Medium Access Sublayer

Medium Access Sublayer

Medium Access Sublayer
Network Layer Logical Link Control Sublayer Medium Access Sublayer Data Link Layer Physical Layer

IEEE 802 Standards

Medium Access Sublayer (cont’d)
Medium access (MAC) sublayer is not important on point-to-point links The MAC sublayer is only used in broadcast or shared channel networks Examples: Satellite, Ethernet, Cellular

Logical Link Control Sublayer
Logical Link Control (LLC) sublayer provides mechanisms for reliable communications acknowledgement, etc

MAC Sublayer: Contents
Fixed Assignment Protocols Demand Assignment Protocols Contention Access Protocols IEEE 802 LANs DQDB and FDDI

Classification of Multiple Access Protocols
Contention Access Fixed Access Demand Access Random access Collision resolution Polling, Reservation, Token TREE, WINDOW, etc FDMA, TDMA, CDMA, Token Bus, etc ALOHA, CSMA, BTMA, etc BTMA: Busy Tone Multiple Access

1. Fixed Assignment Protocols
Static and predetermined allocation of channel access: independent of user activity Idle users may be assigned to the channel, in which case channel capacity is wasted Examples: TDMA, FDMA, WDMA

2. Demand Assignment Protocols
Allocate channel capacity to hosts on a demand basis (i.e., only to active users) Requires methods for measuring the demand for the channel Polling Reservation schemes Token Passing Scheme

2.1 Polling A central controller interrogates each host and allocates channel capacity to those who need it Good for systems with: Short propagation delay Small polling messages Non-bursty traffic

2.2 Reservation Schemes Hosts independently reserve the channel for period of time Reservations are usually piggybacked on data messages passing along the channel Good for systems with : short propagation delay no central controller node non-bursty traffic

Reservation Protocols (cont’d)
Reservation protocol examples: Bit-Map Protocol Binary Countdown Protocol

2.2.1 Bit-Map Protocol Contention and data transmission periods alternate The contention period is divided into slots, with 1 bit-wide slot for each host in the network. If a host wants to transmit a packet, it sets its contention slot equal to 1. Otherwise, it sets it to 0. The slots pass all hosts in sequence, so every host is aware of who will transmit 1 1 1 1 2 4 5 6 1 1 4 data frame

Bit-Map Protocol (cont’d)
But what if there are a large number of hosts in the network? The contention period will have to grow to include them all With a large number of hosts, the contention period may be very long, leading to inefficiency

2.2.2 Binary Countdown Protocol
During contention period, each host broadcasts its binary address one bit at a time, starting with the most significant bit bits transmitted simultaneously are boolean OR’d together Arbitration rule: If a host sent a zero bit but the boolean OR results in a one bit, the host gives up and stops sending address bits Whichever host remains after the entire address has been broadcast gets access to the medium

Binary Countdown (cont’d)
Host Addresses Bit Time

Binary Countdown: Fairness
Stations with the highest addresses will always win. This is good if you want to implement priority, but bad if you want to give all hosts fair access to the channel Solution: Change the address of a host after a successful transmission

Binary Countdown: Permuting Addresses
After host A successful transmission, all hosts with addresses less than host A add one to their address, improving their priority Host A changes its address to zero, giving it the lowest priority

3. Token Passing Scheme A token always circulates around a ring net.
A user grabs a token to transmit data Will discuss details later in the Token Ring LAN

3. Contention Access Protocols
Single channel shared by a large number of hosts No coordination between hosts Control is completely distributed Examples: ALOHA, CSMA, CSMA/CD

Contention Access (cont’d)
Advantages: Short delay for bursty traffic Simple (due to distributed control) Flexible to fluctuations in the number of hosts Fairness

Contention Access (cont’d)
Disadvantages: Low channel efficiency with a large number of hosts Not good for continuous traffic (e.g., voice) Cannot support priority traffic High variance in transmission delays

Contention Access Methods
Pure ALOHA Slotted ALOHA CSMA 1-Persistent CSMA Non-Persistent CSMA P-Persistent CSMA CSMA/CD

3.1 Pure ALOHA Originally developed for ground-based packet radio communications in 1970 Goal: let users transmit whenever they have something to send

The Pure ALOHA Algorithm
1. Transmit whenever you have data to send 2. Listen to the broadcast Because broadcast is fed back, the sending host can always find out if its packet was destroyed just by listening to the downward broadcast one round-trip time after sending the packet 3. If the packet was destroyed, wait a random amount of time and send it again The waiting time must be random to prevent the same packets from colliding over and over again

t : one packet transmission time
Pure ALOHA (cont’d) Note that if the first bit of a new packet overlaps with the last bit of a packet almost finished, both packets are totally destroyed. t : one packet transmission time Vulnerable period: 2t t t0 t0+t t0+2t t0+3t Vulnerable

Pure ALOHA (cont’d) Due to collisions and idle periods, pure ALOHA is limited to approximately 18% throughput in the best case Can we improve this? G = mean Transmission Attempts per Frame T = frame time Zero transmit In (t-T, t+T)

3.2 Slotted ALOHA Slotted ALOHA cuts the vulnerable period for packets from 2t to t. This doubles the best possible throughput from 18.4% to 36.8% Slotted ALOHA is used in low-data-rate tactical satellite communications networks by military forces, in subscriber-based satellite communications networks, mobile telephony call setup, set-top box communications and in the contactless RFID technologies.

The Slotted ALOHA Algorithm
1. If a host has a packet to transmit, it waits until the beginning of the next slot before sending 2. Listen to the broadcast and check if the packet was destroyed 3. If there was a collision, wait a random number of slots and try to send again Note: The undesired user still Intends to transmit in Poisson Points. But it has to wait until the beginning of a slot. Therefore, if the desired user Starts sending at T, vulnerable period Will be [0,T] which maps to start time Of T.

3.3 CSMA We could achieve better throughput if we could listen to the channel before transmitting a packet This way, we would stop avoidable collisions. To do this, we need “Carrier Sense Multiple Access,” or CSMA, protocols

Assumptions with CSMA Networks
1. Constant length packets 2. No errors, except those caused by collisions 3. No capture effect 4. Each host can sense the transmissions of all other hosts 5. The propagation delay is small compared to the transmission time

CSMA (cont’d) There are several types of CSMA protocols:
1-Persistent CSMA Non-Persistent CSMA P-Persistent CSMA

3.3.1 1-Persistent CSMA Sense the channel. If collision occurs,
If busy, keep listening to the channel and transmit immediately when the channel becomes idle. If idle, transmit a packet immediately. If collision occurs, Wait a random amount of time and start over again.

1-Persistent CSMA (cont’d)
The protocol is called 1-persistent because the host transmits with a probability of 1 whenever it finds the channel idle.

The Effect of Propagation Delay on CSMA
packet A B carrier sense = idle Transmit a packet Collision

Propagation Delay and CSMA
Contention (vulnerable) period in Pure ALOHA two packet transmission times Contention period in Slotted ALOHA one packet transmission time Contention period in CSMA up to 2 x end-to-end propagation delay Performance of CSMA > Performance of Slotted ALOHA > Performance of Pure ALOHA

1-Persistent CSMA with Satellite Systems
Satellite system: long prop. delay (270 msec) Carrier sense makes no sense It takes 270 msecs to sense the channel, which is a really long time Vulnerability time = 540 msec (1/2 a second is forever in a network!)

1-Persistent CSMA (cont’d)
Even if prop. delay is zero, there will be collisions Example: If stations B and C become ready in the middle of A’s transmission, B and C will wait until the end of A’s transmission and then both will begin transmitted simultaneously, resulting in a collision. If B and C were not so greedy, there would be fewer collisions

3.3.2 Non-Persistent CSMA Sense the channel. If collision occurs
If busy, wait a random amount of time and sense the channel again If idle, transmit a packet immediately If collision occurs wait a random amount of time and start all over again

Tradeoff between 1- and Non-Persistent CSMA
If B and C become ready in the middle of A’s transmission, 1-Persistent: B and C collide Non-Persistent: B and C probably do not collide If only B becomes ready in the middle of A’s transmission, 1-Persistent: B succeeds as soon as A ends Non-Persistent: B may have to wait

3.3.3 P-Persistent CSMA Optimal strategy: use P-Persistent CSMA
Assume channels are slotted One slot = contention period (i.e., one round trip propagation delay)

P-Persistent CSMA (cont’d)
1. Sense the channel If channel is idle, transmit a packet with probability p if a packet was transmitted, go to step 2 if a packet was not transmitted, wait one slot and go to step 1 If channel is busy, wait one slot and go to step 1. 2. Detect collisions If a collision occurs, wait a random amount of time and go to step 1

P-Persistent CSMA (cont’d)
Consider p-persistent CSMA with p=0.5 When a host senses an idle channel, it will only send a packet with 50% probability If it does not send, it tries again in the next slot. The average number of tries is: 1 8 S i (1-p)i-1 p = p i=0

Comparison of CSMA and ALOHA Protocols
(Number of Channel Contenders)

3.4 CSMA/CD In CSMA protocols If two stations begin transmitting at the same time, each will transmit its complete packet, thus wasting the channel for an entire packet time In CSMA/CD protocols The transmission is terminated immediately upon the detection of a collision CD = Collision Detect a collision (excess current above what it is generating, i.e. > 24 mA for coaxial Ethernet)

CSMA/CD Sense the channel Collision detection
If idle, transmit immediately If busy, wait until the channel becomes idle Collision detection Abort a transmission immediately if a collision is detected Try again later after waiting a random amount of time

CSMA/CD (cont’d) Carrier sense Collision detection
reduces the number of collisions Collision detection reduces the effect of collisions, making the channel ready to use sooner

Collision detection time
How long does it take to realize there has been a collision? Worst case: 2 x end-to-end prop. delay packet A B

CSMA/CA ( = WiFi) If we only sense the medium, we may not use the available chance. (We should care about whether our destination is receiving at the time or not)

CSMA/CA ( = WiFi)

CSMA/CA ( = WiFi) There is a field in the header that determines the length of the packet (RTS/CTS) All nodes receiving RTS will know approximately how long the current transmission will take

CSMA/CA ( = WiFi) If there is collision, the unsuccessful transmitter (i.e. one that does not hear CTS within expected time) waits a random amount of time and tries again later. The algorithm is called binary exponential backoff

4. IEEE 802 LANs LAN: Local Area Network What is a local area network?
A LAN is a network that resides in a geographically restricted area LANs usually span a building or a campus

Characteristics of LANs
Short propagation delays Small number of users Single shared medium (usually) Inexpensive

Common LANs Bus-based LANs Ring-based LANs Switch-based LANs
Ethernet (*) Token Bus (*) Ring-based LANs Token Ring (*) Switch-based LANs Switched Ethernet ATM LANs (*) IEEE 802 LANs

IEEE 802 Standards 802.1: Introduction
802.2: Logical Link Control (LLC) 802.3: CSMA/CD (Ethernet) 802.4: Token Bus 802.5: Token Ring 802.6: DQDB 802.11: CSMA/CA (Wireless LAN)

IEEE 802 Standards (cont’d)
802 standards define: Physical layer protocol Data link layer protocol Medium Access (MAC) Sublayer Logical Link Control (LLC) Sublayer

OSI Layers and IEEE 802 OSI layers IEEE 802 LAN standards
Higher Layers Higher Layers 802.2 Logical Link Control Medium Access Control Data Link Layer CSMA/CD Token-passing Token-passing bus bus ring Physical Layer

IEEE 802 LANs (cont’d) Ethernet Token Ring

4.1 Ethernet (CSMA/CD) IEEE 802.3 defines Ethernet
Layers specified by 802.3: Ethernet Physical Layer Ethernet Medium Access (MAC) Sublayer

Ethernet (cont’d) Possible Topologies: 1. Bus
2. Branching non-rooted tree for large Ethernets

Minimal Bus Configuration
Coaxial Cable Transceiver Terminator Transceiver Cable Host

Typical Large-Scale Configuration
Repeater Host Ethernet segment

Ethernet Configuration

4.1.1 Ethernet Physical Layer
Transceiver Transceiver Cable 4 Twisted Pairs 15 Pin Connectors Channel Logic Manchester Phase Encoding 64-bit preamble for sychronization

Ethernet Physical Configuration (for thick coaxial cable)
Segments of 500 meters maximum Maximum total cable length of 1500 meters between any two transceivers Maximum of 2 repeaters in any path Maximum of 100 transceivers per segment Transceivers placed only at 2.5 meter marks on cable

Manchester Encoding 1 bit = high/low voltage signal
Data stream Encoded bit pattern 1 bit = high/low voltage signal 0 bit = low/high voltage signal

Ethernet Synchronization
64-bit frame preamble used to synchronize reception 7 bytes of followed by a byte containing Manchester encoded, the preamble appears like a sine wave

Ethernet Cabling Options
10Base5: Thick Coax 10Base2: Thin Coax (“cheapernet”) 10Base-T: Twisted Pair 10Base-F: Fiber optic Each cabling option carries with it a different set of physical layer constraints (e.g., max. segment size, nodes/segment, etc.)

4.1.2 Ethernet: MAC Layer Data encapsulation Link Management
Frame Format Addressing Error Detection Link Management CSMA/CD Backoff Algorithm

MAC Layer Ethernet Frame Format
Multicast bit Destination (6 bytes) Source (6 bytes) Length (2 bytes) Data ( bytes) Pad Frame Check Seq. (4 bytes)

Ethernet MAC Frame Address Field
Destination and Source Addresses: 6 bytes each Two types of destination addresses Physical address: Unique for each user Multicast address: Group of users First bit of address determines which type of address is being used 0 = physical address 1 = multicast address

Ethernet MAC Frame Other Fields
Length Field 2 bytes in length determines length of data payload Data Field: between 0 and 1500 bytes Pad: Filled when Length < 46 Frame Check Sequence Field 4 bytes Cyclic Redundancy Check (CRC-32)

CSMA/CD Recall: CSMA/CD is a “carrier sense” protocol.
If channel is idle, transmit immediately If busy, wait until the channel becomes idle CSMA/CD can detect collections. Abort transmission immediately if there is a collision Try again later according to a backoff algorithm

CSMA/CD (cont’d) Carrier sense reduces the number of collisions
Collision detection reduces the impact of collisions

CSMA/CD and Ethernet Ethernet: Ethernet access protocol:
Short end-to-end propagation delay Broadcast channel Ethernet access protocol: 1-Persistent CSMA/CD with Binary Exponential Backoff Algorithm

Ethernet Backoff Algorithm: Binary Exponential Backoff
If collision, Choose one slot randomly from 2k slots, where k is the number of collisions the frame has suffered. One contention slot length = 2 x end-to-end propagation delay This algorithm can adapt to changes in network load.

Binary Exponential Backoff (cont’d)
slot length = 2 x end-to-end delay = 15 ms A B t=0ms: Assume A and B collide (kA = kB = 1) A, B choose randomly from 21 slots: [0,1] Assume A chooses 1, B chooses 1 t=30ms: A and B collide (kA = kB = 2) A, B choose randomly from 22 slots: [0,3] Assume A chooses 2, B chooses 0 t=45ms: B transmits successfully t=75ms: A transmits successfully

Binary Exponential Backoff (cont’d)
In Ethernet, Binary exponential backoff will allow a maximum of 15 retransmission attempts If 16 backoffs occur, the transmission of the frame is considered a failure.

4.1.4 Ethernet Performance

4.1.3 Ethernet Features and Advantages
1. Passive interface: No active element 2. Broadcast: All users can listen 3. Distributed control: Each user makes own decision Simple Reliable Easy to reconfigure

Ethernet Disadvantages
Lack of priority levels Cannot perform real-time communication Security issues

4.1.5 Ethernet Switching Recent development: Connect many Ethernet segments or subnets through an “Ethernet switch” to segment 4 to segment 1 to segment 3 to segment 2

Why Ethernet switching?
LANs may grow very large The switch has a very fast backplane It can forward frames very quickly from one segment to another Cheaper than upgrading all host interfaces to use a faster network

4.2 Token Ring IEEE 802.5 Standard Layers specified by 802.5:
Token Ring Physical Layer Token Ring MAC Sublayer

Token Ring (cont’d) Token Ring, unlike Ethernet, requires an active interface Host Ring interface

Token Ring Configuration

4.2.1 Token Ring Physical Layer
Ring Interfaces Listen and Transmit Modes Channel Logic Differential Manchester Encoding

Token Ring Interface Modes
Listen Mode Transmit Mode one-bit delay To station From station To station From station

Differential Manchester Encoding
Transitions take place at midpoint of interval 1 bit: the initial half of the bit interval carries the same polarity as the second half of the previous interval 0 bit: a transition takes place at both the beginning and the middle of the bit interval

4.2.2 Token Ring MAC Sublayer
Token passing protocol Frame format Token format

Token Passing Protocol
A token (8 bit pattern) circulates around the ring Token state: Busy: Idle:

Token Passing Protocol (cont’d)
General Procedure: Sending host waits for and captures an idle token Sending host changes the token to a frame and circulates it Receiving host accepts the frame and continues to circulate it Sending host receives its frame, removes it from the ring, and generates an idle token which it then circulates on the ring

Token Ring Frame and Token Formats
Bytes SD AC ED Token Format / / unlimited SD AC FC Destination Address Source Address Data Checksum ED FS Frame Format

Token Ring Delimiters SD = Starting Delimiter ED = Ending Delimiter
AC ED SD AC FC Destination Address Source Address Data Checksum ED FS SD = Starting Delimiter ED = Ending Delimiter They contains invalid differential Manchester codes

Token Ring Access Control Field
SD AC ED (Note: The AC field is also used in frames) P P P T M R R R P = Priority bits provides up to 8 levels of priority when accessing the ring T = Token bit T=0: Token T=1: Frame

Token Ring Access Control Field (cont’d)
SD AC ED P P P T M R R R M = Monitor Bit Prevents tokens and frames from circulating indefinitely All frames and tokens are issued with M=0 On passing through the “monitor station,” M is set to 1 All other stations repeat this bit as set A token or frame that reaches the monitor station with M=1 is considered invalid and is purged

Token Ring Access Control Fields (cont’d)
SD AC ED P P P T M R R R R = Reservation Bits Allows stations with high priority data to request (in frames and tokens as they are repeated) that the next token be issued at the requested priority

Token Ring Frame Control Field
SD AC FC Destination Address Source Address Data Checksum ED FS FC = Frame Control Field Defines the type of frame being sent Frames may be either data frames or some type of control frame. Example control frames: Beacon: Used to locate breaks in the ring Duplicate address test: Used to test if two stations have the same address

Token Ring Address & Data Fields
AC FC Destination Address Source Address Data Checksum ED FS Address Fields: Indicate the source and destination hosts Broadcast: Set all destination address bits to 1s. Data No fixed limit on length Caveat: Hosts may only hold the token for a limited amount of time (10 msec)

Token Ring Checksum and Frame Status
SD AC FC Destination Address Source Address Data Checksum ED FS Checksum: 32-bit CRC FS = Frame Status Contains two bits, A and C When the message arrives at the destination, it sets A=1 When the destination copies the data in the message, it sets C=1

The Token Ring Monitor Station
One station on the ring is designated as the “monitor station” The monitor station: marks the M bit in frames and tokens removes marked frames and tokens from the ring watches for missing tokens and generates new ones after a timeout period

Using Priority in Token Ring
If a host wants to send data of priority n, it may only grab a token with priority value n or lower. A host may reserve a token of priority n by marking setting the reservation bits in the AC field of a passing token or frame Caveat: The host may not make the reservation if the token or frame’s AC field already indicates a higher priority reservation The next token generated will have a priority equal to the reserved priority

When a new token is generated (i. e
When a new token is generated (i.e., when a sender finishes sending and releases an idle token), or when a sender sends a data frame, RRR is set to the lowest priority.

Priority Transmission: Example
B D C Host B has 1 frame of priority 3 to send to A Host C has 1 frame of priority 2 to send to A Host D has 1 frame of priority 4 to send to A Token starts at host A with priority 0 and circulates clockwise Host C is the monitor station (priority 0: lowest priority in this example)

Example (cont’d) Event Token/Frame AC Field
A generates a token P=0, M=0, T=0, R=0 B grabs the token and sets the message destination to A P=3, M=0, T=1, R=0 Frame arrives at C, and C reserves priority level 2. Monitor bit set. P=3, M=1, T=1, R=2 Frame arrives at D, and D attempts to reserve priority level 4: P=3, M=1, T=1, R=4 Frame arrives at A, and A copies it P=3, M=1, T=1, R=4 Frame returns to B, so B removes it, and generates a new token P=4, M=0, T=0, R=0 Token arrives at C, but its priority is too high. C reserves priority 2. M bit. P=4, M=1, T=0, R=2

Example (cont’d) Event Token/Frame AC Field
Token arrives at D, and D grabs it, sending a message to A P=4, M=0, T=1, R=0 Frame arrives at A, and A copies it P=4, M=0, T=1, R=0 Frame arrives at B, which does nothing to it P=4, M=0, T=1, R=0 Frame arrives at C, which sets the monitor bit P=4, M=1, T=1, R=2 Frame returns to D, so D removes it and generates a new token with P=2 P=2, M=0, T=0, R=0 etc… Attempt to complete this scenario on your own.

5. FDDI FDDI = Fiber Distributed Data Interface
Metropolitan Area Network (MAN)

Metropolitan Area Networks
MANs are like LANs, only bigger and faster MANs are typically used as campus-wide or regional computer networks LAN < MAN < WAN

FDDI Characteristics 100 Mbps data rate Distances of up to 200 km
Up to 1000 hosts attached Based on fiber optic cabling

FDDI Dual Ring Topology
Two fiber rings: Fiber channels are unidirectional, so 2 are needed

FDDI Fault Tolerance Fault Tolerance
Two rings can be merged into one after failure

FDDI Physical Layer Coding
4B5B (4 out of 5) coding is used Each 4-bit data message requires 5 bits to send Some of the unused 5-bit codes are used for controlling the fiber rings or for synchronization To obtain a 100 Mbps data rate, a 125 Mbps physical layer rate must be achieved

Each group of 4 bit MAC symbols is encoded as a group of 5 bits
16 (of 32 combinations) are for date 3 are for delimiters 2 are for control 3 are for hardware signaling 8 are unused (reserved for future use)

FDDI Frame Format Similar to Token Ring SD FC Data ED FS Preamble
> / / unlimited Preamble SD FC Destination Address Source Address Data Checksum ED FS

FDDI Token Passing Protocol
In Token Ring, a new token is not generated until the frame is received again at the transmitting host In FDDI, multiple tokens may be on the ring simultaneously. After an FDDI host transmits a frame, it may put another token on the ring Timers used to implement priority If a token is ahead of schedule (target token rotation time), all priorities may transmit. If it is behind schedule, only the highest ones may send.

Other FDDI timers Token Holding Timer Token Rotation Timer
determines how long a station may continue to transmit once it has acquired the token. Token Rotation Timer is restarted every time a token is seen. If this timer expires, it means that the token has not been sighted for too long an interval; possibly it has been lost. Token recovery procedure is initiated.

6. DQDB DQDB = Distributed Queue Dual Bus DQDB, like FDDI, is a MAN.
Unlike FDDI, DQDB is an IEEE standard: 802.6

DQDB (cont’d) Topology: Dual Bus Head End Host Host Host Host Head End

53-byte frame called a “cell”
DQDB Frame Format 53-byte frame called a “cell” Busy SLT PSR Rsvd REQ VCI (20 bits) Payl Typ Pri. HCS Seg Typ Sequence Num 53 bytes MID Information (44 bytes) Payload Length CRC-10 8 bits

DQDB MAC Sublayer Head-ends generate cells in both directions
To transmit, a host must know whether the destination is to its right or its left If right, the host must send on one bus If left, the host must send on the other bus A “Distributed Queue” is used to make sure that cells are transmitted on a first-come first-serve basis

Distributed Queueing Each host keeps two counters, CD and RC
Procedure for sending a cell: Set the request bit in a non-busy cell passing on the reverse bus. Copy the RC counter to the CD counter, and reset the RC counter to zero. As the request cell passes on the reverse bus, each host on the path observes it and increments its RC counter When an empty cell passes on the forward bus, the hosts decrement their RC counters and CD counters If RC and CD equal zero when an empty cell passes on the forward bus, it may be used to send the message

Distributed Queueing (cont’d)

A Brief Note: Fast Ethernet
IEEE 802.3u 100 Mbps Ethernet The MAC sublayer for Fast Ethernet is the same as for normal Ethernet Physical layer is slightly different: no more Manchester encoding (4B5B, 8B6T coding are used instead)

Medium Access Sublayer

Similar presentations

Presentation on theme: "Medium Access Sublayer"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

Medium Access Sublayer

Similar presentations

Presentation on theme: "Medium Access Sublayer"— Presentation transcript:

Similar presentations

About project

Feedback