Design Factors

1 The 6-addresses scheme

Since 802.11s is based on 802.11, its frame format is therefore very similar. We will not discuss 802.11 frames here but more details can be found in [8]. Fig 4 shows an 802.11s frame with the corresponding fields. The mesh control field is located at the beginning of the body field of the 802.11 frames. It can be between 6 to 24 octets long and has the following fields:

 A 1 octet mesh flag where 2 bits indicates if the address extension mode is used, and 6 bits are reserved for future use

 A 1 octet Mesh Time to Live (TTL), to indicate when to drop a packet if it cannot reach destination

 A 4 octets Mesh sequence number, to suppress duplicate frames

 A Mesh Address Extension field, between 0 to 18 octets in case of address extension mode

Indeed, 802.11s allows for an extension of up to three more addresses in the Mesh Control Field.

These extra three addresses allow for the mesh network to carry data to and from networks outside the mesh as well as to and from stations inside the mesh. When the 6 address-headers are used, the

Figure 4 - 802.11s MAC Frame Format [9]

11

ordering of the addresses should be from the innermost to the outer-most connection. Figure 5 illustrates this concept:

 Addresses 1&2 are used between end-points of a link between a transmitter station and receiver station.

 Addresses 3&4 are used between end-points of a mesh path (between source and destination MAP, or between a MAP and a MPP)

 Addresses 5&6 are used between end-points of a 802.x communication (source STA to destination STA, inside or outside the mesh network)

Here is a simple example we created to show how the 6 addresses scheme works, in two different cases: intra-mesh flow and extra-mesh flow (Figure 6).

Figure 5 - The 6 addresses scheme

Figure 6 - 6 addresses scheme example

12

In A, STA1 sends a frame to STA2, who is inside the mesh network. STA1 will send a normal 802.11 frame. Address 1 is the address of MAP1, the MAP linked to STA1. Address 2 is the address of the source, which is its own address. Address 3 is the destination address of the frame, which is the address of STA2. MAP1 on the contrary will send a 802.11s frame, and will use the 6-addresses scheme. Addresses 1&2 will be its own address and the one from the next hop MAP/MP in the Mesh network, here MP2. Addresses 3&4 will be the addresses from the Mesh path between the two STAs.

That is its own address and the address of the MAP linked to STA2, MAP3. Addresses 5&6 will be two addresses from the STAs. At MP2, addresses 3,4,5,6 remain the same, since its still the same mesh path and source and destination addresses. Only addresses 1&2 change, since the link is different (link between MP2 and next hop, here MAP3). When the frame arrives at MAP3, MAP3 has to change it from a 802.11s frame to a classic 802.11 frame. It will only use the first 3 addresses. Address 1 is the recipient address, STA2, address 2 is the transmitter address, MAP3, and address 3 is the original sender of the frame, STA1.

In B, the process is the same until the frame reaches MPP3, the Mesh Portal. Because in Figure 3, the Mesh Portal is a gateway to an Ethernet network, the portal has to change the frame from a 802.11s frame to an 802.3 frame. Therefore, MPP3 makes the appropriate changes to the frame and forwards it to its 802.3 interface.

2 EDCA

EDCA (Enhanced Distributed Channel Access) [6] is a new distributed channel access protocol, built on legacy DCF (Distributed Coordination Function) protocol, currently utilized in 802.11-based Wireless Mesh Networks. Together with HCCA, it forms the new access control process developed for Wireless Mesh Network, known as HCF (Hybrid Coordination Function).

EDCA brings quality of service (QoS) features to the channel access process and the MAC layer, by classifying the soon-to-be forwarded data frame into 4 different access categories (AC) or traffic class (Fig 4): AC_BK (background traffic), AC_BE (best effort traffic), AC_VI (Video traffic) and AC_VO (Voice traffic). In each AC, the IFS (InterFrame Space), known as AIFS (Arbitrary IFS) and the contention windows (CW) will be set to a different value, depending on the priority of the AC.

The time is divided in slots (time slots) of duration 20μs and CW and AIFS are counted in number of slots. AC_VO has the highest priority and by default, its AIFS is 2, CWmin is 7, CWmax is 15.

AC_VI has the second highest priority, with default AIFS of 2, CWmin of 15 and CWmax of 31.

AC_BK has the third priority and has AIFS of 3, CWmin of 31 and CWmax of 1023. Finally, AC_BE has the lowest priority, with AIFS of 7, CWmin of 31 and CWmax of 1023.

13 Here is how a station accesses the medium with EDCA:

 First, the station senses the medium during AIFS time. If the medium is sensed as idle, the station initiates the back-off procedure for transmission, if not, the station performs another back-off procedure in CW before sensing the medium again during AIFS. High priority ACs have a small AIFS times, while low priority ACs have longer AIFS times

 The station then performs a back-off procedure, with a starting time set by the CW. The starting time cannot be lower than CWmin, or higher than CWmax. The back-off procedure can be frozen if the medium becomes suddenly non-idle, and can be reactivated if the medium is sensed as idle for AIFS time. In case of retransmission, the starting time of the back-off process is increased exponentially (the value still being between CWmin and CWmax). If the back-off procedure starts at CWmax and the transmission fails, the packet is retransmitted as many times as necessary until it reaches the maximum number of possible tries (dot11ShortRetry Limit or dot11LongRetryLimit). The station then discards it. ACs with high-priority traffic have a low CW, and ACs with low high-priority traffic have higher CW

 At the end of the back-off procedure, the station is awarded a transmission opportunity (TxOp). It is a period where the station enjoys contention free access to the medium and can transmit either one or as many packets as possible during TxOp. The TxOp period is set depending on the priority of the AC.

 At the end of a TxOp, the station waits for AIFS and starts another back-off procedure to get another TxOp.

Figure 7 - Transmission queues in EDCA

14

On top of that, there is also contention process within each station, as there are four transmission queues corresponding to the four AC. This process is called virtual contention and is quite simple: the queue with the highest priority traffic gets the transmission opportunity, the other queues then perform a back-off procedure to try to get another transmission opportunity (Figure 7). Therefore, two types of collision can be experienced in EDCA.

The main difference between DCF and EDCA is that EDCA classifies the traffic in four categories, each having its own set of AIFS and CW values. DCF set the same IFS (called DIFS) and CW values for all the traffic.

3 MCCA

The 802.11s Mac layer implements EDCA by default, with an option called MCCA (Mesh Coordinated Channel Access). MCCA provides contention-free and guaranteed channel access to stations during reserved periods. Here is how MCCA works.

In order to transmit data, a MCCA-enabled node has to reserve transmission opportunities, or MCCAOPs (MCCA OPportunities). A MCCAOP happens during a DTIM (Delivery Traffic Indication Message) interval, the time interval between two DTIM Beacon, and is defined by three parameters: its duration (time duration of the MCCAOP), its offset (position of the first MCCAOP during the DTIM interval) and its periodicity (how many MCCAOP during a DTIM interval). The DTIM interval is divided in slots of 32 μs (Figure 8).

The sender of the request is called the MCCAOP owner (or owner), and the receiver is called the MCCAOP responder (or responder). To set up a reservation, the owner will send a MCCAOP request to the responder, containing the three parameters (offset, duration, periodicity) that it chose for this particular request. Once the responder has received the request, it will check if it is ok or not to accept the request, if there might be a conflict with other MCCAOP reservations. It will send back a response

Figure 8 - Example of MCCAOP reservation with Offset = 7, Duration = 6, Periodicity = 2 [7]

15

with accept or reject code. Every time an owner wants to communicate with a different responder, it has to send a request to this node with specific offset, duration and periodicity parameters.

To reduce the probability of reservation conflict, a node periodically advertises its reservations and the ones of its neighbors. It will generate and transmit the following two elements: a MCCAOP Advertisement Overview Information Element (IE) and a MCCAOP Advertisement IE. The MCCAOP Advertisement Overview IE contains three elements: the MCCA Access Fraction (MAF), the MAF limit and the accept reservation bit. The MAF value is used to limit the number of MCCAOP reservation per group of nodes. It is a percentage calculated by every node, and is the ratio of the duration of all MCCAOPs in the group of nodes by the duration of the DTIM interval. The MAF limit is the maximum value a group of nodes can reach. Upon acceptance of new MCCAOPs, a node will calculate its possible new MAF. If it exceeds the MAF limit, it will reject the request. This is used to allow non-MCCA nodes to access the medium. The MCCAOP Advertisement IE contains three reports: the TX-RX, the Broadcast and the Interfering times (IR). The TX-RX report contains all the MCCAOPs of a node that are individually addressed to it (node is owner or responder). The Broadcast report contains all the MCCAOPs addressed to multiple nodes (node can be owner or one of the responder). The IR report contains all the MCCAOPs of the neighbors of the nodes (node is not involved). With these reports, a node can build a map of its neighborhood MCCAOPs times.

MCCAOP Advertisement Overview IE and MCCAOP Advertisement IE are transmitted during beacon times, for instance during DTIM and TIM beacon in Fig 1. The standard requires that between two DTIM beacons, there must be at least one TIM beacon. It allows nodes to be the most up to date with the MCCAOPs in their neighborhood.

One might be wondering how MCCA interacts with EDCA. Since MCCA is an option of EDCA, it works on top of EDCA. It means that in order to transmit a packet during a MCCAOP, a node must first obtain a EDCA TxOp. To do so, the owner of a MCCAOP has the highest priority among all other nodes, in other words the lowest contention windows. Then when its back-off procedure reaches 0, it will start the transmission.

16 3 Routing

In this section we will first introduce the metric used for the routing as well as a brief explanation on how the routing is done in 802.11s Mesh networks. The part is given for reference purposes only, as it is not directly related to the core content of this thesis. However, we estimated that it should be included since it is a major component of 802.11s.

(1) The Airtime metric

The default routing metric specified in [24] is the Airtime metric. This metric calculates the amount of channel resources (ca) used when transmitting a frame over a particular link. It is calculated by sending a test frame and finding the Airtime cost with the following formula (Figure 9):

where c is the Airtime cost, Oca is the Channel access overhead, Op is the protocol overhead, Bt is the test frame size (in bits), r is the data rate (in Mbps) and ept the frame error rate (for a test frame). r is the data rate at which the node would transmit the test frame, based on the current conditions, and depends on the local implementation of rate adaptation. Ept is the probability that when the test frame is sent, it would be corrupted because of transmission errors. As shown by the formula, the lower the airtime cost, the better.

The airtime metric only takes into account the transmission rate and the error rate for the quality of the link. However, in reality, the quality of a link does not depend only on those two factors but also on the behavior of the receiving node. Indeed, using a link with low error rate and good transmission quality doesn’t mean that the frame will arrive at destination, since it could be dropped at the receiving node if it is congested. Airtime will always privilege link with higher data rate no matter the actual state of the receiving node.

However, the amendment allows for the use of other metric, such as EFT [10] or WCETT [11], as long as all the nodes of a mesh network use the same one. If not, Airtime will be used.

Figure 9 - The Airtime metric formula

17 (2) HWMP

The default routing protocol for 802.11s mesh network, defined in [24] is HWMP or Hybrid Wireless Mesh Protocol. The term “Hybrid” comes from the fact that this protocol combines an on-demand distributed path selection protocol (based on the Ad-Hoc On-Demand Distance Vector protocol or AODV [12]) called Radio-Metric AODV (RM-AODV) [13] with a pro-active tree-oriented routing approach. While AODV works with layer 3 IP addresses and uses hop-count as a routing metric, RM-AODV works with layer 2 MAC addresses and uses a radio-aware routing metric for path selection.

In this section we will describe the main principles behind HWMP, more details can be found in [13]

and [14].

The On-Demand path setup is established by a path discovery mechanism quite similar to the one in AODV. Whenever a Mesh STA needs a path to a destination, it broadcasts a path request message (PREQ) into the mesh network. Similarly the destination will reply with a path reply message (PREP) whenever it receives a PREQ message. Moreover, all Mesh STAs will update their routing tables whenever the received PREQ message correspond to a new or better path from the emitting Mesh STA. Depending if the destination only (DO) flag is set or not, intermediate Mesh STAs can respond to a PREQ message, instead of the destination.

The proactive part of HWMP is set up when one Mesh STA, usually a MPP will periodically broadcast mesh portal announcements (RANNs). The Mesh STA then becomes the root node. The root node can be selected by configuration (manually) or with a selection process (dynamically). Once a root node is established, a routing tree can be built within the network. This tree can be maintained proactively or not, using the registration and non-registration modes respectively. If the registration mode (or proactive tree) is used, Mesh STAs that receive RANNs are registered proactively at the root portal. Upon the reception of a RANN, the Mesh STA sends a route request message (RREQ) to the root, for revalidating the path to the root node. The Mesh STA chooses the neighboring Mesh STA with the best airtime metric to send the RREQ. Once the path is revalidated, the Mesh STA sends a RREP (route reply) message to the root node, to register itself and its associated 802.11 STAs. The topology maintenance is achieved by using directed, unicast RREQ message to Mesh STAs periodically.

The non-registration mode is a “lightweight” version of HWMP topology formation where the routing overhead is kept to a minimum. In this case, when a Mesh STA receives a RANN, it updates its routing table by adding or updating an entry. However, updating an entry in the routing table is only performed when there is newer or better path information. If the received path information isn’t better than the one already in the table, the Mesh STA performs no action. In non-registration mode, the

18

update of the routing table by the Mesh STAs is the only action happening. There is no registration at the root node, hence the name.

The following figure sums up the basic features of HWMP (Figure 10):

Figure 10 - HWMP basic features

19