Randomized
Mutual Eyal
Exclusion
Kushilevitzt
Algorithms O.
Michael
Abstract
with
[4] a randomized
clusion
with
bounded
logarithmic Saias
algorithm
sized
and
versary can
observe
the
interval
and
section.
It
values
of
value ponent
behavior
next
draw
local
shared
so as to discriminate
process.
This
ties
of the
In the
above
pa-
of the
invalidates
1
modified,
using
come
this
same
results.
yields
simple
the
the
In
criti-
this
critical
as well
as the
number
com-
arrange
a chosen
claimed
proper-
ideas
difficulty,
as in
can
and
algorithms
for
the
direct
to
copy
provided
that
commercial
title
of the
that
copying
and/or PoDC e 1992
specific
fee
the copies
advantage,
publication
Machinery.
without
lem
mutual-exclusion
ACM
and its date
is by permission To copy
the
all or part
otherwise,
of this
are not made
material
and notice
of the Association
question:
the
shared
mented? ret ical This
a fee
by use of
action, to
and
a Pi sched-
was suggested in many
for
can
the
this
guarantees
by
papers
[1, 4] and
A solution that
testing
lit-
prob-
freedom
be achieved
semaphore)
and
by
freedom
al.
[1]
What variable
considered should
be
v so that
This
is because atomic
sume
is that
ately
writing
question
is not
of practical
in practice
operation reading
and the
the
size
can
but
also
followof
(deadlock-free,
mutual-exclusion
interest
the
be imple-
only
of
test-and-set what variable
theo-
interest. is not
we really and
as-
immedi-
permission.
‘92-81921B.C. ACM
et.
ing
an
resource.
v (i.e.,
atomic
alone
to one
lockout.
lockout-free)
is given
requires
(this
need
exactly
activities
example,
algorithm
time
shared
problem
use of a one-bit
Burns
is
to
available
there).
. . . . PN be
in which
was discussed
(see, for
cited
time
is an
This
deadiock
from
for Computing
or to republish,
well-known
PI,
variable
v itself
[2] and
is an
the
or distributed for notice and the
copyright
appear,
the Let
their
v is always
then
erature
randomization
*Research supported by research contracts ONRNOO01491-J-1981 and NSF-CCR-90-07677. tAiken Computation Lab., Harvard University and Computer Science Dept., Technion. e-mail:
[email protected] .edu . ~Aiken Computation Lab., Harvard University and Institute of Mathematics, Hebrew University of Jerusalem. e-mail:
[email protected]. edu . Permission
the
determinis-
some
coordinate
to
Dijkstra since
from
employ
to do so).
from
granted
than
[1] for
with
section
test-and-set
setting
uled
deal
that
to
a shared
[4] is
essentially [4],
in
problem:
a critical
allowed
They
of {4], so as to over-
obtaining
Thus,
in
we
processes
is
the
against
paper
rnutua?-ezclusion
N
algorithm
established
size
Introduction
access
paper
smaller
in
algorithm. present
a shared
algorithms.
execute
and
the
of considerably
about
round
schedule
ad-
conclusions
variable,
tic
employing
given. the
of the
variables
waiting,
lower-bound
a
of processes
closing
randomized
of the
that
an opening
then
their
of the
was
in the
the
can
variable,
postulated the
ex-
employing
out
between
cal section
mutual
[5] pointed
scheduler
per
waiting,
shared
Lynch
for
Rabin*
bounded
variable In
Revisited*
0-89791
-496 -119210008
/0275
. ..$1 .50
275
it
can
be done
very
fast
so that
no other
operations
an assumption variables.
Burns
tion
shared
this
Rabin
et.
[4],
presented
for the problem
shared
variable.
the following
[1]
proved
is required,
a
an O(log
algorithm
on
lemma:
any
1
say Pl, ...7 with