Optimal
Response
Time
Retrieval (Extended
Ling
Tony
Lawrence
E-mail:
Doron
Laboratory
Lawrence
deals with
for retrieving
with
replicated
to this
The paper
(a)
access path
problem.
worst
load, (b)
case response
schemes, in this this
where
present
and also discuss
the replication
bound.
replication
We then
scheme
case response
a network
(or
all)
flow
reside
bound
under
for
scheme that
can also greatly
reduce
for
write
transactions.
of these
disadvantages
partial
replication
cated
on a small factors,
path
for
the
identity
the average
time.
Introduction
Recently
there
improve or
disk
all
of
Such
the
many that
or
addition
applications is that
In
general,
choose with
1/0 that
may
to read utilized minimum
location
[BG88,
lead
be used
in
performance
case of disk
may
the
be The
page replicated from one
seek time
on r disks,
one of the
which based
can
r disks
service
on the
current
the
for
disks
we may
list
request arm
LM92].
[CR93].
36
this
The
same
tiles
time
for servicing that
must
be
by
the
we
choose
disks
others and
on
of an
problem in
to as
are
less
disk
request
The
is based
that from
on the
given
tiles
resolution
the
to read
replication
In tiles
with We
on-the-fly
disks
across
above.
requested
determines which
to allow
improve
explained
space.
and
response “hiccups”.
of the
low
less of a
disks
be minimal
greatly
replicate
they
occupy
algorithm
the The
some
reasons
or
consists
without
could
such
more
from
terrain
aerial
m parallel
resolutions
request
must
of
of
with
fetched
we
the
visualization
by a terrain
a set
viewed
replicating
for
decision
basic
workstation. of the
time
an on-line
be
each request
response than
disks
a transaction
are of different
visualization
of
more
the
choice
a collection
stored
can
display
frequency,
of
computed
on
the
is motivated
where
tiles
a subset case
balance nature
be
the for
A visualization
that
the
found
that
load
affect
path
work
are
area
as needed. of
Since pages
to
we study
access
page. of its
paths
the
rn
possible
last
dynamic
need
be
all
each
the
may
paper
tiles)
to
this
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. SIGMOD/PODS 94- 5/94 Minneapolis, Minnesota USA (3 1994 ACM 0-89791 -639-5/94/0005..$3.50
and
application
sent
We
failures
specifically,
a continuous *The support of the Defense Advanced Research Projects Agency, as well as the support of the Department of Energy under contract DE- ACO3- 76 SFOO09S is gratefully acknowledged.
may
n pages.
may
to
the
specify
of the
there
access
an access
must
environments.
(called
detail
each
to
paths
In this
repli-
frequently
disks,
n pages
until
Due
an optimal
the
are more
to retrieve
wait
transactions
dynamic
that
m parallel
to find
as disk
path.
photos that
disk
fly
concurrent
More
for this
used
disks.
access
visualization
currently
the
system,
as
improved
on the
of finding
failures.
reason
the
such
known
availability
load
efuse
repli-
variable
corresponding
to
it is important
the
with
a transaction
needs
the may
each page is only
servicing
paths
of disks
access
by
such
a transaction
from
disks
supported
It is well
to higher
are read-intensive.
a copy or
are
for
access
to reduce
times.
with
disk
repli-
response
applications
requesting
of the
data
slower
disks
pages
system
to
and
order
where
more
combinations
on
[BG88].
as mirrored
and
to
portion
disks
systems.
may
for a requested
least
are known
efforts
some
multiple
[RS92]
of fault-tolerant disks
research
replicating
across
sets
replication
alternative
many by
methods
shadow
vendors data
been times
database
replication
[PBD93]
In
have access
that
a transaction
alternative
In
of the
are replicated
arrives,
1
such
costs some
schemes subset
In a multi-disk
how
disadvantages
storage
time
on the
results
to show
obvious
as higher
fects
accessed
all replication
use simulation
also
such
cation
of the disks and
a lower
of a request
an
may be employed
are heterogeneous, We
related
of finding
it into
some
access
Laboratory
94720
CA
[email protected]
are
cation
system
two results
problem
time
lower
efficient
a multi-disk
how our method
environments servers.
the
by transforming
a preexisting
different
of finding contains
We solve
It’e also indicate
in dynamic have
data.
problem:
optimal
the problem
a set of pages from
Berkeley
E-mail:
[email protected]
There
work
plans
Rotem
Berkeley,
CA 94720
Abstract This
Data*
abstract)
Chen
Berkeley
Berkeley,
of Replicated
more
developed for the map
each tiles. and
Notation
Meaning
n
The
number
The .;j
total
number
1 if page
Response
R RoP~
Optimally
obtainable
t]; u] 7’7
current
Throughout represent in
the
mind
have
to be a disk
case,
unit
rest
Section a
the
graph
representing
of copies total
some
paper
disk
on
systems. the
discuss
the
above.
an
i should
Put
Mathematically,
this that
In
subject
formal
have
to the
to
a
of
data
over
from
multiple
a copy to
up
from
is used of pages
disks.
disks.
of page
come
pages
disk
m
with the
a way disks,
to retrieve that
corresponds
Assume
i on disk
j,
that and
and
that
and and
The
this shows
value
pages
that
as few pages
need
to be read
to the
response
any
can
taking
the
choose
discusses
.Ytj
other
be =
that
~
time
We
the
this for
(Rij)
~
