2005 | B2 |
| 6909705 | Lee et al. | Jun 2005 | B1 |
| 6912373 | Lee | Jun 2005 | B2 |
| 7058050 | Johansson et al. | Jun 2006 | B2 |
| 7164667 | Rayment et al. | Jan 2007 | B2 |
| 7190690 | Tuomela et al. | Mar 2007 | B2 |
| 7200130 | Forstadius et al. | Apr 2007 | B2 |
| 20010021639 | Kaku | Sep 2001 | A1 |
| 20020055978 | Joon-Bo et al. | May 2002 | A1 |
| 20020085719 | Crosbie | Jul 2002 | A1 |
| 20020097681 | Treister et al. | Jul 2002 | A1 |
| 20020097696 | Kossi et al. | Jul 2002 | A1 |
| 20020159401 | Boger | Oct 2002 | A1 |
| 20020159409 | Wolfe et al. | Oct 2002 | A1 |
| 20020176390 | Sparr et al. | Nov 2002 | A1 |
| 20020181427 | Sparr et al. | Dec 2002 | A1 |
| 20030002442 | Flammer, III et al. | Jan 2003 | A1 |
| 20030012176 | Kondylis et al. | Jan 2003 | A1 |
| 20030012193 | Novaes | Jan 2003 | A1 |
| 20030026356 | Brommer | Feb 2003 | A1 |
| 20030114204 | Allen et al. | Jun 2003 | A1 |
| 20030185169 | Higgins | Oct 2003 | A1 |
| 20040001442 | Rayment et al. | Jan 2004 | A1 |
| 20040014491 | Weigand | Jan 2004 | A1 |
| 20040137924 | Herscovich et al. | Jul 2004 | A1 |
| 20050122925 | Chitre et al. | Jun 2005 | A1 |
| 20050181725 | Dabak et al. | Aug 2005 | A1 |
| Number | Date | Country |
|---|---|---|
| WO 2004017568 | Feb 2004 | WO |
| Number | Date | Country | |
|---|---|---|---|
| 20040090943 A1 | May 2004 | US |
| Number | Date | Country | |
|---|---|---|---|
| 60421930 | Oct 2002 | US |
