Professional Documents
Culture Documents
Sender
Receiver
Packet
Packet
CS SAP
MAC
SDU
Service-Specific
Convergence Sublayers
SDU
MAC SAP
PDU
PDU
Privacy Sublayer
PHY
PHY SAP
PDU
PHY LAYER
Airlink
Figure 1.1: WiMAX Protocol Architecture
PDU
An IP Packet PDU arrives at the WiMAXs MAC layer is called a MAC SDU.
Classification
Classification is a process by which a MAC SDU is mapped onto a particular
connection for transmission between MAC peers. The mapping process associates a
MAC SDU to a connection (with a CID) and to the service flow characteristics (i.e. a
list of QoS parameters) of that connection which provide the QoS for the packet.
A classifier is a set of matching criteria applied to each packet entering WiMAX
network. It consists of some protocol-specific packet matching criteria (e.g.
destination IP address), a priority, and a reference to a CID. If matched, the packet is
then delivered to the SAP for delivery to the connection identified by the CID.
Implementation of each specific classification capability (e.g. IPv4-based
classification) is optional. Several classifiers may refer to the same service flow. In
this case the classifier priority is used to order the application of classifiers to a
packet. Downlink classifiers are applied by the BS to packets it is transmitting, and
uplink classifiers are applied at the SS.
Priority of Classification Parameters
service flow (encoded using type/length/value formatting scheme)
802.3
802.1P/Q
IPv4/v6
TCP/UDP
1 byte
TLV1
TLV2
TLV3
TLV4
CID
SFID
2 and
bytes
and
4
SDU (or CS PDU) to become a MAC PDU. Usually TCP traffic will arrive
encapsulated within an IP packet which in turn is encapsulated within an Ethernet
frame.
PHSI (optional)
Packet PDU
(a) A MAC SDU
e.g.
PHSI 0
CRC (opt)
1.3
Between the PHY and MAC layers is a TC sublayer which transforms variable-length
MAC PDUs into fixed-length FEC blocks (plus possibly a shortened FEC block, i.e. a
PAD, at the end of each burst). The TC PDU is sized to fit in the FEC block currently
being filled. It starts with a pointer byte at the start of the payload indicating where
the next MAC PDU header starts within the FEC block. This allows resynchronization to the next MAC PDU in the event that previous FEC block had
irrecoverable errors. The TC layer can be considered to be a sublayer of the PHY.
Pointer
Transmission
MAC PDUs are transmitted in the PHY bursts each may carry many concatenated
MAC PDUs spanning over multiple FEC blocks. The FEC used in WiMAX is RS
GF(256) with variable block size and error correction capabilities. This is paired with
an inner block convolutional code to robustly transmit critical data such as frame
control and initial accesses.
Basic connection for short and critical MAC and RLC messages,
Primary management connection for longer and more delay-tolerant
messages such as authentication and connection setup. The secondary
management connection transfers standards-based messages such as DHCP,
TFTP, and SNMP.
Transport connections are unidirectional to facilitate different UL and DL
QoS and traffic parameters.
Each SS has a standard 48-bit IEEE MAC address, which serves as an equipment
identifier because the primary addresses used during operation are the CIDs. BS has
48-bit ID, not MAC address, of which is a 24-bit operator indicator.
Figure 2.2: Generic MAC header format (HT: header type, EC: encryption control,
EKS: encryption key sequence, CI: CRC indicator, CID: connection ID, HCS: header
check sequence)
MAC Subheaders
Incoming MAC SDUs from corresponding convergence sublayers are re-formatted
according to the MAC PDU format, with necessary fragmentation and/or packing,
before being transmitted over one or more user connections in accordance with the
MAC protocols. On arriving at the peer MAC Layer at the receiver, MAC PDUs are
re-constructed back to the original MAC SDUs to be delivered to the MAC
Convergence Sublayers. WiMAX incorporates the fragmentation/packing process
with the bandwidth allocation process to maximize flexibility, efficiency, and
effectiveness of both processes.
Fragmentation is the process in which a MAC SDU is divided into many MAC SDU
fragments for transporting in multiple MAC PDUs. Each connection may be in only
one single-fragment state at any time. Packing is the process in which multiple MAC
SDUs are packed into a single MAC PDU payload. Either a BS or an SS may initiate
both processes. WiMAX allows simultaneous fragmentation and packing for efficient
use of bandwidth.
Except for bandwidth request MAC PDUs which contains no payload, MAC PDUs
contain either MAC management messages or convergence sublayer data, and
therefore have 3 types of MAC subheaders:
The grant management and fragmentation subheaders may be inserted in MAC PDUs
immediately following the generic header, and the packing subheader may be inserted
immediately before each MAC SDU; all these insertions will be indicated by the Type
field in the generic header. The use of subheaders can save up to 10% of system
bandwidth.
An FCH burst
DL burst #1
Authentication and
registration
Optional
Sink 1
SS1
802.16
BS
SS2
Internet
TCP
SSN
N
SS3
Sink 2
WiMAX network
Figure 4.1: A typical broadband wireless access (BWA) scenario using WiMAX
4.1 Bandwidth request-grant mechanisms
When an SS wishes to transmit, it sends a request message to the BS informing the
latter of its class and traffic specifications (see Table 4.1 below). The BS collects
requests from all SSs then performs bandwidth allocation to the SSs using appropriate
scheduling (granting) schemes in order to satisfy the QoS requirements of the SSs.
Since the BS controls the access to the network in the UL, upon requests from the SSs
it grants bandwidth to the latter on demand. There are four request-granting
mechanisms used for bandwidth allocation for UL:
(i)
(ii)
(iii)
(iv)
To date, there are only a few ns-2 simulator modules for WiMAX MAC layer
simulation, e.g. [4], [7], and [8].
4.3 Service Classes and QoS Parameters
Since the BS has the knowledge of all requests and available resources, DL
scheduling is much simpler than the UL scheduling. Neither DL nor UL Scheduling
nor Admission Control and Traffic Policing schemes are specified in the 802.16
However, the
Unsolicited Grant Service (UGS) has the highest priority and is used to support CBR
traffic with strict delay requirements such as PCM voice and VoIP (without silence).
Since these packet services are synchronous (i.e. fixed length, fixed data rate), they
are not required to request bandwidth for each packet, but instead the BS periodically
assigns slots according to traffic specifications once the connection is established.
The grant size is calculated by the BS based on the minimum reserved traffic rate
averaged over time, and the grant interval follows closely the packet arrival pattern
plus an offset upper bounded by the tolerated jitter. Therefore scheduling for DL
UGS traffic is not required nor is necessary.
QoS specification parameters for the traffic are therefore: minimum reserved traffic
rate, maximum delay/latency, and tolerated jitter.
Real-time Polling Service (rtPS) has the next highest priority and is used to support
VBR traffic (variable-size data packets but generated periodically) with less stringent
delay requirements such as MPEG multimedia streams and VoIP with silence. Since
the size of arriving packets is not fixed in rtPS, BS has to poll the connection of this
class periodically (unicast polling) to ask how much bandwidth is needed.
Traffic QoS specification parameters are: minimum reserved traffic rate, maximum
sustained traffic rate, and maximum delay/latency.
Non real-time Polling Service (nrtPS) , for traffic without any specific delay
requirement, e.g. FTP traffic, has higher priority than the lowest priority BE class. It
is the same as BE class except it may have additional bandwidth allocated through
non periodic polling. nrtPS therefore, is not likely to be starved.
Traffic specification parameters are: minimum reserved traffic rate, maximum
sustained traffic rate, and traffic priority.
Best Effort (BE) class is used to support traffic with no QoS such as HTTP and
TELNET traffic. This class is allowed to use only contention-based bandwidth
request. There is a fixed number of mini slots in the UL subframe that all SSs have to
share to place their requests (Figure 4.2). An increase in the number of slots can only
be made at a decrease in the bandwidth available for the transmission of the actual
data. There is possibility that BE traffic is starved by the lack of bandwidth.
Collisions of requests can corrupt some slots and contention back-off can cause
serious delay to some SSs.
Traffic specification parameters are: maximum sustained traffic rate, and traffic
priority.
Applications
VoIP
For CBR or CBR-like
service flows (SFs), e.g.
T1/E1
QoS Specifications
Max sustained rate
Max latency tolerant
Jitter tolerance
activity
nrtPS
For nrtSFs with better than
Non-real-time
Polling best-effort service, e.g.
Service
bandwidth-intensive file
(contention
requests transfer
allowed,
polled
less
frequently)
BE
For BE traffic, e.g. generic
Best-Effort Service
traffic, data transfer, web
(contention
requests browsing, etc.
allowed,
QoS
not
guaranteed)
Voice
with
detection (VoIP)
present at the queue and never misses the opportunity. Packets from intermittent
traffic such as email is quite often absent at the queue, thus missing the RR polling.
Suppose the BS allocates a total amount of resource total to N SSs each requesting a
amount i. Under a certain allocation scheme each SS is allocated an amount i, so
N
that total i . Suppose there is a fair share amount fair that all SSs should
i 1
receive. The max-min fairness criterion states that an allocation is fair if:
(i) no user receives more than its request, i.e. i min( fair , i ) ,
(ii) no other allocation scheme satisfying condition (i) has a higher minimum
allocation, and
(iii)
Condition (2) remains recursively true as we remove the minimal user
and reduce the total resource accordingly total ( total min )
In [1] a FQ algorithm is proposed.
Since UGS and rtPS classes have strict delay and bandwidth constraints, scheduling
for them is usually fairly deterministic and performance relies more on admission
control rather than on scheduling. Existing real-time (deterministic) scheduling
schemes such as Weighted Fair Queuing (WFQ) [1] and Weighted Round-Robin
(WRR) [2] have proved to be very suitable for these two classes. Priority is used by
some authors for the scheduling weight of the flow class, e.g. UGS (1), rtPS (2) ,
nrtPS (3), BE (4), in Multi-class Priority Fair Queue (MPFQ) scheduling. Others use
the percentage connection bandwidth of a connection to express its weight in a WFQ
scheduling. Higher bandwidth is given higher priority or weight. Two-tier, e.g.
MPFQ followed by WFQ, scheduling has been proposed by some authors.
This is called normalised bandwidth asymmetry [5] [6]. Thus the bandwidth
asymmetry is less serious because of the TCP ACKs (typically 40 bytes) is much
shorter than data packets. TCP behaves normally when k=<1.
Since reliable transport protocols such as TCP uses the arrival rate of ACKs in the
reverse channel to control packet flow in the forward direction, congestion in the
reverse channel may lead to poor performance of the TCP traffic. Their performance
is affected by the asymmetry of the network. Furthermore, TCP is ack-clocked, i.e.
relying on the arrival of ACKs to make the next progress and fully utilize the
available bandwidth of the path. There are many versions of TCP congestion control
and error control but they are all based on the sliding window mechanism, slow start,
congestion avoidance and fast retransmit or fast recovery. Since TCP increases the
window size by counting the rate of ACKs received, a small bandwidth ACK path (i.e.
slow rate of ACKs arrival) can significantly slow down the growth of the TCP sender
window during a slow start, independent of the link bandwidth in the direction of data
transfer. Also in this situation, the reverse channel becomes congested and its buffer
can overflow, loosing ACK packets, hence seriously affecting the performance of
TCP. This is the situation in most wireless networks including WiMAX. Note that in
WiMAX, the ratio of DL:UL bandwidth can be adjusted by the network operator, in a
10ms TDD frame, DL:UL = 2ms:8ms for example. Frame size affects the TCP roundtrip time (RTT). The one-way end-to-end delay is roughly equal to the frame size
plus a fixed delay representing delays in other parts of the network from the SS to the
BS, and is determined during initial ranging. The maximum tolerable delay in an
802.16 network is specified as 40ms.
In wireless networks variable delay paths can make the senders estimate of the
round- trip time (RTT) inaccurate and therefore increases the TCPs retransmission
timeout value. In WiMAX, it is difficult for the SSs to estimate the bandwidth
required for TCP flow to make request to the BS, due to the dynamic changes of the
TCP sending rate under congestion control. Bandwidth request can be omitted and is
replaced by some simple parameter (e.g. recent sending rate of packet connection [3]).
TCP performance over DOCSIS (Data-over-Cable Service Interface Specification)
has been studied recently for CATV in [6]. Since 802.16 MAC layer protocol is
almost identical to DOCSIS MAC layer protocol, we expect similar effect of WiMAX
MAC layer scheduling on the performance of TCP traffic as is in DOCSIS. The BS
and SSs in WiMAX correspond to the CMTS (Cable Modem Termination System)
and CMs (Cable Modems) respectively, in CATV. The TCP-over-DOCSIS analysis in
[6] is carried out for TCP Reno using delayed ACK mechanism for congestion
control.
References
[1] A. Demers et al., Analysis and Simulation of a Fair Queuing Algorithm, ACM
SIGCOMM Computer Communications Review, vol. 19, 1989.
[2] M. Katevenis et al., Weighted Round Robin Cell multiplexing in a GeneralPurpose ATM Switch Chip, IEEE Journal on Selected Areas in Communications,
vol. 9, no. 8, pp.1265-1279, Oct. 1991.
[3] Seungwoon Kim and Ikjun Yeom, Performance Analysis of Best Effort Traffic
in IEEE 802.16 Networks, IEEE Transactions on Mobile Computing, 2007.
[4] An ns-2 module for simulating WiBro based on IEEE 802.16, August 2006,
available at
http://cnlab.kaist.ac.kr
[5] H. Balakrishnan et al., The effects of asymmetry on TCP performance, Mobile
Networks and Applications, vol. 4, pp.219-241, 1999.
[6] Wanjiun Liao, The Behavior of TCP Over DOCSIS-Based CATV Networks,
IEEE Transactions on Communications, vol.54, no. 9, pp.1633-1642, Sept. 2006.
[7] Jenhui Chen et al., The Design and Implementation of WiMAX Module for ns-2
Simulator, Taiwan 2006
[8] Richard Rouil, The Network Simulator NS-2 NIST add-on IEEE 802.16
model (MAC+PHY), National Institute of Standards and Technology, September
2006
[9] G.S. Chu, D. Wang, and S. Mei, A QoS Architecture for the MAC Protocol of
IEEE 802.16 BWA System, IEEE Conference on Communications, Circuits, and
Systems and West Sino Expositions, pp. 435-439, July 2002
[10] G.S. Paschos et al, A heuristic Strategy for IEEE 802.16 WiMAX scheduler
for Quality of Service,
[11] Alexander Sayenko et al., Ensuring the QoS Requirements in 802.16
Scheduling, Proceedings International Workshop on Modeling Analysis and
Simulation of Wireless and Mobile Systems MSWiM06, pp.108-117, October 2006
Malaga, Spain
[12] C. Cicconetti et al., Quality of Service Support in IEEE 802.16 Networks,
IEEE Network, pp. 50-55, Mar/Apr 2006
Professor Thong Nguyen
Sydney 16 February 2007