Escolar Documentos
Profissional Documentos
Cultura Documentos
A Media
Access
Vaduvur
Protocol
Bharghavan and
for
Wireless
LANs
Department
of Electrical University
Engineering of California
Computer
Science
bharghav@cs.berkeley. Alan Demers Palo {demers, Scott Alto Xerox shenker, Shenker
Research Corporation
lixia}@parc.xerox.
Abstract
In has digit for LAN tocols Xerox with Karn off. leads cantly recent years, a wide including will In variety of mobile adequate a new paper Alto access by and in such exchange algorithm. computing and of network generation media being first which Center. uses devices personal y wireless proat start by RTSbackperanalysis a signifi-
lation will
results be more
only
to
particular insight
technology, emerged, these for the portables, require this Palo media refined exchange issues palmtops, Wireless typically cess. the For multiple by
we expect
access
protocols in the
channel
al assistants. devices a single MACA later packet technology. Corporations [9] and Using
Providing
as either
token-based
or multiple
we choose on MACA, first [3]. to guide We of its design followthe note the carthat, secfairly, entereach the incan contention proUsing our call geis
1 Our
channel
wireless
a Multiple posed design, the based ing is rier in in at nealogical on the sense contrast fact, the Third, That learn access Karn packet-level resulting
algorithm key
simulations, MACAW,
proposal we relevant
to a new different
an RTS-CTS-
others
DS-DATA-ACK
message backoff
includes
Ethernets,
dependent; access
Introduction
years, a wide including as stand-alone intended a new to generation variety of mobile While machines, as full of wireless network wireless computing digital first of these the many devices assisnew portables citizens. technolfor
is irrelevant to allocate must protocol rather independently. propagate periods, this be just so that means provides initiated levels access
ond. In recent have tants, were devices and designed are emerged, palmtops, personal prise. device media
learning congestion
about
propagate
portable
explicitly congestion should In particular, not to attempt 5 sections. on PARCs protocol. MACA; data for
network network
cormectivitY
ation
Consequently,
formation for
contention should
devices the to
cent end effectively. bandwidth While device. formance it is merely there vide MACA our presenting our
local enabling
mobile
area
net-
(LANs) be initially,
are expected
to be a crucial
technoldevices
proposed
enhanced these
traditional
media thus media protocols One the Computer
will
office settings where such and most heavily, utilized. network key questions paper, Our a media LANs. In this of the is how
The resource; shared access purpose. use Palo some wireless in in the
in a wireless one is controlled. in wireless goal network Science Center performance access
issues.
This
2 we first
and
on media
background
3, discuss changes
protocol
modifications simulation
different
network
Corporations
Alto
figurations. We discuss remaining design issues and summarize our findings in Section 5.
1 We make expect in future in-depth work to revisit the token-based
in Section
of the
and
media
protocols.
specific
approach
and
a more
comparison.
Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given
that copying is by permission of the Association of Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. SIGCOMM 94 -8/94 London England UK @ 1994 ACM 0-89791 -682 -4/94/0008 ..S3.50
212
2
2.1
The Palo radio
Background
PARCs
Computer Alto t ethnology effects, LAN. which complete and all The built wireless (the All the base
or base besides
station, explicit
about that,
the
presence
of other
devices our B,
communication. in the absence hear this effect our of our access of noise, a station symmetry, of noise. overprotocol testing to a sinmultiple access This where Second, the frequently. or recovery algorithm, (CSMA). argue, approach In small ap First, is the cell This in curthe a it the if a station the may we design; but different these We are gamut and token multiple is carrier properties that the its station interfere will consider that done A can A. The with the
Nano-Cellular
Science [7]; and LAN portable description). base The near-field tow base with that each Given stations range signal r- its
Radio
Network
at Xerox 5 MHz frequency for consists and devices is a single is between Corporations near-field eliminates of base pads, (see 256kbps a pad together at the is region). small defined very small cell 3 to very [8] stafor chanand same 4 merapidly We thus (roughly and interin a a by an a then and
It is important technology station (e.g., our factor noise in noise in sources However, riding to in tolerate There gle channel. access preach multiple especially mobile pect size, would One rently next Karn the these or over B can
Research
Center thus
presence
of noise
simulations
infrastructure computing
we make most
are installed
we design
custom
a noise-free
base station Ethernet). signal ters, (X obtain, 6 meters cell .-3, and
are connected pads transmit decays transmission strength in the far-field a very sharply are
approaches two
we ex-
in diameter)
boundaries:
a nanocell. interference A range receive a receiver tions, station; (w Perhaps given are that typically this 2 meters in is not our but this collision of two signal multi-cell
is negligible, is quite when occurs transmitting from to only either reception cleanly occur This this stations than whose a minimum
aggregate
bandwidth is in the
environment
high. a receiver and reception to cleanly when stacloser is large of x to and distance over be relatively range of the the is unable
token-based
we discuss e for
following is in-
power
10db or more).
surprisingly, the base no gives
requires
a distance ceiling
1.5.
pads floor.
CSMA
CSMA, if the every station (CSMA again). the that at the carrier for in Figure hear station that more 1. However, signals testing station senses schemes Carrier signal is, the it collisions is the receiver and does the not the differ sense strength occur that sender provide Two the hear carrier then as to in presence the at the of are the examples B but A and C). not before the station the to vicinity receiver, two or when transmitdefers transcolnot more not of the
achieve,
detects
carrier
higher in a cell
transmission is tried by
pad-to-base capture
attempts
avoid
environment, design occurs station station, signal The a major capture which of if within one and range model no pad quite any only in and but rather factor
a primary
consideration. a receiver unable of the sharp rare and two if in our to decay is in cleanly slightly is out-of-range interfering environment, design. leads are either a station is exactly our range different for this below. is much known. source for the of a pad way
pads
transmitter; Since
Interference transmitting transmitting closer the makes do not simple or receives transmitter we often umptions other, stations. dios. is still in any tification the in that the It far and stations signal. it interference make model both other
when
constitutes typically
receive presence
because
in signal
strength and to
configuration by symmetry,
rather
in our
C can
A (and,
Ignoring out-of-range
a very
B can
both
in-range one active assof each base rabut model jusFirst, easier However. location to know from a pad,
and base 3 These in future then can ressons work, to we make are merely intuitive guides for design. We hope, explore a vahd the tokencomparison bssed approach between the more fully. Only two approaches.
another, of it.
(xJ
Station each B can other. to terminal hear C attempts exposed to A when assume (perhaps not detect the A to
@J)@)
hear A both hidden scenario A and while C, but A and C reterminal results if scenario transmit A is transmitting B is trans-
accompanied base stations is within for but to poor our not range
additional
no two
of two model
is an extremely so poor realistic. simulations, of the the there the access of is pads.
near-field used
it
B. An
to transmit. to thus B. to When some Cs C is other carrier station ready station), sense A did was to
of some
if
perhaps Station
various no
devices independent
carrier necessary
commences since
transmission;
produces provide
a collision
information baae
We stations. will
There
is no way there
to refer
information
it is leaving
a cell
except
through
Furthermore,
the term stat]on
is no
to both
213
hidden nario. An that when fore out ment essary though sion. Carrier sions can that carrier ance
it.
This terminal to
is the
classic
sceCTS
In would
Figure defer
I, would
C the
from
A rather
A sending
during the
t ermiB, to the exat to the elicit by and This hear stacolhde station
rral scenario, transmit desired Thus, change receiver, from other thus tion with B, it the could Bs CTS If station will collision mission. algorithm
station be free
fer transmission
during behavior. in enables not stations crucially then the receiver collide
transmission. carrier-sense, stations The to role that impending that to a CTS schedule that B.
is exactly RTS-CTS
earlier, did
contrast nearby
stations). it
provide station Bs
or interfere information
whose
reception they
be used range
as an
indication the
about
with
at the be not
information Because
depends
on symmetry; we assume transmission not time and uses this then the hear out binary
if a station
the provide
A does
information,
to seek
assume retrans(BEB)
on MACA,
backoff
MACA
proposed to also the to that MACA traditional similar in Link and general description types (RTS) of the and it this sends would of the the packet receiving defers (which this used brief for use CSMA to the WaveLAN, Access and in packet media protocol and both Protocol does fixed-size to B; this data currently replies contains receiving station time (this as well ). Any of in both station station data with length some Any not radio access as an scheme in Here algorithm many packconwe proA resemble [2]. of the specify signaling RTS packet (which of the alter[9]. [3] the we
retransmission
time.
3 Our able
detads these
Designing
purpose in
MACAW
here is to re-evaluate and then wireless produce LAN. some The design
investigation
proposed
design suit-
choices
MACA in [9],
for
a revised
algorithm,
many defining
description
([9] is itself
MACA ets. When
as p~esented
Appendix MACA. algorithm 30 occur number A this the a CTS of BO via bytes the if defines
gives
the
of short, packet
pseudo-code We mention the for control transmission station after domly tween The ter functions, an RTS, does the end
A wishes
to transmit
to station transmission. deferring with the the after includes as the station the the length CTS, the
B, it sends If station
a Request-to-Send t ains the length B hears explain (CTS) posed defers CTS time a CTS packets). With defer ceive avoid the every may long the CTS the RTS, below), packet; data all
retransmissions.
in response period.
to its
immediately packet at
overhearing the
an RTS
where and
backoff
counter.
transmission
turnaround overhearing expected RTS an station the assures and RTS can CTS data CTS will rewill Since us that
F,m..
is received function
counter
is adjusted
transmission
hearing hearing
transmitting
Fd.c(z) F tnc: BO := F,. C(BO). For BEB, F,nc(s) = MIN[2z, BO maz ] , w h ere BOmin resent and ate other lator ing base viding strength from to the ducing the lower and upper bounds for the respectively. We our cation For our simulations simulations The simulator we have of the we have
BO~,~
and
chosen protocol
BOm,n = 2
to evaluof
BOma. = 64.
use packet-level design of the decisions. network simulator we use is a modifiused in a number
of colliding all
is in range transmissions an RTS but sender mission, they the data but are not
is possible, in-range
the area). Notice that stations that hear a CTS because they are in range of the of the after receiver the CTS they can has commence been sent; collide transsince with harm,
studies (for example, [5]) of wired networks. The simuis event-driven and contains the following components: generator The space (which simulator into source size. small at each to the can For be the can generate TCP, and approximates cubes cube center made then according of the arbitrarily data UDP, the to cube. streams media the small accordand by dithe due by redistance to various stations. the the statistical models), 1P, pads, computing Errors
of range
a traffic
without
in range
of the
receiver
cannot
transmission.
the reception of the CTS; of the RTS this includes
at
4 This turnaround time is the time from the receiving antenna to the transmission system delays as well as radio
operating
transients.
simulations
mentioned
in this
214
base station
/? o
Figure in range base 2: station
EIEilia
Table 1: The throughput, in Figure 2. in packets per second, achieved
PI
celI arrows pads second configuration and both indicate are and each pads the
o
other
P2
all stations data of the data for are to the data
by the
streams
where direction U DP
of each
are sending
UDP
traffic
the
channel. evenand the counter behavior if then thereby keeps pads beis no will
1, when transmits
transmission). of 64 packets
generating
at a rate transport.
a single
are using
BOmac ).
in [11]. but every will its
is very
the our
cubes
are A
in size. are can of cubes. 6 feet be below either signal At station is greater and least refer to to 10 dB the the the a pad Whenever is added end than during set on 512 of here, 32 or bytes, Simulawith use a null protocol: exthe are to that to one of can correctly some than the packthe 64 and base or a a to the
pads
backoff that
simulations, station resides sending, signal the packet (the of the the from For
likely
a cube.
phenomenon there
the
of the
current the
nearby signal
decreasingly backoff, capture one other the pad backoff in the its and this views own there leads of the this,
transmission, receive threshold the entire We ets going station. devices packets the tions sum
receiving strength at 10 feet) by to sender simulations a constant data CTS, etc.) 500 The the basic our the total total turn and packets at times.
one pad backoff one ambient all pads. on its congestion having cell. algorithm the a station
dynamic
is driven counter would exlevel Each own inwidely by curhears counter. are from 1. Thus, in all this The havby of al-
signal
is greater
lower
pads,
packet use
transmission term a particular most second. packets run of the data at All (RTS,
of congestion pad is doing experience formation; varying including rent Thus, range pads value in of have a packet, To rectify
which
receiver
based
generate per
rate
of congestion modified a field into counter. value scenario each right counter. the
control
are typically period time. investigate algorithm protocol We should access also always the provide fairness6 easiest
between
and
seconds,
in the packet of the it our each the are copies single other, same shown
header
a warmup turnaround We the change. media and not often media user deliver backoff
of 50 seconds. two clearly fair over way areas and should access and optimal to optimize sharing the state to of
Whenever
deliver when
algorithm throughput ing the the media Above gorithm adjustment the tion to
the is now
completely
compatible,
they
throughput
produced basic
a fairer
in shared
resources. we modified allocate to the of the structure fairly. The it it An can BEB both also upon of the slightly backoff backs off bandwidth backoff protocol. rapidly; additional minor improve crdculaquickly backtransbackoff suca
is to eliminate exclusively).
computation
3.1
Recall
Backoff
that the to the level For in a cell,
Algorithm
MACA backoff minimal We now of fairness as depicted other (and
are value a precise
efficiency adjusts
extremely
uses
exponential after after that the every every this simple case does where
is doubled
when a collision is detected and OR counter to BOmin immediately mission. counter; cessful have period relevant for the This produces rather one-cell we return backoff there is high. such wild oscillations, are counter several to increase large in our simple transmission a minimal when media of contention
reduces the a successful in the after where This and we must a cell every
exchange.
in some in the
example,
consider
there
Figure base
on the
2; the station)
properties little effect
within pad
of each
5 These
however of the 6 We
based should
we have factor
=
have
simulation do not
algorithm; by 1:
upon
Fine(z)
a collision, JflN[l.5z,
definition
of fairness,
this
by a multiplicative
and
= MAX[Z 1, BcL+I. This rnultb~cative inlinear decrease (MILD) still provides reasonably
215
base station
base station
(a
3: station. per
cell
where data
all
stations data
are to the of 32 Figure in two base range 4: of the station. per A single cell and and configuration The the base third pad UDP where station all stations data to data at a rate are to the of 32 of each pads, Each second other. is sending data
other. and
stream
at a rate
stream
is generating using
packets quick by not ing to escalation resetting repeat the in the the backoffs in when to the data two for BEB identical that with the backoff contention is high it avoids after in base but havevery Figure station. we can tion stations packets extent streams of to nications are likely This and BEB copy PI-B P2-B P3-B P4-B P5-B P6-B Table by the 2: The throughput, in F@re 3. MILD copy each then backoff We in pads data of MILD was essentially low interfere a clear the all counter tested
BOm,n
counters relative
escalation
state
that
in that to
cell
performance
we want
to treat
all streams
(as opposed flow That This all wireless thus can base
recall we want to the since source per the and treat contention each station,
from and
advantage
themselves.
on the
two-pad the
is especially
because
go through of
resetting
counters
streams. be implemented for send each (e.g., stream for after for streams be done the stacounter, destinatime. winner, streams a slight algothe If wait N independently to data pending can queues
performance.
notion running
by keeping, queue.
a station
a deferral destinations, when tion how tion. more second, achieved time, rather
3.00
3.05
in
as N co-located This data on the an the have one RTS shortest the same of them may pending, backoff to this wait as the of this, have stream station.
as follows.
to determine,
wait picks
then one
of these
shortest
packets
randomly
streams
originating advantage As 3.2 Let Our the head equally tion is sending is sending equally to can rithm of fair allocation packet BO of the allocates Consider pads; he pads, of the and of bandwidth. queue which packet the the the the at each cent rols at the
multiple bandwidth
Multiple
us return initiaJ
Stream
to the
Model
notion backoff This there to two base gives and in
produces
among
competing design had (and queue. transmitting 4 where packets to the station a single FIFO parameter design stations. are three oft half station.
streams.
station, of
Multiple
transmission
retransmissions)
in Figure
throughput, in Figure 4.
in
packets
per
second,
achieved
packets
Allocating bandwidth
streams
pad-to-base-station
bcme-station-to-pad
stream
streams,
a quarter this to
The data
to each of the two in Table 3 shows that each transmitshare and an equal 3.3 In the this role
a single (pad
queue
Basic
Message
Exchange
the four basic RTS-CTS-DATA The and the CA material need for meson a
of the
section of link
of the to
sage
exchange
changes.
achieve; of
functionaht
y (as in CSMA/
defining setting
a general
definition at the
accepted
community;
is beyond
moment.
However,
completeness.
216
3.3.1
Many electronic transport UDP vide
ACK
of the layer was reliability. or by wait, are the timeout both timeout of the that upon applications require these used in In transport period local and at the Thus used reliable applications the MACA, by layer, TCP of 0.5sec, long can haul link-layer we have an from then data the the on mobile use TCP when noise, This data the devices, of data. such At to has to as the to prosuffer be a sighave chosen to is the
can
mail,
delivery
which that
previous
simulations)
5: other. each
two The
both also
pads in
are in of
respective
range
corrupted current
to their a rate
is generating
as many
implementations
second
transmissions. be much to fit the the faster short basic only station only that matters. hear to hear its However, a returning any replies when Cs CTS. and transmission When thus B is initibenefit if C can is useless. get any simple
In contrast, because time packet, is not uled the the its the there not We a sirde v. modeled less of nation. original recovery when head
g~o
recovery media.
the
periods exchange is returned completion sender, If the the instead sending the The
be tailored
scales ACK,
amended
transmitting, ating and rapidly. relevant directional at both We defer because C cannot so does There carrier-sense defer carrier ensure ACK). chose carrier a station Every sion RTS-CTS transmissions (the that This sensing sends station is about a transfer does not With
C is unable
to include
acknowledgement to the sender If the indeed then ACK been when returns increases arrives when if does on is is sched-
7 Moreover,
C does
response, uni-directional
backoff
counter with
transmissions however, exchange, that data. not the was CTS, data. One it the
by the but
correctly
ACK
packet is sent,
retransmission
associated times
of a CTS. an RTS,
B is transmitting the RTS exchange transmitting to this useless slot slot do not CSMA/CA which sending Data-Sending this packet and that
no CTS counter
before
sender backoff
decreases exchange
RTS-CTS
a success can
is a successful simulated
RTS-CTS the
but
RTS s. A station detects air [2], not packet know that a data stations slot has of clear protocol does a DATA will
effects stream.
transmission
time
Dad-to-base-station as a given size) Table is not probability received the DATA noise layer. TCP inclusion vs. 40.41 1 packet results. and that
cleanly resulting exchange, level The is much of the in the Given it ACK no
a slightly
in throughput ACK
is included
exchange until
transmisdefer passed.
to the PPS
is only algorithms
overhearing packet
the likelv
in 1000 can
RTS-CTS-DATA-ACK
46.72
RTS-CTS-DS-DATA-ACK
23.35 22.63 in packets per second, achieved
throughput, transmissions.
we have
augmented
RTS-CTS-DATA-ACK
o
The throughput, in Figure 5.
streams
Table
E!miima
4: The TCP presence throughput, data in packets between per second, and achieved by stream a pad a base station of noise.
Error
Rate
RTS-CTS-DATA
RTS-CTS-DATA-ACK
have
examined two-cell
the
of Figure
this 5.
in each
simple
configuration terminal the the period key after is that the it identify
pad-to-base-station with one the next and to then the without the DS pads to fupacket the effecharm after the it is the the
stream. DS
throughput DS
a single in the
proceeds
without packet
contention before
therefore
note sender have after will that does included no not collide longer
compete
nOt
DS
original 1), analysis we argued of the exposed that the even of the terminal configuration C should of the receiyer the
does
ACK
If
terminal A, and
assumption
to transmit B, it is out
because of range
it is in range
transmitting transmission
hearing
ACK
from
217
6: other.
two The
cell
both also
are
in of
7: other.
two Base
cell
pads in Pl,
are
in of
of their
in range at
of their
their a
B1 is sending of 64 packets
to pad
and is using
respective
pads,
rat e of 64 packets
transport.
generating
for to most
large
comA secondary creasing We the ceives it then oft ral can problem it on solve both is that never of it the behalf B1s backoff a response problems B1. respond contention packet several to the first RTSs responds RRTS, to by (due period during with the Whenever counter from having a station to deferral), and the RTS). normal sends deferThe If mest o t he sender keeps P 1. P1 do reinbecause receives these of next cannot
an ongoing
is sending is essenend up data results pad and and must this fin-
picking
usually
transmitting transmission in a collision. its This the following RTS send ish. in this informs of the add
middle
contention transmissions
o-Send
(RRTS) received
requires
(if it has received only responds in response times, long that a fair
immediately to the
the
is commenced. occurs.
Stations enough The results access the 7 shows from does the
another
provides
synchronizing
information
second to the
3.3.3 Consider where media. resulting porating B 1-PI B2-P2 (in that this riod. when receiving deferring can to arrive fact, the
RRTS
the each The from all stream stream it two-cell two first the the column version amendments is receiving section, same are all this confimration data in ~treams Table of the completely of its deDicted alo~e can in fully protocol so far. while throughput. configuration Figure 5 except streams packet of an only RTS of the RTS, it way (in pesize time the is B1 the Figure l~ad 6. the
however,
two-cell access
configuration protocol so far. the This incorporatThe P2-B2 is because by sending The its in is when gaps
of the
throughput
6 shows
access
media
we have
is almost
denied
As in the
previous is the flows B2-P2 this, to initiates pad to the during data the
hear
it
due
configuration reversed) wins relatively transfer respond a transfer very and short the by with to
a transfer
one of the initizd large data most a CTS P2. in The its
case After B1
stream) to the of
contention
due that
of synchronization contention
as compared
control
packets,
has The
sending
is irrelevant have
P 1 cannot
because
the
transmission
successfully data
between
pleted CTS.
transmission
m
Table by the 7: The throughput, in Figure 7. in packets per second, achieved streams
Table
by the
6:
The
throughput,
in Figure 6.
in
packets
per
second,
achieved
streams
3.3.4
So far there
Multicast we have is a unique RTS-CTS others collisions only there cannot CTS. by discussed can be unicast packet. viable are likely we have multiple and being transmissions, For multicast for since the receivers where data multiwith such use an a packet,
for each
key
is again but
the
of synchronizing B2 during way that of the of the very knowing DS packet base
B 1 is trying periods
coordinate having
to collide avoided
Notice other
solve hear
this any
problem
because
a multicast
transmission
streams
message
exchange.
218
. _
Figure cell two Cl counters 8: are A in leads two to cell
cl
L_._
@l
all
I
__ I
~/~\. PI
(offline)
P2
P3
where The
the of
pads
in the Figure range The data sending 9: A single base to the celJ configuration stations is sending base of 64 packets P1 is turned and data station. per off also in to each Each second after 300 where range pad, stream and all and using pads each are in other. pad is for
range loaded
copying values
leakage cells.
between
differently
base station
is generating UDP
RTS hearing
followed stations and DATA design, stations that not how work
by
the that
The length
a multicaat
transport.
seconds.
address, following This those those will signal out CTS rounds
transmission. flaws sender but in the control without as CSMA. will not unicast yet defer, the the
no overhearing, backoff in C2. one copied value counter all value when the counter, However, another the two by leading from leading algorithm from different the P5 in
the
and and
the
much
are within
leakage backoff by now idle in Cl B2 have traverse have collisions. number problems. to another, accurately our
is delivered
by the receiver
CTS
We have
messages involving
finally Similarly,
case of multicast
transmissions
of contention.
time.
also
transmissions to wasteful creates one region counter in a cell. levels we use a single
an artificially
Backoff
to the
Algorithm
random MACAW of RTSS the they on the therefore earlier, The be the occur. nature
Revisited
of multiple a backoff are delayed The the access, algorithm, for a random and of this value collisions in are which
backoff
unavoidable. transmission of slots, collisions delay value media. As stated a high put crucial algorithm rat ely scribed to MILD level. tions we that single ways tention others. poorly tion and with near ally The other, C2. C2. the For also Notice how overall role reflect in to to streams.
uses
to reduce once
probability backoff reflect the goal and backoff both the the counter;
regions assured bient far There a CTS reflects a RTS close or the CTS there off the high the they ure (or
of congestion;
we are reflects
no longer
backoff level
is based should
of the backoff
is a second implicitly there fail to (so will sender is trying has In These made left
problem assumed However, return the never will any the cases, even
to evoke an
both plays
has been
a collision, a CTS.
of througha backoff accuBEB sta3.1, ensures level wireless is heavy can the contention perform cells, the ( P5) Cl is to configuraborder by a is alLAN, conin 3. I we de-
congestion.
source RTSthough if to any a for region; in Figbase the P1. a cerbefore As we turned to
the sender
is corrupted), backoff
we modified
attempt,
necessarily
to reach but
stations value
RTS both
models
though
if congestion There
phenomena the ambient by the the and pad trying pad the base pads
a single
homogeneous. is for The in such depicted Cl C2 border to their near the the
number
congestion
configuration
single
9. There
is sending
P 1 is turned to communicate
to the
MACAW
transmit pads
stations are
number 3.2,
generating and
within
of each If there
interval
so they
packets.
of the
transmission.
219
base the
station
counter, Every increased streams out faster driven problem when The this high the the a high
an is 1.5. the
finally the
successful). appropriate
Given ends
that
adjust usual
backofi all
adjustment To nicate backoff except tion, each with value. now and packet
algorithms. fairness, same This there header. results the enables station to it. overall pad. the yet is stations station by our backoff backoff column commusame scheme, each ends stainto the by alare the receiving a separate the this This also The with right use the
achieve
carries
transmisBecause the values. high value comis also staand but always eventuoverall 8.
counter increases
is reduced to very
is achieved insert
counter
we now
Because transmission of the backoff municates by the copied tion. the the ally and
is reasonable
of Table
copying
throughput congestion
with used
problem backoff to
is exacerbated base
unresponsive a given
copying Successful
algorithm,
specific
copied
value the
3.5
Preliminary
Evaluation
a detailed this
of MACAW
description to quantify in a cell In under and decrease a data which see by and noise. below, superior under that the rate MACAW is 78~0 this Note station. protocol of the the with Table the MACAW overhead a single 9, we original addi-
dominate thoughput
additive backoffs
is low,
of Table
opportunity to a base
MACAW a pad
stream the
packets-per-second RTS-CTS-DATA
RTS-CTS-DS-DATA-ACK the DS 870. which rate more presence However, than and ACK roughly MACA
217kbps, Table by the 8: The throughput, in Figure 9. in packets per second, achieved a data pacity. often the All three examples or receiver, and Figure strate to send separate earlier, requires stream. reflect sender. rately, but to each to then be used end that to the us The the These so that should in requires, not we must different backoff to (Figure 8, noise next to the sender 9) dLscussed in this section demonbetween station stream. of in the can when the at each the should As both message backoffs maintain we discussed exchange ends and estimated by the algorithm whether CTS sender we know we do In our not a CTS (this not returning the but of at the the The decisions results at when being and that not not know reat if at any In can is
BEB suffer has, the as Wo
is 84~0 channel as we
capacity.
of roughly
streams
compensated
performance
of congestion
differentiate pads. Each for for value at nature used both levels
used a Table
MACA
I RTS-CTS-DATA
] 53.07
MACAW
9: The
I RTS-CTS-DS-DATA-ACK
throughput, single in stream. packets per
I 49.07
second, achieved
hi-directional
congestion
by a uncontested
should sepascenarios were from 10, four There C2 near other, they (P1-P5) of its The second, MACAW We remark using of to over handle data two The which pads only and the and can one C3 we used to simple. more scenario all (P5), (P6) is in border overhear their data rate the own motivate In has similar near which which range the this three to is near straddles of both are others station. to and data for each base B2 our various network cells Figure the the and within as shown 8. with Cl C2. design conin conC2 with beThe of howof the base to
should
information
coming,
backoff
section
we present
be combined t ransmission.9 in the being there backoff fails, received is congestion RTS at any at the fails; at in is not the change sender
complicated
congestion
figurations. Figure tains contains Cl. tween pads each ever, pads station B3. per the First, ment ity
is somewhat
an RTS-CTS the RTS ceived. we know the must there we will the change Appendix determine nation
8 In algorithm
exchange
border
is received
is one
receiver.
(P1-P5)
streams in each
sender
P6 is sending
a UDP stream
receiver an CTS
generation
algorithm
Table the
10 shows
performance
if the only
sense,
determi-
we compare
be definitive
this faster is
algorithm. on MACAW 37~0 in congestion a few over simulation yielded a superior compensated the streams an improve-
a problem than It
decreases
increases,
the problem we describe here. Of course, the we have observed, problems of its own 9 we combine the congestion information backoff values.
throughput. overhead.
MACAWs a fairer
increased division
MACAW
has yielded
of throughput
220
, _
I
____
,__
_____
in and
its the
all P7
other can
in
their B 1 other.
P 1 and
MACA.
of throughput.
P7-B4
respectively
35~o
of the
through-
I I I
I C3 .
Figure levels 10: A configuration with
I I
32% and
mobility
with MACAW they receive only In this simulation, the impact of in either case.
MACA
I MACAW 2.39
.
three
.
cells
_
with varying
P1-B1
0.78
of congestion.
MACA
MACAW 3.45 3.84 3.27 3.80 3.83 3.72 3.72 3.59 7.82 7.80 25.16 per second, achieved In This Section data 3.3, 4 Table by the 11: The
TmnP2-B1 P3-B1 P4-B1 B1-P1 B1-P2 B1-P3 B1-P4 P5-B2 B2-P5 P6-B3 Table bythe 10: The throughput, Figure
9.61 2.45 3.70 0.46 0.12 0.01 0.20 0.66 2.24 3.21 28.40 in packets 10.
Eki_E!J
throughput, in Figure in packets per second, achieved 11.
streams
Future
Design
Issues
simulation the basic of including an ACK only way by packet to achieve the ACKS packets number the would on the had request did it We not have the queue the in it, could (by for of the sending wait CTS. results RTS-CTS the which sup-
ported
adding shows ACK, exchange unless a bit onto which arrived had not for the in the
exchange. functionalbasic For in(through be piggyincluding most a stream station for the the a reat this. in every
isty of the message stance, setting backed field same 0.59 able can cell. packets to cope In for per with due MACA across higher data media second There Cl) and Cl the flanked a coffee (PI is due open error into the seconds - P4). to the area. rate into to and is 9.60 MACAW, any two second, packets highly the the maximum in in the MACA, Third, from losing difference same the cell between is only is and much of the Cl is about a area and tions the in Cl, by is there maximum MACAW too effect even C3 cells though where achieve cently a station would queue then piggy-backed throughput difference shield streams while per streamsin
is not the
specifically packet the than the at station one could sending subsequent
sender
Whenever
second.
When
congestion,
packet, of ACKS. as exback of here to exstaexin either note ways fully that data
uncontested
throughput In both congestion has only the small PARC. (cell C3) cell in a much one same The
neighbour.
a receiving of the
cells
utilization. as depicted cells, (C4). which Each in Figure represent of the office source electronic effect from period of 3.3.1). (which 11, simulates at Xerox open (C2 cells in the the pads cell noise Pad P7 an Computer the in C2, There We of coffee presence 0.01 room the offices Science of two Laboratory researchers There are four
functionality not
of link-
portion
implement Similarly, changes. alently collisions. sensing of such The we have, ply the only
C2 and
achieved exchange
this
is also simulate
pattern. many We
equivas
whiteboard
which have
intermediate
Section
clean
space
an uncontested is 2000
carrier
sense
mechanisms. Figure 7 poses a problem to which This requires some distribution but since it none seems of the difficult stations to in supthe
configuration
long).
station per of
Each
its
data
pad
stream
rate hearing
to the
of 32 the
as yet,
information
no answer.
to
cell,
generation each
of synchronizing
information, B1
packets
second.
221
@J@J
_,
P
I
\
o 0
B4
l__
__l
o
_
PI
o
BI
I f J
..---
C2 I
at PARC.
Figure
11:
A configuration
based
onpartof
the
Computer
Science
Laboratory
congested Another with our The provement However, complex, scratched our options Unlike protocols tions our want in violate to
are aware problem RTS-DATA algorithm the initial performance the and design in surface both
that
the The
demand media
for in
backoff, algorithms
we have
discussed
is exceedingly We have the other any behavior algorithmic media Such access variato not more only are For service considered approaches Various or work. for and evaluating and fair locfllzed connonwell of two that neisuch who such Bewhat and is not altokenreservations,
is derived
MACA
[9]. Our
is at the
in understanding
understanding not
First,
receiver,
realization, suggests
Karn
[9] and
use of
others approaches an
[4,
RTS-
[9], we did
involved symmetry and so far we hope networks best-effort from to
12], argues
(CSMA), CTS-DATA reasons, ACK change. Second,
power. psss
exchange. then an
of performance ex-
generalized
an RTS-CTS-DATA-
RTS-CTS-DS-DATA-ACK a homogeneous varies according We It is inadequate parameter. for each was taken collisions. levels must should rely often notion packets. networks should The phenomenon. to the instead and
carefully. moving their We not from wireless service have not a single also instance, as well such other based service, tion
Rather, congestion duced end for each Third, lective direct to the could ment.
the
of congestion
locaintrothen which
of the
moving
[3] supports
separate of the
parameters Care
as asynchronous than
service
stream learning
was experiencing congestion each views from not only shared access station
approach.
schemes,
those
are possibilities Furthermore, the performance of that such However, For and up can both location set tings cept. defined, cells, ties ther a pad are
we hope we stated was as an in setting inst ante, therefore base can the range stations receive same of one bandwidth.
congestion, of the
chance
our
highly backoff
a homogeneous
utilization
overheard
Ethernet, the
is a well-defined
be relevant
geographically of a wireless a pad hear other allocation of the that both when
homogeneous
border (in
information
contention
of providing be initiated
information.
observation
of data. receivers
congestion
is mainly
location,
sender
less, because they cause on a final design choice, policy we want to implement.
cannot contend data transmissions so that it congestion being These performance design is still are many the is in the
know when the the RRTS packet bandwidth packet also data when allows was not the our
of congestion.
5
The bile will
Summary
emergence computing play in recent years devices suggests important particularly of a new generation that indoor wireless role in our in traditional of moLANs set-
when
communicated. changes the which media remain have access As significantly protocol. improved However, in Section
an increasingly infrastructure,
telecommuoffice
preliminary.
we discuss unresolved.
4, there
nication
issues
222
References
[1] D. less Allen,
1. When data
A packet
is
in to
state sets
and a random
wants timer
to
a to
LANs,
Wire-
the
paper 2. When
from state machine of bytes For state timer, state state time and IDs to Data and receives a RTS (CTS) A, sets B packet packet and state. a CTS Data packet packet the a timer transmits number a Clear-To-Send send.
802.11/93-xx.
[2]
G. Sidhu,
AppleTalk,
R.
Andrews,
and Wesley,
A.
Oppenheimer,
Inside
Addison-
1989.
[3]
K.
A Hybrid Wireless MAC Protocol Supporting Delivery SerAsynchronous and Synchronous MSDU vices, IEEE 802.11 Working Group paper 802.11/9192, September, 1992.
Biba,
to Wait
3. When
from to
B, and
[4]
IEEE
802.11
4. When
from
a Data state
packet
and
resets when it
the
to IDLE.
[5]
Clark, S. Shenker, and L. Zhang, Supporting Rea2in an Integrated Services Packet Time Applications Network: Architecture and Mechanism, Proceedings of ACM SIGCOMM 92, August, 1992. in a Datagram S. Deering, Multicast Routing work, Tech. Report No. STAN-CS-92-1415, University, December, 1991.
5. If A receives
state, the The it sender Deferral clears
is in state.
CONTEND packet to
transmits WFData
a CTS
[6]
InternetStanford
state
QUIET
state,
A to hear from As
[7]
A. Demers, S. Elrod, Chris Kantarjiev, and E. Rlchley, A Nano-Cellular Local Area Network Using Near-Field Virginia Tech Symposium on Wheless RF Coupling, Personal Communications, to appear. C. Kantarjiev, A. Demers, R. Frederick, and R. Krivacic, Experiences with X in a Wireless Environment, Proceedings of the USENIX Mobile & LocationIndependent Computing Symposium, 1993. P. Karn MA CA - A New Channel Access Method for Packe Radio, ARRL/CRRL Amateur Radio 9th Computer Networking Conference, September 22, 1990. K. S. Natarajan, Access Control 802.11 Working 1992. C. C. Huang, and D. F. Bantz, Media Protocols for Wireless LANs, IEEE Group paper 802.11/92-39, March,
current
state,
B to are the
Timeout
rules
following. state to Send of A and A then and the timer packet the and expires, which goes requested
[8]
A is in CONTEND a Request IDs to send. the station of bytes state. other IDLE state, state. of rules. when
transmits
a timer
[9]
a timer
expires,
a station
[10]
order Timer
precedence
is
Deferral
rules,
MACAW Message
in A MACA, rules, pad
[11]
S. Shenker Some Conjectures on the Behavior of Transmission Control of RanAcknowledgment-Based dom Access Communication Channels, Proceedings of ACM Sigmetrics 87, 1987.
Limitations C. Rypinski, oian Applications, IEEE per 802.11/91-46a. of
B.1
As
Exchange
MACAW Deferral Wait Wait Wait rules is in to
and Control
can also and can For For CTS be
Rules
described rules. one of eight Wait and for Wait QUIET. states: ConFor in terms of
rules,
running
MACAW
[12]
CSMA
802.11
(WFDS),
For
ACK
The A The
Control A packet
MACA
Control and rules data and Let that the Backoff rules The B. Let station stations only A, govern A wants C and and the rules carrier govern to transD be two only
1. When
data the
state sets
wants timer
to
a to
Deferral
2. When
packet packet.
IDLE a timer
state and
and
receives to
a data
A, then (WFDS)
it
a Clear-To-Send goes
D hears
DataSend
A pad running MACA can be in one of five states: IDLE, CONTEND, Wait For CTS (WFCTS), Wait For Data (WFData) and QUIET. The Control rules that goven the state transition are the following.
3. When
from DS WFACK
B, it clears
followed state by
packets a timer.
223
B.2
Each
Backoff
station
and Copying
keeps the
Rules
variables:
following
When
from then
transmits
an ACK
1. my.backoi%
value
at this
station.
6.
WFACK resets
from
B, it
state
to IDLE,
local.backoff estimated
station
as
timer. B is in packet again it IDLE state of CTS. packet CTS and state state. state goes and packet, to the receives it WFCTS a Request state to and sets and when to it is in CONTEND goes to and receives last time, a RTS it sends for a
7.
acknowledged
the
8.
(ESN) exchanges
transmits
packet
the value.
sender,
: the number
of retransmissions.
a timer
receives
9.
an RTS,
it goes
to
When a pad P hears a packet, other than an RTS, from Q to R, P updates its estimate about Q and Rs contention levels by copying the local-backoff and remote.backoff In addition, P values carried in the packet, respectively.
also that copies the Qs backoff station level because precisely, remote-backoff, ignore value aa its around they own Qs the may backoff, backoff carry neighborhood. not the assuming supposedly RTS correct Q is a nearby congestion More backoff, == backoff backof f RTS, are ignored values. (local. packet Qs therefore
10.
is in IDLE Send value. rules are the a RTS to the for the sender,
(RRTS)
transmits
a RTS
a timer
The
Deferral
following. from Bs from the A to B, it goes from and sets a timer CTS. A to B, it state, Data and packet goes and from then packet If else = local_backoff; ! = I. DONT_KNOH) = remote-backoff; if (remote-backoff Rs my-backoff When a Pad
packet QUIET
ESN)
current
state,
A to hear
2. When
its sufficient hear
current Bs
QUIET
sets
a timer
= local-backoff; P receives eqnumber a new In should values the either be the own retry-count. when of its transmission case, and a packet haa increased, or the that Here with correct from Pad either Q to the P, if
3. When
its value
packet QUIET
from As
B to A, it goes from and sets a timer data. from D, and it goes from
exchange= completed. the packet and resets used value other the backoff
current
state,
is initiating
handshake P updates
B to hear
values
4. When
its value The 1.
packet QUIET
increases
current
state
sets
a timer
exchange.
temporarily
attempting
Timeout
rules
is synchronized is done. if the value at which value value f, packet for hand,
handshake P assumes backoff from own backoff backoff its Qs values backoff
is in
state goes
and
the
timer CON-
a random
to the
2. When
expires, form packet sion state For
is in either
state a RTS
the to
timer per-
end
has occurred,
as the
or a RRTS transmis-
(local-backof
If
sender-initiated containing number sets D and to IDLE any to the other IDLE
! = I. DONT-KNOW) = remote_
receiver-initiated
transmission,
a RRTS
state. state, state when and a timer resets expires, the timer a station value. .
3. From
goes
Ps
.1..
Qs if
/*
with is
*I _count * ALPHA;
backoff
= local_backoff
(remote-backoff
224
Ps else
local-backoff
= (local-backoff = rny.backoff;
+ renrote.backoff)
- Qs
backoff;
to Q, it assigns
the
parameter and
in the
rernotelrackoff,
following
If
should it be at in communicating
ESN;
Qs backoff += retry-count * ALPHA; If reached max-retry_count, Ps local-backoff used vith Q = MAX-BACKOFF; Qs backoff = I-DONT-KNOU;
225