Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
  • H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements
  • H04L41/5003—Managing SLA; Interaction between SLA and QoS
  • H04L41/5019—Ensuring fulfilment of SLA
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L67/00—Network arrangements or protocols for supporting network services or applications
  • H04L67/50—Network services
  • H04L67/60—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources
  • H04L67/61—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources taking into account QoS or priority requirements
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
  • H04L41/08—Configuration management of networks or network elements
  • H04L41/0894—Policy-based network configuration management
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
  • H04L43/00—Arrangements for monitoring or testing data switching networks
  • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
  • H04L43/0852—Delays
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04W—WIRELESS COMMUNICATION NETWORKS
  • H04W28/00—Network traffic management; Network resource management
  • H04W28/16—Central resource management; Negotiation of resources or communication parameters, e.g. negotiating bandwidth or QoS [Quality of Service]
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04W—WIRELESS COMMUNICATION NETWORKS
  • H04W40/00—Communication routing or communication path finding
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04W—WIRELESS COMMUNICATION NETWORKS
  • H04W84/00—Network topologies
  • H04W84/02—Hierarchically pre-organised networks, e.g. paging networks, cellular networks, WLAN [Wireless Local Area Network] or WLL [Wireless Local Loop]
  • H04W84/10—Small scale networks; Flat hierarchical networks
  • H04W84/12—WLAN [Wireless Local Area Networks]

Definitions

• the present invention relates generally to data traffic over a multi-hop network. More specifically, the present invention relates to managing admission and routing of data streams transmitted over a multi-hop 802.11 based network (or a similar network) that provides traffic-shaping and rate-adaptation services at each hop.
• a mechanism for both controlling admission and making routing decisions is a desirable tool for efficient network management.
• such a mechanism determines for each new user which access point to associate with and which route, among the multiple potential routes, traffic and packets associated with the user should follow to reach their respective destinations.
• resource availability e.g., bandwidth and the number of access points
• QoS Quality- of-Service
• a new user may be denied admission when the required resources are not available, or when user QoS requirements (for both the new user and existing users) would not be satisfied if the new user is admitted.
• LNPWZ06 transmission control for multihop wireless backhaul networks with QoS support
• a routing scheme of the system determines proper routing of the user's data packets to their destinations. Routing in a multi-hop wireless network has been extensively discussed in the literature. In earlier wireless routing protocols (e.g., AODV, DSR, DSDP, and TORA), routing is handled independently from the lower layers, with path discovery being performed in a best-effort fashion without regard for system performance or QoS. Such routing algorithms are intended mainly for Mobile Ad Hoc networks (MANETs), where finding a connected path has priority (see, e.g., http://www.ietf.org/html.charters/manet-charter.html).
• MANETs Mobile Ad Hoc networks
• a method and apparatus control admission of data streams over a multi-hop 802.11 network.
• the admission control method specifically takes into consideration the joint effects of multiple mechanisms, such as traffic-shaping (e.g., aggregation and bursting) and routing or forwarding packets through multiple wireless links or hops. Because an admission decision impacts both the aggregation and bursting capabilities of the system, and thus the behavior of MAC and PHY layers, these MAC/PHY layer factors are considered directly in admission control. Additional considerations may include those for existing mechanisms already in use in the system, such as adapting the physical transmission rate at each hop to channel conditions (e.g., signal-to- noise ratio).
• the method views the nodes
• the "state" of a communication group includes the implicit or directed states of lower layer mechanisms, such as the aggregation or bursting levels each node in a group will use to transmit traffic.
• an access point is selected for association by the new user .
• This will be the access point with which the user's terminal communicates into the multi-hop network.
• a route from that access-point to the GAP is assigned.
• the route may be implicitly assigned upon admission because of a simple network topology.
• the system may consider a number of possible routes based on the admission decision.
• it can be that the user terminal is connected to an access-point which is itself a GAP, and thus no route is necessary.
• admission and associated parameter settings
• a method of the present invention not only takes into account admission and routes as a decision independent of lower layers, but directly takes into account the lower-layer traffic-shaping mechanisms used for transferring packets between access-points. It does do by either effectuating an implicit change in traffic-shaping mechanism used by each node in the system, or by directly changing the traffic-shaping mechanism used by each node in the system.
• An implicitly directed change would include, for example, eliciting a known deterministic (or average) reaction of the access point to a new flow being routed through it.
• the admission and route decision defines which access points receive additional traffic, and thus the reaction can be predicted.
• the system may have access points that implement a rule which aggregates, when possible, one incoming packet from each flow into a common next hop packet, sends the packets in burst transmissions, or performs both, when the flows have the same next-hop access point destination.
• a directed change specifies how each node should handle incoming flows to aggregate and burst in common transmissions.
• the admission and routing decision implies a change in state variables in each communication group.
• the state variables associated with a communication group may include, for example, an aggregation level (e.g., the number of packets that may be multiplexed together in a larger packet), a bursting level (i.e., the number of packets that can be sent in a transmission burst) and a physical-layer (PHY) transmission rate that are used by the nodes in accessing the common physical medium under contention.
• an aggregation level e.g., the number of packets that may be multiplexed together in a larger packet
• a bursting level i.e., the number of packets that can be sent in a transmission burst
• PHY physical-layer
• One advantage of the present invention is that desirable traffic-shaping parameters for each communication group are determined at the same time the best admission and routing strategy are determined for each new user.
• the present invention does so jointly in providing an admission control method for traffic, e.g. voice transmission, over a multi-hop 802.11 network.
• the method takes multiple parameters into account in order to make a decision and ensures that the system remains stable.
• the parameters may include options (e.g., aggregation level, bursting level, and transmission rate) and constraints (e.g., the number of users, access points, or sensing range of each user).
• the method computes the load of each of the communications groups in the network given each set of options and constraints, as determined by a potential admission and routing decision, and compares it against the wireless resources available to each communication group. Note, a load is calculated for each wireless resource being used by the system, i.e. there is a load calculated for each communication group.
• a given option (admission, routing, and traffic shaping option) one then checks the load required to handle the number and type (size, bursting, etc) of packets that would be transferred in each communication group against the available wireless resources. By doing so one can check whether or not the system operates in a stable fashion for the given option. Furthermore, among options that can operate in a stable fashion, the method has a measure of load on the system for each communication group. Among these options for stable operation, one can then select an option based on other criteria, e.g., the least number of hops, the highest transmission rate, the lowest load over all communication groups, or the lowest average load.
• an option may additionally include, beyond routing, admission, bursting and aggregation, an implicit or direct control of the physical layer data rate used by each transmission in a communication group.
• the present invention is better understood upon consideration of the detailed description below and the accompanying drawings.
• the present invention is applicable, for example, to a voice-over-IP (VoIP) application on a multi-hop 802.11 network.
• VoIP voice-over-IP
• Figure 1 illustrates a method that takes admission, routing and traffic-shaping into consideration in the process of admitting a new user, in accordance with one embodiment of the present invention.
• Figure 3 shows alternative system 300 to which the present invention is applicable.
• Figure 4 shows an example of using both aggregation and bursting at a transmission opportunity to improve efficiency.
• FIG. 5 shows a system with two GAPs which is not tree-rooted and for which the principles also apply).
• the present invention provides, in an 802.11 -based or similar network, a method to admit and route users.
• the method routes the user's traffic through multiple-hop wireless links, determines and provides (implicitly or explicitly) traffic-shaping options at one or more of the hops.
• the admissibility condition is also a stability condition that each communication group has to satisfy for a given set of parameters or options (e.g., aggregation level, bursting level, and connection rates).
• the solution determines an acceptable set of parameters (more than one solution may exist) and the route and traffic profile across the network, for each user and communication group, without overloading or violating Quality of Service (QoS) constraints imposed for existing and new users.
• QoS Quality of Service
• a central or "control” entity collects data regarding the state of the network and computes the loads for each communication group as would occur under different potential options.
• This control entity can exist anywhere in the network; in fact, it may be collocated with or implemented in the GAP.
• the control entity may collect information from and may take advantage of entities within the network that include automated mechanisms for adaptation (e.g. some communication groups may be able to adapt certain parameters autonomously as mentioned above, based the links they support).
• RAP relay access point
• a method under the present invention takes into account multiple parameters (e.g., aggregation level, bursting level, and connection rates) for selectively admitting users.
• system load may be determined from calculating the relative channel occupancy time in each communication group. System load may be used as a metric for stability. A stable system results when the communication groups can support the system load. For voice traffic, for example, if the time interval between voice packet generation for each user is x milliseconds, and the time required to transmit all packets generated over the x milliseconds interval is y milliseconds, for all traffic associated with users having links passing through a given communication group, y being greater than x, the system is unstable.
• the stability condition is deemed only a limiting condition. That is, there are many possible admission and routing options that can support a user and enable the system to perform in a stable fashion, providing the required quality of service to all users.
• a "system load" vector may be provided to adapt the traffic-shaping options under a given admission or routing selection.
• the system load vector may also be used to indicate unstable communication groups.
• the network supports a number of mobile clients, providing each mobile client a route (i.e., a set of one or more connected links) to at least one GAP. Clients may connect to any one of the four access points RAP1-RAP3 or the GAP.
• the packets exchanged between the GAP and the mobile client may traverse only one-hop to the GAP (e.g. in Figure 2 clients that connect directly to the GAP over links labeled "1 "), or multiple hops to the GAP.
• the basic network topology illustrated in this detailed description is a tree- structure, the method of the present invention is applicable to more general network topologies.
• access points can participate in several communication groups. One implementation provides an access point with different wireless network interface cards that allow them to receive and transmit simultaneously in each group.
• a single interface card is used with appropriate time-sharing between its use for multiple communication groups.
• Each communication group may be characterized by the following state variables: (a) the aggregation level used in each of the outbound flows to other terminals, RAPs or GAPs;
• the packet generation rate e.g., in packets/second
• the data rate e.g., in mega or kilo bits/second
• these parameters may be used to estimate a "load" on each communication group, according to one embodiment of the present invention, as discussed below.
• Many of these parameters are traffic-shaping parameters.
• Other state variables are possible, and include, for example, the present buffer levels at each transmitting terminal or MAC mechanism.
• While communication groups are assumed not to interfere with each other in the wireless medium (for the most part), communication groups affect each other through the nature of their respective flows. For example, in Figure 2, the state of one communication group "4" affects the state of communication group "1". This is because the packets sent to (from) mobiles in group 4 do eventually terminate (originate) at the GAP, and therefore do use the link RAP3 ⁇ -->GAP in group "1". Furthermore, the traffic-shaping options of a first communication group on one RAP or GAP can affect the traffic-shaping options of other communication groups or links involve in forwarding the traffic of the first communication group.
• the traffic-shaping settings and state of RAPl applied on its group "2" link to RAP2 affects the QoS in communication group "2".
• the selected traffic-shaping options may even have an effect on the QoS seen in group "1" since the flow from RAPl ->RAP2 eventually flows to the GAP over a group "1" link.
• traffic-shaping inbound traffic i.e., the number and the types of packets
• inbound traffic is collected together for outbound transmissions.
• one focus application is VoIP traffic.
• Stability with respect to the packetization interval of the voice codec may be used as a performance parameter for an VoIP application.
• the following model assumes that the network support only VoIP applications and considers stability with respect to the packetization interval. If all users use the same coder and packetization interval, the method of the present invention described below computes a channel occupancy time (i.e., the "load") of each communication group as the expected (or maximum) time to transfer on the links in the communication group one packet per user that uses this group.
• a user's route to the GAP may include multiple links and thus involves multiple communication groups.
• each access point i.e., at the GAP and at each of the RAPs
• packets are collected for transmission over the wireless links along the assigned routes.
• the collected packets to a common destination may be processed as a group in various ways to allow efficient use of the wireless resources.
• an access point may choose between (i) several options of "aggregating" multiple voice or transmission packets into larger single transmission packets and (ii) using a single transmission opportunity (e.g., an opportunity granted after contention in the Carrier Sense Multiple Access-CSMA scheme of 802.11) to "burst" multiple transmission packets over the wireless medium.
• a single transmission opportunity e.g., an opportunity granted after contention in the Carrier Sense Multiple Access-CSMA scheme of 802.11
• Using both the common transmission packet payloads and the transmission opportunities efficiently reduces overheads associated with data transmission over the 802.11 wireless interface.
• overheads can consume more than 2/3 of the total available wireless resource, in some instances, leaving 1/3 or less of the wireless resource for actual data transmission.
• the relative overhead per bit of payload information transmitted over an 802.11 link is reduced.
• system parameters e.g., "aggregation level”, “bursting level”, “physical layer rate, and "signal-to-noise ratio on the link" affect the resource utilization efficiency of the wireless resource each communication group has over the wireless medium.
• Figure 4 shows an example of using both aggregation and bursting at a transmission opportunity to improve efficiency.
• a burst of two transmission packets i.e., packets 401 and 402
• acknowledgement of packet 403 separated from each other by a short interframe spacing (SIFS).
• Transmission packets 401 and 402 are each formed by aggregating three voice data packets.
• the aggregation level on flow "i" is indicated by "A(i)' ⁇ and the bursting level by "B(i)" as in LKRP06 and RLKP07.
• each packet in a burst transmission may use a different aggregation level.
• A(i,j) we use to refer to the aggregation level of the j-th packet in a burst of flow "i", where "j" goes from 1 to B(i). All "A(i,j)” and “B(J)” values can also be statistical quantities.
• an access point is selected for the new user to associate and a route to the GAP is assigned.
• the route may be implicitly assigned upon admission due to the simple topologies. Admission (and associated parameter settings) is correctly managed such that the resulting system parameter values achieve desirable performance (e.g. maximized efficiency).
• a method of the present invention not only takes into account admission and routes, but also defines (implicitly or explicitly) a traffic-shaping mechanism (e.g., aggregation and bursting level) for transferring packets collected at one access point to the next access point.
• Figure 1 illustrates a method that takes admission, routing and traffic-shaping into consideration in the process of admitting a new user, in accordance with one embodiment of the present invention.
• the method first collects information regarding possible admission decisions and routes that are to be, or can be, considered for admitting the user into the system.
• the method collects the existing state of the network (e.g., the aggregation levels, the bursting levels, the SNRs, and the number and types of users in each communication group).
• the admission control algorithm includes primarily the following steps: (i) decompose or note the map of the multi-hop network into communication groups; (ii) compute the load in each communication group for different possible route and traffic-shaping options (steps 103-105); (iii) check the stability condition for each communication group for every possible scenario (step 106); (iv) consider further load attributes if stability is achieved for more than one scenario (step 107); (iv) admit or reject each user to the network, based on the best solution.
• the search for stable scenarios is conducted by theoretically computing the load vector (vector consisting of load values for each communication group) for each admission and routing scenario.
• a channel occupancy time is computed for that communication group.
• Each scenario takes into account the constraints of the multi-hop network.
• the constraints may include the numbers of access points, users, and current connections, and the sensing range of each user. Setting of proper constraints generally limits the search space.
• the users may have several access points in their sensing range and the network allows re-association of the users (without violating the stability condition), then scenarios where users who are already admitted into the network are re-associated to different access points may be investigated.
• the state of the system may change as result of a new user's admission.
• multiple changes may result from each admission option, and multiple admission options exist.
• the number of users and the types of users in each communication group may change, and a change in one communication group typically leads to a change in another communication group.
• Such changes follow directly from the fact that the traffic for a new user is carried by the assigned route or routes to and from one of the GAPs, the possible routes based on the admission options considered or selected.
• a method of the present invention determines how each change in a communication group (e.g., whether directed centrally or as a result of self- adaptation) changes traffic-shaping settings (e.g., aggregation, bursting levels and PHY rate on its links).
• Traffic shaping has a strong influence on system performance, and admission decisions affect traffic- shaping.
• the method may have options to re-associate existing users (i.e., changing the APs with which the user are associated and their routes to the GAP) as part of the admission consideration and the change in system state.
• a new "load” may be computed for each communication group to provide a new network state. These potential “load” measurements can then be used for making admission control decisions and to select between routing options and traffic-shaping options, where many options exist for a given admission option.
• the computed "load” indicates a relative (fractional) or weighted channel occupancy time for each communication group.
• the "system load” for a given option may be expressed as a vector of load values, with each value in the vector corresponding to the load for a communication group.
• the new "system load" for a given admission, routing and traffic-shaping option may in fact contain load values that show instability in one or more of the communication groups (i.e., the communication groups cannot support the admission option). For example, the load on one or more communication groups exceeds a pre-determined limit for acceptable operation (e.g., a load which exceeds the wireless medium's ability to deliver an acceptable QoS in delay or packet loss). Alternatively, a number of potential admission and routing options may allow all communication groups to operate acceptably.
• the final admission and routing decision among these possibilities may be made based on other criteria, such as the least number of hops, the highest signal-to-noise ratio, minimization of the maximum load over all groups, and load balancing, given the current constraints of the network.
• loads are calculated for each communication group.
• load components Some components are incurred per packet transmission, such as 802.11 PHY and MAC headers, while others are incurred per transmission opportunity, such as contention overheads.
• Still other load components are invariant to how many packets are transmitted, or how many transmission opportunities are used, such as the actual voice (or media) data being transmitted.
• the former two types of loads are intimately connected with traffic-shaping, since the loads can be amortized, using the techniques of aggregation and bursting, over multiple voice-packets.
• Aggregation level A(i) refers to the number of packets (e.g., voice packets) that are aggregated into a single 802.11 -packet in flow i.
• Bursting level B(i) refers to the number of 802.11 -packets that are transmitted per transmission opportunity in flow i. If different aggregation levels are used for different packets in a burst one can use the notation "A(iJ) " as described above to refer to the aggregation level of the "j-th" packet in a burst.
• PHY ( ⁇ ) refers to physical layer data rate that is used to transmit the z-th flow.
• PHY(i) takes values of 1, 2, 11, ..., 54 mega-bits-sec (Mbs).
• Figure 4 also assumes the aggregation includes exactly one voice packet from each user in the communication group.
• a (i) is more realistically the effective aggregation level (i.e., some 802.11 -packets may contain two or more voice data packets from a single user; but each user provides on the average one voice packet per 802.11 -packet).
• aggregate level A(i) and bursting level B(i) are assumed the same at every transmission opportunity. Although other values of aggregation level A(i) and bursting value B(i) are possible. B(i) need not be the same at every transmission opportunity. B(i) may be, for example, an average value over many bursts.
• Each admission and routing decision includes an assumption regarding a level of data packet traffic that results on each link in each communication group per packetization interval. For example, for VoIP traffic, a 2-way constant bit-rate (or constant packet rate) traffic may be assumed. If one voice packet is generated per call per direction within each frame-interval ("voice-packet-interval"), the number of voice packets transported within that communication group is given by 2 x the number of users using links in the communication group. It may also be 4 x the number of users using links in the communication group, at a RAP, if inbound and outbound likes use the same group, i.e. where the voice packet appears in both inbound and outbound links in the same wireless interface using the same channel. Of course, the assumption single voice packet per voice-packet-interval per direction is invalid for a variable bit-rate service, or where a single user may use multiple routes.
• a load to be borne by the communication group may be associated with each admission and routing decision based on the number of packets (per packetization interval) supported by flows in the communication group. This number defines the possible aggregation and bursting parameter options "A(i) " (or “A(Ij) ”) and "B(i) " that may be considered for flow "i” in the group. Such parameters may also be implicit on the operation of the nodes in response to the new traffic it sees, as discussed above. An option may consider a joint selection of admission, routing and set of parameter choices for all flows in the communication group. A load for the group can then be calculated.
• the parameters A(i), B(i) and PHY(i) are used to illustrate calculating a channel occupancy required to support that load. This calculation is done for all groups affected by the admission and routing possibility (option) being tested. LKRP06 and RLKP07 provide in greater detail the calculation described below.
• V(i) refers to the number of bits per voice packet within the z-th flow.
• V(i) may be provided the same fixed value for each user to simplify calculation of the load, although the analysis can be extended to cases where different users generates different bits per voice packet.
• V(i) may be used to model not only speech data generated by the speech or audio encoder, but also other overhead quantities such as (a) Internet Protocol (IP) header overheads; (b) Real Time Transport (RTP) overheads; User Datagram Protocol (UDP) header overheads, and (d) any other overhead (on average or amortized over users) within the payload of each 802.11 packet.
• Packet error rate P(i) refers the packet error rate of each transmitted 802.11 packet in the z-th flow, which is a function of V(i), PHY(i), and A(i). See, e.g., the "LKRP06" and "RLKP07" article mentioned above.
• Transmission probability p(i,j,m) refers to the probability that it takes u r ⁇ " transmit opportunities to transfer the "y " packet in a burst for flow i.
• the load (i.e., "channel occupancy time") for each packetization interval consists of three additive parameters: (a) successful transmission time T s ; (b) unsuccessful transmission time Tf ; and (c) the time when nodes are not transmitting but are in contention for the wireless medium, known as the back-off time Tt.
• Successful transmission time T s refers to the time spent successfully transmitting packets.
• Unsuccessful transmission time Tf refers to the time wasted due to transmission failures.
• Back-off time T b refers to the time spent in back-off mode by each MAC in the communication group.
• Successful transmission time T s includes 3 components: (a) the time spent transmitting voice (or another application payload) bits on each link, (b) the time spent transmitting an acknowledgement (ACK) packet, and (c) additional overhead, such as interframe (IF) spacing.
• the time spent transmitting an application payload e.g., voice
• V(i), A(i), and B(i) are fixed and the same for each user on link i, V(i) ⁇ A(i) ⁇ B(i) bits are transmitted during each packet-interval (e.g., one voice-packet per user) over the link.
• one or more ACK packets may be sent per transmission opportunity to cover all packets burst transmitted during that transmission opportunity. For example, if there are no hidden terminals (i.e. all nodes in a group can sense transmissions from all other nodes) and if there are no other channel impairments resulting in bit or symbol losses, it is shown in LKRP06 and RLKP07 that if the first packet in a burst transmission is successful, then all other packets in the burst are successful. Therefore, for a set of B(i) burst packets on ACK would be used to acknowledge the burst.
• one ACK packet is provided for transmission of V(i) ⁇ A(i) ⁇ B(i) bits.
• Traffic- shaping via both aggregation (A(i)) and burst processing B(i) amortizes the overhead over multiple voice packets.
• one ACK packet may be sent for each 802.11 -packet transmitted in a burst.
• the first case is when the system choses to do so (this practice may not conform to the 802.11 standard).
• the second case is when there are channel impairments such as hidden terminals or bit or symbol losses due to noise on the wireless channel.
• individual packets in a burst can be lost, terminating the burst transmission and resulting in retransmissions of the lost packets as described in RLKP07.
• the parameter "p(i,j,m) " helps to describe the average number of ACK packets that are transmitted. For greater detail, see, the RLKP07 article.
• the aggregation mechanism helps to amortize ACK overheads from multiple speech packets by creating a single ACK for an aggregated transmission. Bursting can further amortize ACKs over multiple aggregated packets, depending on the ACK mechanism and channel impairments.
• IFS inter-frame spacings
• DIFS Distributed IFS
• Time wasted due to transmission failures can result from collisions in the transmission medium, or errors in transmitted information due to channel impairments. Failures can happen at each transmission opportunity and for each packet transferred within a burst of transmission. Depending on the channel impairments, aggregation level A(i) can affect the packet loss rate. Tf depends on the number (or expected number) of transmission failures per transmitted burst (i.e., on transmission probability p (i,j,m)). See, e.g., the articles LKRP06 and RLKP07 for a description of how to calculate time 7), and the 802. lle/D 13.0 standard and the 802.1 Ia standard, for the overheads relevant to each failure.
• the main time not included is time for ACK packets that are missing from 802.11 -packets, or which are not correctly received.
• time 7/ is often given as an expected value, and not an deterministic value as in time T s . See, for example, the LKPR06 and RLKP07 articles.
• Other ways for calculating a value representative of time Tf includes the distribution of the Tf value, or the probability that time Tf exceeds a predetermined time value.
• Time T t spent in back-off mode by each MAC in the communication group represents the time wasted while the channel is idle, or when MAC mechanisms in the system are in a random back-off state, contending for access to the channel.
• Time T t is a central part of the Distributed Coordination Function (DCF) mode under the 802.11 standard, which is based on a Carrier Sense Multiple Access (CSMA) scheme. See, e.g., the descriptions in the 802. lle/D 13.0 standard and the 802.11a standard.
• DCF Distributed Coordination Function
• CSMA Carrier Sense Multiple Access
• the total system load is then estimated by taking into account all links and users, and adding all the individual contributions to the values T s , Tf and T b . Once the totals are known over all traffic within the communication group, the load for the communication group is given by the sum:
• This time-based load may be converted to a relative value.
• a "relative load” may be used, given by:
• the relative load for all communication groups may be either strictly less than, or at times sufficiently less than or equal, to 1.
• the time load and the relative load values given above for each communication group form a vector of load values for a admission option.
• the relative load of each communication group for each option can be used to compare options. For example, the maximum load or the average load estimated across communication groups for each option can be used to compare options. Comparison of options can be refined using other metrics, such as the number of hops of each flow or delays.
• Danjue Li is an intern at DoCoMo USA Labs working towards her Ph.D. at University of California Davis.
• the other authors are employees of DoCoMo USA Labs power sources (lamp posts, etc.), the fact that 802.11 is a widely used standard-based solution, and the possibility of using only a few gateways to provide (wired) connectivity outside of the net.
• such networks do suffer from similar inefficiency problems as single-hop systems due to inherent overheads in accessing and transmitting on the wireless medium [1] using the Distributed Coordination Function (DCF).
• DCF Distributed Coordination Function
• multi-hop scenarios such inefficiencies can propagate with flows across the network resulting in bottlenecks and even larger losses in efficiency relative to single-hop scenarios.
• Section II we discuss the inefficiency problem of the current DCF scheme and provide a survey on related works. Then we present our proposed joint Traffic shaping, rate Adaptation and Routing (TAR) mechanism for improving the efficiency of 802.11 multi-hop networks in Section III.
• Section IV we provide a theoretical framework to analyze the voice capacity of a multi-hop network without channel impairments, and then in Section V, we use this theory to explore adaptive TAR configurations as a function of channel impairments.
• the relative benefits of both aggregation and bursting i.e., relative decrease in overheads " ⁇ " and "/3" can clearly be seen.
• Section VI we compare the theory with ns simulations, and look at how and why they differ.
• Section VII we discuss in Section VII how to manage such networks in a practical setting, where wireless channel quality and voice traffic patterns change constantly or are uncertain. Issues of network planning and admission control are all jointly discussed in this section. Finally, in Section VIII, we summarize the work and discuss some open problems.
• the inefficiencies of the DCF scheme can be represented by two factors “ ⁇ ” and "/3".
• the a factor refers to the impact of the overheads incurred by the idle periods used for contention resolution and the wasted time slots due to packet collisions [I].
• These types of overheads are inherent to the distributed nature of the DCF access mechanism.
• ⁇ , 0 ⁇ a ⁇ 1 we denote ⁇ , 0 ⁇ a ⁇ 1, as the ratio of such overheads to the raw bandwidth of the wireless medium. If the bandwidth is W, then (1 - a) W is the effective bandwidth that can be used for successful transmissions.
• the ⁇ factor is affected by the maximum transfer unit (MTU) of the MAC layer.
• MTU maximum transfer unit
• a packet exceeds the MTU size its payload is fragmented into multiple packets. All fragments are sent as consecutive transmissions in a single transmission opportunity. Therefore at each fragmentation boundary additional /3-related overheads are added and the ⁇ factor has discontinuous jumps. See illustration in Figure 1 [I].
• 802.1 Ie [5] extends the original 802.11 standard to provide additional mechanisms for bursting multiple packets in the same transmission opportunity to reduce the a factor.
• 802.1 In [6] [7] in addition to bursting, different types of aggregation are used at the MAC layer as well as the PHY layer to further reduce both a and ⁇ factors.
• Wang et al. [8] proposed to aggregate packets from multiple users at an access point (AP) and use a multicast transmission to better utilize transmission opportunities in the downlink direction. Header compression and silent suppression have also been investigated in the past for voice traffic to improve system capacity [3]. Although they focus on reducing the amount of data to be transmitted instead of improving the efficiency of the DCF mechanism, silence suppression in particular does have the side effect of affecting a overheads by reducing contention.
• each access point aggregates (i.e., multiplexes) arriving flows and transmits these aggregated packets as bursts within the same transmission opportunity, at a transmission rate adapted to the noise in the wireless channel.
• Adequate routing allows this traffic shaping to take place.
• Aggregation can efficiently suppress both ⁇ and ⁇ factors by reducing the number of contending flows and sharing the MAC/PHY headers, frame separations and control overheads. Bursting focuses on suppressing a overheads only and is thus less effective than aggregation. As shown in a previous paper [1] using a larger aggregation level can potentially increase voice capacity. However, this is only true when the impact of channel impairments, such as bit error, is negligible. If packet losses due
• bit/symbol/packet-errors are non-negligible, aggregation might lose its advantage over bursting.
• bit/symbol-errors and a given error rate
• packets generated at a large aggregation level suffer from higher packet loss rates than shorter packets.
• the packet loss rate becomes large enough, the overhead created by retransmitting long erroneous packets can exceed that saved by reducing a and ⁇ factors per each transmission.
• bursting smaller (possibly aggregated) packets may be a better choice. Therefore, by combining aggregation and bursting together one can try to balance the two strategies, achieving better system performance.
• parameters such as the bit/symbol/packet loss rate depend on the underlying PHY rate. When traffic shaping loses its advantages because of high packet loss rates, one may try to balance this loss with the PHY rate adaptation options available in 802.11.
• the model encompasses the following components: (1) wireless VoIP clients that act as both the source and sink of VoIP flows; (2) Wireless TAR-enabled RAPs that can forward and process many VoIP flows; (3) A TAR-enabled GAP through which VoIP sessions are established with hosts outside the network.
• wireless VoIP clients that act as both the source and sink of VoIP flows
• Wireless TAR-enabled RAPs that can forward and process many VoIP flows
• a TAR-enabled GAP through which VoIP sessions are established with hosts outside the network.
• each RAP/GAP is equipped with multiple wireless interface cards, one for each communication group that the node is participating in.
• Each WNIC can operate simultaneously in parallel, Le. receiving on one interface and transmitting on the other one at the same time. To avoid interference, we require each communication group operate orthogonally to each other.
• each voice call includes one uplink flow (i.e. flow running from a RAP to the GAP) and one downlink flow (i.e. flow running from the GAP to a RAP), and these two flows start close in time.
• uplink flows and downlink flows are shaped independently.
• RAPs will take advantage of the idle time required by the GAP to fulfill all their backoff requirements to minimize the medium idle time of the system.
• the main idea underlying the analysis is to count the number of packet exchanges that can be held in D 3 by calculating average wireless resources, i.e. long term average of channel occupancy time T required to support one packet exchange per user.
• This approach has been originally proposed by Hole et. al [14] to predict voice capacity of WLANs.
• We extend such analysis by explicitly including consideration for traffic shaping strategies, channel impairments and different PHY transmission rates.
• the "exchanges" mentioned above are a pair of bursts of aggregated voice packets as shown in Figure 4, one burst for uplink traffic and one burst for downlink traffic.
• T mainly includes the time of successfully transmitting the burst of aggregated packets, T s , the time wasted due to collision-caused transmission failures, Tf, and associated backoff time T&, i.e.
• RAPs and GAP can be configured using same TAR parameters in a balanced case or different ones in an unbalanced case.
• T s consists of the transmission time of voice packets T ⁇ , the transmission time of ACK message T ac k, and any necessary inter-frame spacings (IFS). Since packets can be dropped due to collisions, the long term average of T 3 depends on the probability of having a successful transmission.
• m denote the number of undergone collisions/retransmissions before a success as m
• T ⁇ Tbeader + T d ata, and T header , T d ata, and T ack are defined by:
• J- header ⁇ PHY ⁇ JJT ⁇ , (3)
• T ack T PH y + -* ⁇ - (5)
• Tw LAN ⁇ T(IHt) ⁇ A(i), B(i)) (8)
• Wasted time due to transmission failure (a) when a packet exchange fails due to collision, failure can only happen to the first transmission in a burst;(b) when the packet exchanged fails due to channel errors, failure can happen any time during the bursting procedure and here shows k-th aggregated packet fails its first transmission during the bursting process and it is successfully transmitted after m retransmissions;
• TWLAN has to be less than assigned wireless resource D 8 , i.e. TWLAN ⁇ D s .
• D 8 assigned wireless resource
• Amax can be computed using 1 D B R S +S( 8 IP ⁇ VDP+IO ' P) j •
• a and B denote the size of candidate sets for aggregation level and bursting level, respectively.
• the two-stage searching can reduce the complexity from 0 ⁇ NR B NR ) to O( ⁇ B).
• T f ' and T s ' are updated values for a TAR-disabled scenario, respectively, by setting A(i) and B(i) in (2) and (4) to be 1.
• IEEE802.1 Ib provides four different PHY modes at 2.4GHz with each corresponding to a different modulation scheme (Table II).
• PCCKM ⁇ p(S,), (21) where Q ⁇ x) -rfc ), A j is the Hamming Distance between a codeword i and a codeword j, and r c is the code rate, which is 1 for modulation schemes used in IEEE8O2.1 Ib.
• Ps channel SER beed by (18) or (20), depending on its modulation scheme
• Pc is the collision probability defined in Section IV
• T e (R(i), A(i), B(i)) be the long term average time of channel occupancy time of node i when packet transmissions can suffer from both collision and channel errors.
• R(i), A(i) and B ⁇ i) are similarly defined as in Section IV, except that R(i) is no longer fixed to be l lMbps. Instead, it is a variable determined by measured channel quality.
• T e can be computed as:
• T e (R(i), A(t), B( ⁇ )) E(Tf) + E(T f ) + E(Tg), (23)
• E(Tf) is the time used to successfully transmit the burst of aggregated packets when considering both collisions and channel errors.
• E(TJ) is the wasted time slots due to collisions and channel errors, and E(T ⁇ ) is the time spent on performing backoff.
• channel errors can cause a retransmission to any packet in a burst, not just the first one, and retransmission limit M max is applied to each of them as well.
• T WL ⁇ N ⁇ r « (fl(i),A(i), ⁇ (i)) (28)
• ns-2.26 implementation with an implementation of 802.1 Ie EDCF [19] by adding a traffic shaping buffer on each node.
• This buffer is used to accumulate enough packets before performing aggregation and/or bursting.
• buffering deadline to be equivalent to 30ms, i.e. the packetization interval of voice codec. It means that every 30ms, node i aggregates a single packet from each of A(i) users into a super packet and then bursts out B(i) of those super packets after contending for the channel. This buffering delay depends upon the relative transmission/arrival times of each packet in the first hop.
• Table III, IV, and V compare voice capacity predicted by using our proposed model to voice capacity obtained through ns simulations when three different traffic shaping schemes are implemented: a) joint aggregation and bursting; b) aggregation only; and c) bursting only.
• both model-predicted voice capacity and simulation-obtained voice capacity are evaluated for different NR value, i.e. TVR € ⁇ 1,2, ...15 ⁇ , and all nodes are operating at llMbps.
• Table III, IV, and V also give out corresponding A ⁇ i) and B ⁇ i) settings for achieving those voice capacities.
• AU simulation results hereinafter are average results from multiple simulation runs.
• Figure 8 plots the voice capacity for different traffic shaping schemes.
• our proposed joint traffic shaping scheme outperforms the two schemes which consider aggregation and bursting exclusively.
• N R certain point
• Figure 9 shows the model-predicted performance versus simulation-obtained performance when channel SER is 5 • 10 ⁇ 4 . Again, we can see that when taking channel errors into account, our proposed model can still be very accurate in predicting voice capacity of the system. Compared to Figure 8 where we have negligible channel bit error, Figure 9 shows much lower voice capacity for schemes where aggregation is used to shape the traffic. This happens because compared to original voice packets, aggregated packets are much longer and as a result more vulnerable to-channel bit errors. Frequent retransmissions triggered by packet losses increase the a overheads, causing voice capacity degradation. However, such performance degradation does not appear to the bursting-only scheme.
• Figure 10 shows voice capacity for a wide range of signal-to-noise-ratio (SNR), where NR is fixed to be 4 and different transmission rates are considered.
• SNR signal-to-noise-ratio
• Figure 11 illustrates the traffic shaping settings deduced from the model for different SNR values. From Figure 10, we can see that when nodes are operating on high data rates, i.e. 5.5Mbps and HMbps, if SNR value drops to be below certain threshold, voice capacity deteriorates very fast. This is due to that the system tries to reduce the aggregation level to maintain a better error resilience, as shown in Figure 11.
• aggregation level does not always decreases monotonically. Instead, at certain points, higher aggregation level is used for a lower SNR value. Such fluctuations happen because at those points, there exist more than one traffic shaping settings which can yield the same voice capacity, and the
• Rate adaptation For a TAR-enabled system, when wireless channel quality deteriorates, the system chooses to switch to a modulation scheme which can provide better error resilience. Among four modulation schemes existing in 802.11b, BPSK has highest error resilience but lowest bit rate, while CCKl 1 has highest bit rate but is the least resilient to channel errors. Therefore, switching modulation schemes based on wireless channel quality is actually requiring nodes in the system to be able to adapt their transmission rates. Previous literature has demonstrated the efficacy of rate adaptation on improving wireless system throughput. In the rest of this session, we examine the impact of rate adaption on voice capacity of a TAR-enabled system, especially its joint force with traffic shaping.
• (a), (b) and (c) are results for a system without implementing any traffic shaping strategy, a TAR-enabled system with balanced traffic shaping settings, and a system with unbalanced traffic shaping settings, respectively.
• Solid curves in this figure are voice capacity without performing rate adaptation and blue dotted curves are capacity after adapting PHY rate.
• system capacity decreases very fast when operating at llMbps in SNR range 9.IdB - 11.9dB or 5.5Mbps in SNR range 6.OdB - 9.OdB.
• Figure 16(b)(c) we observe the similar patterns in Figure 16(b)(c). However, by performing rate adaptation, the system becomes more resilient to the deterioration of channel quality, as illustrated by blue dotted lines in all three plots in Figure 16.
• FIG 17 shows the generic architecture of a TAR-enabled node.
• SNR signal-to-noise ratio
• each node can estimate its wireless channel quality. SNR measurements will then be collected by TAR control unit. Depending on whether the system implements centralized control mechanism or distributed one, SNR measurements will be processed differently.
• TAR control unit of each RAP forwards its collected information to the GAP and retrieve updates on TAR configuration ⁇ . e. aggregation level, bursting level, PHY transmission rate and routing decision) from the GAP.
• each RAP processes SNR information locally to obtain TAR updates by looking up the prebuilt table to determine the best traffic shaping settings and PHY mode for the next transmission attempt. Meanwhile, call manager updates its serving list using received routing related information.
• Managing wireless mesh networks includes two parts: how to admit new users into the system (i.e. which AP to associate each newly joined user to) and how each AP operates given the users they support.
• System status is decided by many variables, i.e., transmission rates of each RAP/GAP, traffic shaping settings used by the system, number of users existing on each AP and etc.
• the paper provides a theoretical framework that includes many of the main parameters of interest, e.g. aggregation, bursting, PHY rate, and symbol error-rate, on each link in the multi-hop network.
• the admission and routing choices implicitly define which links are present and what aggregation and bursting settings are possible.
• the framework therefore provides a tool to help understand, manage and plan such networks, allowing one to identify potential bottlenecks and heavily loaded areas in the network.
• Section II we briefly discuss the inefficiencies of the current 802.11 DCF mechanism, leading into our proposed Joint Traffic shaping, Admission, Routing and Rate-adaptation (JTARR) approach.
• Section III we provide a theoretical framework to analyze the loads in a multi-hop network for the case of no channel impairments, describing in Section IV how to use these loads to give an upperbound on VoIP capacity.
• Section V we extend this theory for cases with channel impairments.
• Section VI we compare the theory with ns-2 simulations, looking in detail at joint tradeoffs of parameters.
• Section VII we discuss briefly how to manage such networks in a practical setting, showing how the framework can be used to address a number of different objectives, including maximizing VoIP capacity.
• Section VIQ we summarize the work and discuss some open problems.
• JTARR JOINT TRAFFIC SHAPING, ADMISSION, ROUTING AND RATE ADAPTATION
• JTARR Joint Traffic shaping, Admission, Routing and Rate adaptation
• Aggregation is the process by which payloads from different IP packets are included into one 802.11 transmitted packet. This can reduce the number of contending transmission packets, and amortizes MAC/PHY-headers and control overheads over multiple voice packets. Bursting is the process by which multiple 802.11 packets are transmitted in a single transmission opportunity. This focuses on suppressing contention/idle-slot overheads and is thus less effective than aggregation. However, if packet losses due to bit/symbol/packet-errors are non-negligible, then aggregation may have disadvantages since it relies on using longer packets.
• bit/symbol/packet loss rate depends on the underlying PHY rate
• Our goal is to present a framework quantifying the general tradeoffs of potential joint settings.
• a communication group is defined as a set of flows that contend with each other for the same wireless resource.
• Such a group is defined by a given wireless channel in a given geographic location.
• a single flow represents a number of VoIP links with the same origin and destination (in a given communication group) that are bundled together through aggregation and/or bursting to use common transmission opportunities. Such links can originate/terminate at clients or access points.
• access points are equipped with multiple wireless interfaces. Interfaces can operate in parallel, i.e., receive and/or transmit on interfaces simultaneously.
• Figure l(a) In this example we consider using 4 channels and two gateways where access points may have up to two interfaces.
• the channel that a flow/link uses is synonymous with the term communication group.
• RAPl, RAP2 and GAP2 each have one interface on Channel 1 that is used for connections between these access points.
• RAPl has an additional interface on Channel 4 for connection to GAPl.
• Each access point uses only its "Primary Channel" to communicate with users in their service area. Given the constraint of 4 channels, GAPl uses Channel 4 for this purpose and RAPl uses Channel 1 for this purpose.
• RAP2 and GAP2 use additional channels, channel 2 and 3 respectively, to serve users directly. Additional channels could be used to create more interfaces, or different primary channels. Alternatively, additional channels can expand the coverage area as shown in Figure l(b) where groups re-use the same channels but are geographically separated.
• the 2-hop network in Figure 2 has the following components:
• RAPs have two interfaces and the GAP has one interface. If an additional non-interfering channel is available the GAP could use it to serve its clients directly thus creating an additional (K + l) th group.
• each voice call includes one uplink direction and one downlink direction, and packets travel over one or two hops.
• each voice call includes one uplink direction and one downlink direction, and packets travel over one or two hops.
• the load estimates we can determine the feasible set of options for each group.
• channel occupancy time and "load” interchangeably.
• N n nodes nodes, node( ⁇ ), .. ., node(N n ), which can be mobile clients, RAPs, and/or GAP(s).
• node(l), . .. . ,node(K — 1) are RAPs
• node(K) is the GAP
• node(K+ 1), .. . , node ⁇ N n ) are clients that connect directly to the GAP.
• the situation may be different.
• We use the index i 1, . . . , Np to index the Np flows being supported by this group.
• the bursting level B(i) denotes for flow i how many aggregated packets are burst in a single transmission opportunity.
• the A ⁇ i,j) value is chosen so that the length of any aggregated packet does not exceed the maximum transfer unit (MTU) limit of an 802.11 packet.
• MTU maximum transfer unit
• the maximum value we denote as A max .
• R(i) and H(i) denote the PHY layer data rate and PHY layer basic rate used for flow i.
• Tg the expected load generated by the flows in the group in a D 3 interval of time.
• the load comprises three components: 1) T 8 , the time for successfully transmitting bursts of aggregated packets; 2) Tf, the time wasted due to collision-caused transmission failures; and 3) T b , the associated time spent in random backoff when contending for the channel. These times are often statistical.
• T 8 the time for successfully transmitting bursts of aggregated packets
• Tf the time wasted due to collision-caused transmission failures
• T b the associated time spent in random backoff when contending for the channel.
• Pftrst(iJ) Pfaa(i,j - l)
• T s (i) denote the time spent on successful transmissions of voice packets from flow i.
• T d ataiiJ $$- (D s R s + (IP + UDP + RTP) - 8) (7)
• the parameter values used above for 802.1 Ib can be found in Table I. If the j — 1 th packet fails, then the j th packet becomes the first in a number of subsequent burst attempts.
• T d ⁇ t ⁇ ih j would be a statistical quantity, reflecting a range of possible values depending on what sized/type packets, at a given time, are aggregated and burst together in the flow. Individual values have to conform to constraints such as the MTU size, and in (10) the expected value E ⁇ T d ⁇ t ⁇ ⁇ hj)] is ultimately the quantity of interest. For simplicity in the exposition we will focus on the CBR case with constant D s and R s .
• TA TO denotes the ACK timeout, i.e., the time a sender has to wait before declaring its packet as lost.
• T ATO T PHY + SIFS + ⁇ [18]; ⁇ is defined in Table I.
• p(m, i,j) the expected number of such failed first transmissions is given by:
• FiiiJ E"TM x+1 rnp(m, i,j) (12) Note, under the collision-only assumption j > 1 packets never fail if they are transmitted as the second, third, etc. in a burst. This is implicitly assumed by (12). With (12),
• a transmitting node may see one of two cases.
• the node on getting a packet, the node senses the channel is free for a period DIFS. In this case no backoff is performed and the packet is transmitted.
• the channel was not free for a period of DIFS.
• the node proceeds with contending for the channel using a random backoff. Assume the transmitting node has failed to transmit the first packet in a burst after m attempts, and let U ⁇ be the length of the r th backoff used in units of time slots. The cumulative time spent on backoffs for flow i after m such attempts is:
• the first term accounts for the probability that a backoff is triggered by packet j.
• the second term counts for any r, r > 1, backoffs.
• E[T b ] we bound E[T b ].
• the group-wise analysis allows us to compute an upperbound on the voice capacity (based on a lowerbound on load) for many systems by extending the approach in Section IV-A. Indeed, in order to support calls in a system Tg has to be less than the packetization interval D s for all groups Q.
• Tg has to be less than the packetization interval D s for all groups Q.
• N R — K — 1 RAPs each with a different primary channel, and a single GAP using a (N R + l) th channel.
• RAPs connect to the GAP using this (N R + l) th channel.
• the voice capacity is given by the maximum number of calls that can be supported by flows in the 2 nd -hop, i.e. flows GAP ⁇ RAPs and GAP ⁇ clients using the ./V R -f 1 th channel (group).
• each RAP is supporting one uplink RAP ⁇ GAP flow.
• the number of voice-packets in D s seconds supported by these uplink flows is given by
• the voice capacity is the maximum value of (20) such that the system load does not exceed D s .
• Tg often reduces to a form where the GAP is the node achieving the maximum in (14) and (17).
• B(i) B Vi and A(i, j) — A Vz, j.
• the capacity can be obtained by an exhaustive search on B and A.
• RAPs can use different parameters and/or support varied numbers of clients the search space can be large and a capacity analysis can be less meaningful.
• Often in looking at bounds for such "unbalanced' cases we search locally around optimal solutions for balanced cases.
• Tg is actually a lowerbound on load. In a practical system one should allow for a budget ⁇ > 0 and test Tg ⁇ D s — ⁇ , as discussed in Sections VI-A and VI-B.
• T ⁇ E[Tt] + E[f ⁇ + E[f ⁇ l (25)
• E[Tf] is the time used to successfully transmit aggregated packets when considering both collisions and channel errors
• E[T j ] is the wasted time slots due to collisions and channel errors
• E[T ⁇ ] is the time spent on random backoff. Values E[T j ] and E[f ⁇ ] are lowerbounds for E[Tf] and E[T ⁇ ] respectively.
• a collision can only happen to the first packet in the burst. Symbol errors can affect any of the packets in a burst. Thus only the first aggregated packet transmitted in a burst sees both impairments.
• rf(i,j) (SIFS + T v (i,j) + T AT0 ) and ⁇ f (i,j) is defined in (11).
• Section IV describes the analysis of voice capacity for a 2-hop network.
• P E > 0 By simply substituting T g for Tg in the analysis.
• ns-2.26 extended with an 802.1 Ie EDCF package [19].
• the buffers are used to accumulate packets before aggregating and bursting functions.
• Both theory and simulations use the parameter values in Table I.
• a RAP say node(A;*)
• a client is a member of only one flow.
• U k * be the index set of all uplink RAP ⁇ GAP flows in S k* serviced by node(A;*).
• RAP serves C(A;*) clients
• parameters satisfy ⁇ i&Ak , A(i,j) — C(A;*).
• the GAP accepts one packet from each client every D s msec and groups those destined for a common RAP into one or more flows.
• R(i) l 1 Mb/s the constraint ⁇ ) i6 ⁇ .
• ⁇ j ii A(i,j) — C(h*) ⁇ 15 represents the first-hop limit.
• the tables show the A and B achieving the maximum number of calls for each (balanced) scenario. If different RAPs support different numbers of clients, using different traffic-shaping parameters, one could find cases supporting a larger number of clients. However, the search space considering all cases is too large to simulate and present.
• a RAP uses only one uplink RAP ⁇ GAP flow, and the GAP only one downlink GAP ⁇ RAP flow per RAP, to service calls.
• multiple flows are often used to support uplinks and downlinks, i.e.
• Figure 5 shows the model-predicted performance versus the simulation-obtained performance with an i.i.d. SER rate of 5 • 10 ⁇ 4 .
• SER rate 5 • 10 ⁇ 4 .
• SNRs signal-to-noise-ratios
• the figure shows "balanced-cases" only. A few points are labeled with the corresponding SER and the selected "balanced-case" parameters (A, B) that maximize voice capacity 1 . What the figure shows is that below a certain SNR the aggregation levels and voice capacities drop dramatically within a 2 to 3 dB SNR range. Beyond this range switching to a lower PHY rate (and a larger A) is preferable. Given the higher first-hop limit (15 calls) in the 11 Mb/s case, bursting is useful to allow RAPs to service up to 12 clients per RAP without increasing the aggregation level beyond 6.
• rate adaptation mechanism does however depend on the rate adaptation mechanism.
• Many systems have existing PHY rate adaptation mechanisms. If these mechanisms maintain a low SER, e.g. a SER ⁇ 10 ⁇ 5 switching before the sharp drops in voice capacity, the rate adaptation mechanism operates independently and the system could simply set B(i) and A(i, j) according to the minimum SER rate. If there is access to the SER, e.g. though channel strength measurements, etc., and to the adaptation mechanism then there is a 2 to 3 dB SNR region in which capacity can be increased using finer adaptations of B( ⁇ ), A(i,j) and R(i).
• system load calculation to manage the system depends on the objective. If one wants to minimize the total system load, i.e. min ⁇ g Tg, then a strategy that limits the number of hops by admitting users first to GAPs when possible, and then first-hop RAPs, etc., makes sense. For example in Figure l(a), userl, user2 and user3 would be admitted to either Gl or G2. If one wants to minimize the maximum load on a GAP, then admitting some users to relays first helps. Here if no other users are present in Figure l(a) then admitting users2&3 to RAPl, using G2f ⁇ RAPl to support an aggregated flow for users2&3, and admitting userl to Gl could be a good choice.
• MAC Medium Access Control
• QoS Quality of Service

Landscapes

• Engineering & Computer Science (AREA)
• Computer Networks & Wireless Communication (AREA)
• Signal Processing (AREA)
• Mobile Radio Communication Systems (AREA)
• Data Exchanges In Wide-Area Networks (AREA)
• Telephonic Communication Services (AREA)

Priority Applications (1)

Applications Claiming Priority (4)

Publications (2)

Family

ID=39527035

Family Applications (1)

Country Status (3)

Cited By (1)

Families Citing this family (27)

Family Cites Families (26)

• 2007
  • 2007-12-13 US US11/956,141 patent/US8089970B2/en not_active Expired - Fee Related
  • 2007-12-14 WO PCT/US2007/087677 patent/WO2008076945A2/en not_active Ceased
  • 2007-12-14 JP JP2009541626A patent/JP5027886B2/ja active Active

Also Published As

Similar Documents

Legal Events