Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
Computer Science Technical Report Archive
/
USC Computer Science Technical Reports, no. 625 (1996)
(USC DC Other)
USC Computer Science Technical Reports, no. 625 (1996)
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
Placemen t of Data in MultiZone Disk Driv es
Shahram Ghandeharizadeh Douglas J Ierardi Dongho Kim Roger Zimmermann
Departmen t of Computer Science
Univ ersit y of Southern California
Los Angeles California F ebruary Abstract
This pap er describ es a placemen t tec hnique for the la y out of les across the surface of a m ulti
zone magnetic disk driv e to maximize its bandwidth It argues for a le system that monitors the
frequency of access to the les and mo dies the placemen t of les in order to resp ond to an ev olving
pattern of le accesses Sucha le system is particularly useful for rep ositories whose primary
functionalityis to disseminate information suc h as the W orldWideW eb along with thousands
of ftp sites that render do cumen ts graphic pictures images audio clips and fullmotion video
a v ailable on the In ternet Emplo ying soft w are tec hniques to enhance the disk p erformance is
imp ortan t b ecause its mec hanical nature has limited its hardw are p erformance impro v emen ts at
to percen tay ear
In tro duction
While micropro cessor and net w orking tec hnologies ha v e b een adv ancing at a rapid pace during the
past decade disk p erformance impro v emen ts ha v e been negligible resulting in the IO b ottlenec k
phenomenon The sp eedup for micropro cessor tec hnology is t ypically quoted at a rate of to percen t ann ually While disk storage densities are impro ving at a high rate from to p ercen t
ann ually p erformance impro v emen ts ha v e been rated at only to p ercen t ann ually R W A tec hnique emplo y ed b y mo dern magnetic disk driv es to impro v e their storage densit y is zoning Zoning in tro duces a disk with regions that pro vide dieren t transfer rates Dep ending on the c har
acteristics of a disk the fastest zone migh t pro vide a transfer rate to p ercen t higher than that
of the slo w est zone In this pap er w e study soft w are tec hniques that impro v e the p erformance of
This researchw as supp orted in part b y the National Science F oundation under gran ts IRI IRI
NYI a w ard and CD A and a HewlettP ac k ard unrestricted cashequipmen t gift
Spindle
Head
Arm
Platter
Cylinder
Arm
assembly
Figure A disk driv e
m ultizone disks b y con trolling the placemen t of data across the zones T o the best of our kno wl
edge no related w ork has app eared in the literature due to the recen t in tro duction of m ultizone
disk driv es W e sho w that one can enhance the p erformance of a m ultizone disk b y assigning the
frequen tly accessed les to the zone with the highest transfer rate W e use the term he at CABK to denote the frequency of access to a le
The rest of this pap er is organized as follo ws Section pro vides an o v erview of m ultizone
magnetic disks and transfer rates from t w o commercial disk driv es Assuming that the size of a le
system equals the storage capacit y of a magnetic disk driv e
Section describ es an optimal la y out
strategy that maximizes the p erformance of the disk driv e Section describ es an online reorganiza
tion tec hnique the enables the le system system to resp ond to an ev olving pattern of disk accesses
In addition this section describ es the design of a heat trac king mo dule that estimates the frequency
of access to eac h le Section quan ties the tradeos asso ciated with the prop osed tec hniques using
a tracedriv en sim ulation study The obtained results demonstrate signican t impro v emen ts using
the prop osed designs Section eliminates the assumption that the size of a le system equals the
storage capacit y of a disk and describ es t w o alternativ e designs for data la y out Our conclusion and
future researc h directions are con tained in Section Multizone magnetic disks
A magnetic disk driveis a mec hanical device op erated b y its con trolling electronics The mec hanical
parts of the device consist of a stac k of platters that rotate in unison on a cen tral spindle see R W for details Presen tly a single disk con tains one t w o or as man y as sixteen platters see Figure This assumption is relaxed in Section
Zone Size MB T ransfer Rate
Zone Size MB T ransfer Rate MBps
a HP C F ast Wide disk b Seagate STW F ast Wide disk
T able Tw o commercial disks and their zoning information
Eac h platter surface has an asso ciated disk head resp onsible for reading and writing data Eac h
platter is set up to store data in a series of trac ks A single stac k of trac ks at a common distance
from the spindle is termed a cylinder T o access the data stored in a trac k the disk head m ust be
p ositioned o v er it The op eration to rep osition the head from the currenttrac k to the desired trac k
is termed se ek Next the disk m ust w ait for the desired data to rotate under the head This time
is termed r otational latency In a nal step the disk reads the referenced data This time is termed
tr ansfer time Its v alue dep ends on the size of the referenced data and the transfer rate of the disk
T o meet the demands for higher storage capacit y magnetic disk driv e man ufacturers ha vein tro
duced disks with zones A zone is a con tiguous collection of disk cylinders whose trac ks ha v e the
same storage capacit y T rac ks are longer to w ards the outer p ortions of a disk platter as compared to
the inner p ortions hence more data ma y b e recorded in the outer trac ks While zoning increases the
storage capacit y of the disk it pro duces a disk that do es not ha v e a single transfer rate The m ultiple
transfer rates are due to the v ariable storage capacit y of the trac ks and a xed n um ber of
rev olutions per second for the platters T able presen ts the storage capacit y and transfer rate of
zones for t w o commercial disk driv es Both are SCSI fast and wide disks with a gigab yte storage
capacit y
Optimal zone la y out
Assume that the disk consists of a sequence of blo c ks partitioned in to zones Z
Z
Z
k
where
the ith zone Z
i
has a transfer rate of tfr Z
i
MBs The zones are ordered from fastest to slo w est
so that tfr Z
i
tfr Z
i for i k
T o impro v e the a v erage service time sustained b y a diskresiden t le w e prop ose to k eep the
hottest les the most p opular les with the highest frequency of access in the fastest zones
Sp ecically w e prop ose the follo wing Optimal Zone OZ La y out for a giv en set of les under the
assumptions that w eha v e an accurate estimate for the probabilit y of access of eac h le f
x
denoted
heat f
x
and the size of the set of les is no more than the capacit y of the disk First order the
les b y their heats from maxim um to minim um Then la y out the sequence of les con tiguously starting with zone Z
and pro ceeding progressiv ely to w ards the innermost zone Ob jects ma y span
m ultiple zones and no blo c ks are left uno ccupied
The OZ la y out assigns the hottest les to the fastest zones As one migh t exp ect this la y out in
fact minimizes the exp ected or a v erage service time for a le from the le system As discussed in
the previous section the actual service time of an y le includes time for seeking rotational latency
and transfer time Since ev ery access requires at least an initial seek and latency w e shall exclude
this amoun t from our calculations and fo cus on the transfer time and additional latencies that arises
from the fragmen tation of les Ho w ev er in the prop osed la y out all les are stored con tiguouslyso
none of these additional costs are observ ed
Without loss of generalit y assume that eac h le o ccupies only one zone If not w e can apply
the follo wing argumentb y considering eac h fragmen t of the le separately The actual transfer time
for a le f
x
in zone Z
i
is size f
x
tfr Z
i
so the exp ected transfer time for a diskresiden t le is
X
i
X
f x Z
i
heat f
x
size f
x
tfr Z
i
where the sum is computed o v er all zones Z
i
and les con tained therein
Lemma The la y out prop osed byOZisoptimal that is it minimizes the exp ected transfer rate and th us
the exp ected service time of eac h request when the giv en heat v alues pro vide an accurate estimate
for the probabilit y of access to anyle Pro of W e claim that the quan tityin is minimized when the heat of eac h le in Z
i
is greater
or equal to the heat of eac h le in Z
i for all i This is pro v ed b y con tradiction Assume for a
con tradicion that w ein factha v e an optimal le la y out in whichsome le f
in zone Z
i
that has a
lo w er heat than le f
in zone Z
i Then w e could increase the a v erage transfer rate bysw apping
blo c ks of f
and f
pro ving that this la y out could not be optimal Since an optimal la y out m ust
exist the argumen tsho ws that it can only b e ac hiev ed when the les are laid out so that the fastest
zones ha v e the hottest les Section describ es a tracedriv en sim ulation study that quan ties the p erformance impro v e
men ts that can b e ac hiev ed in this b estcase scenario
Dynamic online reorganization
It is not alw a ys the case that the static la y out describ ed in the previous section will be p ossible
b ecause it requires that all heat v alues to b e kno wn at the time that the la y out is realized on the
disk les are not added to or deleted from the le system and the heats remain constan t In
man y situations one w ould exp ect b oth the con ten t of the system andor the p opularit y of its les
to ev olveo v er time In this case w een vision a le system whic h main tains statistical records of
the p opularit y of its les and uses these computed heat v alues to reorganize data on the disk
migrating the hotter les to faster zones in order to appro ximate the la y out obtained using OZ
Accum ulating heat statistics
The le system m ust learn ab out the heat of les b y gathering statistics from the issued requests
Our design emplo ys the past pattern of access to estimate the iden tit y of les referenced frequen tly
in the future
Its learning pro cess is as follo ws The le system main tains a queue of timestamps for ev ery
le as w ell as an estimated heat v alue All the queues are initially empt y and the heat v alues are
uniformly set
to
n
where n is the total n um ber of les Up on the arriv al of a request referencing
le f
x
the curren t time is recorded in the queue of le f
x
Whenev er the timestamp queue of le f
x
b ecomes full the heat v alue of that le is up dated according to
heat
new
f
x
c K
P
K i t
i t
i
c heat
ol d
f
x
This is the case in our sim ulation and w ould b e true in an y situation where there is no reliable information a v ailable
on the exp ected heat of les Ho w everifsuc h information is a v ailable it could b e used in initializing the system
where K is the length of the timestamp queue c is a constan tbet w een and and t
x
is one individual
timestamp After the up date is completed the queue of this le is ushed and new timestamps can
b e recorded
This approac h is similar to the concept of the b ackwar d k distanc e used b y the authors of OO W in the LR U k algorithm The t w o sc hemes dier in three w a ys First the heat estimates are not
based on the in terv al bet w een the rst and the last timestamp in the queue but are a v erages o v er
all in terv als Second the heat of a le f
x
is only up dated when the timestamp queue of f
x
is full
therefore reducing o v erhead And third the previous heat v alue heat
ol d
f
x
is tak en in to accoun t
when heat
new
f
x
is calculated The ab o v e measures balance the need for smo othing out short
term uctuations in the access pattern and guaran teeing resp onsiv eness to longer term trends The
constan ts k and c could th us b e used to tune the mo dules b eha vior to a particular application
Migrating les among zones
In Section the optimal la y out for les w as attained b y ensuring that eachw as laid out con tiguously
along the surface of the disk th us a v oiding le fragmen tation Ho w ev er since the le system ma y
no w rearrange les on the disk there is the lik eliho o d that les will b ecome fragmen ted causing
the system to incur additional seeks and latencies So while the system migrates hotter les to
faster zones to impro v e their transfer time it ma y also degrade the exp ected service time of les b y
fragmen ting them
T o deal with these issues w e prop ose the follo wing p olicies for reorganization
On eac h access to a le f
x
if itis disco v ered that f
x
s heat has c hanged then it can b e mo v ed
or pr omote d to a hotter zone An y v alid data residing in its target lo cation will b e sw app ed
in to f
x
s curren t lo cation So in eect w e attempt to impro vethe p erformance of the system
only b y sw apping les pairwise although the target of the sw ap ma y in v olv e one or more
les or fragmen ts of les Moreo v er the sw ap is initiated to impro v e the transfer time of a
hotter le nev er to in ten tionally degrade the transfer time of a colder one
T o deal with fragmen tation that can arise eac h le that is promoted to a faster zone m ust
remain con tiguous In other w ords if the hot le that is targeted for migration is curren tly
con tiguously laid out then it remains con tinguous after its transfer to the new lo cation If the
le w as initially fragmen ted then it is reassem bled in a single con tiguous sequence of blo c ks
once mo v ed Note that the sw ap ma y result in the fragmen tation of the colder les curren tly
residing in the faster zone y et the p olicy tries to insure the in tegrit y of the more p opular les
on the disk
Giv en these p olicies the system m ust still c ho ose p oten tial target lo cations for the migrating le and
for eac h p oten tial target m ust determine whether the prop osed sw ap is w orth while These problems
are resolv ed in the follo wing w a y Initially up on a reference to le f
x
whose heat has c hanged w e determine the zone that it should
occup y with OZ sa y zone Z
optimal
Assume f
x
resides in zone Z
x
If Z
x
and Z
optimal
refer to the
same zone then no migration is necessary Otherwise f
x
is a candidate for migration from Z
x
to
Z
optimal
The system searc hes Z
optimal
for the coldest con tiguous sequence of blo c ks that can hold
f
x
T o calculate the heat of a sequence of blo c ks w e note that eac h blo c k of the disk that holds v alid
data inherits a heat from the le that o ccupies it whic h is just the frequency at whic h that blo c k
is accessed from the device The total he at of a sequence of blo c ks is then just the sum of the heat
of all blo c ks in the sequence So if the dierence in total heat bet w een f
x
and the curren t target
exceeds a preset threshold the le is sw app ed in to this lo cation This heat dierence is prop ortional
to the amountbywhic h the total service time is exp ected to decrease if the sw ap is made excluding
the time for an y additional seeks that ha v e been in tro duced
If the dierence do es not exceed the
threshold Z
optimal
is reset to Z
optimal the next slo w er zone and this pro cess is rep eated This
pro cess terminates when Z
optimal
Z
x
A more detailed description of the algorithm is presen ted in Figure The pro cedure is con
serv ativ e to the exten t that it attempts to main tain the con tiguit y of the more p opular les and
nev er tries to mo v e a le bey ond its placemen t in the optimal static sc hedule This is an attempt
to promote les while a v oiding comp etition with other presumably hotter les that should reside
in the faster outermost zones The pro cedure also relies on the congurable threshold parameter
denoted THRESHOLD in Figure that is used to tune the sensitivit y of the reorganization pro
cedure a sw ap is done only if the gain in exp ected service time exceeds this threshold F or smaller
v alues of this parameter les are migrated frequen tly in resp onse to small v ariations in heat as the
parameters v alue is increased les will migrate only if their heat c hanges b y a large amoun t or
if the les are themselv es large In this w a y the parameter helps to quan tify the tradeo bet w een
the onetime cost of mo ving a le and the exp ected reduction in service time that will b e observ ed
during future accesses
A more accurate measure of the c hange in service time can b e computed b y estimating the service time for b oth
the curren t and prop osed la y outs of the les in v olv ed in the sw ap w eigh ting these b y the les heats and determining
the eect up on the o v erall a v erage service time This approac h is discussed further in Section
move f f
if the le alr e ady r esides in the fastest zone
then return A t this p oin t all the heats should b e curren t curr ent lo c ation blo cks curr ently o c cupie d by f curr ent zone slowest zone in which any fr agment of f r esides tar get zone b est static plac ement of f for curr ent he ats
while curr ent zone tar get zone f
tar get lo c ation a se quenc e of size f blo cks of minimum total
he at in tar get zone if total he at curr ent lo c ation total he at tar get lo c ation THRESHOLD f
swap the c ontents of curr ent lo c ation and tar get lo c ation break g
else
curr ent zone g
return g
Figure Pro cedure for attempting to promote a le during online reorganization
T race driv en ev aluation
Weemplo y ed a trace driv en sim ulation study to in v estigate the tradeos asso ciated with the prop osed
tec hniques The trace is a sequence of le accesses to a ftp site at the Univ ersit y of California
Berk eley skftpCSBerk eley EDU It w as accum ulated o v er a p erio d of mon ths from to The le system consists of les and is gigab ytes in size The trace con tains
neither the creation nor deletion date of a le only the references to retriev e a le It consists of
requests A few les are hot and referenced more frequen tly than the others On the a v erage
a request retriev es kilob ytes of data
Sim ulation mo del
F or the purp oses of this ev aluation w e represen ted a disk driv e as an arra y of blo c ks The size of
eachbloc k corresp onded to a ph ysical disk blo c k and is b ytes long A zone o ccupied a con tiguous
n um ber of blo c ks The sim ulator main tained the n um ber of blo c ks that constituted a le and the
lo cation of eac h blo c k As w e did not ha v e the zone c haracteristics of a Gigab yte disk driv e w e
emplo y ed the zone c haracteristics of the Seagate STW disk for this ev aluation The storage
capacity of this disk is of the le system size W e increased the size of eachzoneb y a factor of
four to store the le system
W e emplo y ed the analytical mo dels of GSZ to represen t the seek op eration and latency of
a magnetic disk driv e These mo dels estimate the c haracteristics of the Seagate STW disk
and are similar to those of R W W GPW The minim um and maxim um seek times are and
milliseconds resp ectiv ely The a v erage rotational latency time is milliseconds When a le
transfer w as initiated the sim ulator reads its ph ysically con tiguous blo c k in one transfer eliminating
seek and rotational latency for t w o or more ph ysically adjacen t blo c ks A seek is incurred if a le
is fragmen ted across the surface of the disk The sim ulator computes the n um ber of seeks incurred
when reading a le W e assume all the references are directed to the le system and do not observ e
a memory hit
Optimal zone la y out
W e analyzed the p erformance of the system with the OZ la y out F or comparison purp oses w e
analyzed the w orst case scenario that la ys the les on the surface of the disk in the rev erse order
assigning the frequen tly accessed les to the innermost zones The a v erage service time of the system
is milliseconds and milliseconds with optimal and w orst case assignmen t resp ectiv ely The
OZ la y out pro vides a p ercen t impro v ementrelativetothe w orst case
The a v erage service time of the disk driv e based on a random assignmen t of les to the disk
is estimated to be msec This n um ber is computed based on the probabilit y of a request
referencing data stored in a zone for zone Z
i
this probabilityis
siz e of Z
i
disk capacity
the time required to
retrievethe a v erage n um ber of b ytes p er request from zone Z
i
av er ag e r eq uest siz e
tf r Z
i
and the a v erage
seek and rotational dela ys This a v erage estimate w as v eried b y running the sim ulator sev eral times
assuming a random assignmen t of les OZ pro vides a p ercen t impro v emen t as compared to this
a v erage service time
Heat trac king
In order to quan tify the impact and eectiv eness of the conguration parameters K and c Equa
tion of the heat trac king mo dule w e analyzed its p erformance b y computing the ro ot mean square
rms deviation W al of the estimates computed b y the online heattrac king mo dule from the
o v erall v alues computed o v er the duration of the run That is w e consider the qualit y factor Q Q s
P
N
i E f
i
R f
i
N
where N is the total n um ber of les on the disk E f
i
is the estimated heat v alue for le f
i
com
puted using the trac king mo dule and R f
i
denotes the real heat v alue calculated as the fraction
number of accesses to f
i
total number of accesses
Of course this is not necessarily the only measure to use here W e are in
eect trac king a mo ving target the c hanging p opularit y of les Nev ertheless this v alue can
pro vide a go o d estimate of howm uchc hange there is among the heats of individual les throughout
the run and ho w resp onsiv e or unresp onsiv e the heattrac king mo dule is to lo cal uctuations
The deviation Q w as computed after ev ery requests during the sim ulation of the trace
W e in v estigated the follo wing c v alues and F or eac h of the c v alues the heat
statistics queue length K w as set to and for dieren t exp erimen ts The results indicate
that for this particular trace the best ie smallest o v erall deviation can be ac hiev ed with either
K and c or K and c although in general the rms deviations are relativ ely
consisten t A small c v alue that fa v ors the more recen t history of the frequency of access is more
accurate at system startup time ho w ev er its margin of error increases as the n um ber of issued
requests increase Short heat queues ie either K or K lead to more resp onsiv e heat
trac king ho w ev er they also increase the uctuation in the estimated heat v alues resulting in a
higher margin of error
Online reorganization
W e analyzed the online reorganization tec hnique b y assuming the w orstcase scenario for the initial
placemen t of les on the disk Next the sim ulator w as in v ok ed to learn the heat of les K c in Equation and w as allo w ed to migrate the frequen tly accessed les from the slo w er zones
to the faster ones using the algorithm dev elop ed earlier W e tallied the a v erage service time of the
disk for eac h requests Finally w e computed the p ercen tage impro v emen t in service time
with reorganization as compared to the w orstcase la y out with no reorganization Figure sho ws the
obtained results In this gure ev ery tic k on the xaxis represen ts requests appro ximately
da ys of op eration The yaxis represen ts the p ercen t impro v emen t F or comparison purp oses
this gure also sho ws the p ercen t impro v emen t
in service time with OZ relativ e to the w orst case
Assuming that Sla y out is a function that denotes the a v erage service time of the system with a la y out sa y OZ
as compared to w orst is computed at S wor st S OZ S wor st This n um ber is alw a ys less than
100000 200000 300000 400000
Accumulated number of requests
0
10
20
30
40
50
60
OZ
On-line reorganization
%
Percent improvement
Figure P ercen t impro v emen t with online reorganization and optimal
la y out with no reorganization
During the rst requests the reorganization mo dule impro v es the a v erage service time
b y p ercen t This impro v emen t increases to p ercen t during the next requests b ecause
the heat trac king mo dule pro vides a b etter estimate of le heats The reorganization mo dule ap
pro ximates the p ercen t impro v ementpro vided b y optimal It cannot pro vide a b etter impro v emen t
b ecause it is appro ximating the heat of les while OZ has organized the les based on p erfect
heat statistics that is heat statistics corresp onding to global a v erages across the en tire trace The
amoun t of fragmen tation attributed to the reorganization pro cess is negligible A t the end of sim ula
tion the reorganization pro cess resulted in four h undred noncon tiguous c h unks due to fragmen tation
as compared to zero at system startup
Discussion
The sim ulation results of the previous section assumed that the size of the le system w as ab out equal
to the capacit y of the disk and th us that the les w ere pac k ed in to zones lea ving no free space
In the case where the le system is smaller than the disk capacit y the system ma y c ho ose bet w een
t w o alternativ es when assigning les to the zones it can either try to ll completely eac h of the
fastest zones to tak e adv an tage of their higher transfer rate or distribute the les more ev enly
The prop osed online reorganization pro cedure ho w ev er made no assumptions in this regard
across the disk lea ving free space in eac h of the zones Filling up eac h of the fastest zones completely
has the adv an tage of utilizing the whole space of these zones and therefore maximizes the a v erage
observ ed transfer of the disk for its residen t les This can poten tially giv e the greatest p ossible
p erformance enhancemen t Ho w ev er this approac h migh t suer from the follo wing limitations
greater di!culties during online compaction le creation and up date whic h are more lik ely to lead
to the fragmen tation of les
On the other hand if w e do not ll up eac h zone completely there is the lik eliho o d that online
compaction le creation and le up date will b e more e!cien t and eectiv e and that fragmen tation
can be minimized during these op erations b ecause of the a v ailable free space within zones By
comparison the Unix F ast File System MJLF groups the cylinders of a hard disk driv e as cylinder
groups It then tries to lo calize related les within a cylinder group while also spreading unrelated
les across cylinder groups in order to impro v e p erformance b y minimizing seek time T o ac hiev e
this it usually reserv es percen t of the space as free space to do compactions In addition it
is lik ely that the initial stage of ob ject promotions can be made c heap er and that the p ossibilit y
of m ultiple mo v es b oth pingp onging and domino eects can be reduced Ho w ev er this strategy
do es not fully utilize the capacit y of the fastest zones Therefore it cannot exp ect to ac hiev e the b est
p ossible p erformance
It is lik ely that these tradeos can best be managed in an application dep enden t w a y F or
example the amoun t of free space reserv ed in eac h zone could be tuned to the pro jected stabilit y
of le system ie ho w often les are created deleted and up dated and ho wm uchv ariation in the
heat of les is exp ected
In the trace driv en sim ulation w e also did not deal with the issue of le creation since le
creation times w ere not explicitly represen ted in the trace Nev ertheless the creation of new les
can be handled within the framew ork dev elop ed ab o v e Sp ecically if w e are giv en a prosp ectiv e
heat for the newly created le w e can determine from this what its appropriate zone should b e If w e
ha v e free space for the le in this zone it is inserted there otherwise the reorganization algorithm
is applied to this le If this also fails the le can then b e placed in to the zone with the most free
space whic h is most lik ely the innermost or slo w est zone Up dates can be handled in a similar
manner
Conclusion and F uture Directions
This study describ es an optimal placemen t tec hnique for the assignmen t of les to the zones of a disk
driv e This tec hnique striv es to assign the frequen tly accessed les to the fastest zones in order to
maximize the disk bandwidth In order to resp ond to ev olving access patterns w e outlined an online
reorganization pro cess that monitors the frequency of access to the les and migrates the frequen tly
accessed ones to the faster zones up on reference Our trace driv en sim ulation study demonstrates
that the optimal la y out impro v es the a v erage service time of the system eectiv e heat trac king
mo dules can b e constructed and the online reorganization pro cess can appro ximate the p ercen t
impro v emen t pro vided b y the optimal organization As discussed in Section the reorganization
pro cess can b e netuned based on the c haracteristics of a target application
As part of our future researc h activit yw ein tend to analyze alternativem ultizone disk driv es and
traces obtained from other ftp sites In addition w ein tend to analyze the concept of striping using
m ultizone disk driv es When a le is strip ed across m ultiple disks its retriev al time is determined
b y the fragmen t assigned to the slo w est zone of the participating disks Is it imp ortan t to con trol the
assignmen t of dieren t fragmen ts of a le to the zones of a v ailable disks" Do es this impact the size
of a fragmen t" What is the impact of this on the paritybloc k and its computation" Some of these
issues are addressed in the con text of heterogeneous disks CRS and w e in tend to extend these
designs to m ultidisk en vironmen ts consisting of m ultizone disk driv es W e in tend to in v estigate
these design decisions along with the implemen tation details of the prop osed ideas using the Unix
op erating system
References
CABK G Cop eland W Alexander E Bough ter and T Keller Data Placemen t in Bubba In
Pr o c e e dings of A CM SIGMOD pages CRS L T Chen D Rotem and S Seshadri Declustering Databases on Heterogeneous Disk
Systems In Pr o c e e dings of V ery L ar ge Datab ases GSZ S Ghandeharizadeh J Stone and R Zimmermann T ec hniques to Quan tify SCSI
Disk Subsystem for Multimedia USC T ec hnical Rep ort Univ ersit y of Southern
California MJLF M Mc kusic k W Jo y S Le#er and R F abry A Fast File System for UNIX A CM
T r ansactions on Computer Systems August OO W E J ONeil P E ONeil and G W eikum The LRUK Page Replacemen t Algorithm
for Database Disk Buering In Pr o c e e dings of A CM SIGMOD pages
R W C Ruemmler and J Wilk es An In tro duction to Disk Driv e Mo deling IEEE Computer Marc h
W al P R W allace Mathematic al A nalysis of Physic al Pr oblems Do v er W GPW B W orthington G R Ganger Y N P att and J Wilk es OnLine Extraction of SCSI
Disk Parameters In Pr o c e e dings of A CM SIGMETRICS pages
Abstract (if available)
Linked assets
Computer Science Technical Report Archive
Conceptually similar
PDF
USC Computer Science Technical Reports, no. 598 (1994)
PDF
USC Computer Science Technical Reports, no. 628 (1996)
PDF
USC Computer Science Technical Reports, no. 650 (1997)
PDF
USC Computer Science Technical Reports, no. 627 (1996)
PDF
USC Computer Science Technical Reports, no. 634 (1996)
PDF
USC Computer Science Technical Reports, no. 610 (1995)
PDF
USC Computer Science Technical Reports, no. 766 (2002)
PDF
USC Computer Science Technical Reports, no. 592 (1994)
PDF
USC Computer Science Technical Reports, no. 659 (1997)
PDF
USC Computer Science Technical Reports, no. 615 (1995)
PDF
USC Computer Science Technical Reports, no. 623 (1995)
PDF
USC Computer Science Technical Reports, no. 612 (1995)
PDF
USC Computer Science Technical Reports, no. 630 (1996)
PDF
USC Computer Science Technical Reports, no. 685 (1998)
PDF
USC Computer Science Technical Reports, no. 619 (1995)
PDF
USC Computer Science Technical Reports, no. 602 (1995)
PDF
USC Computer Science Technical Reports, no. 699 (1999)
PDF
USC Computer Science Technical Reports, no. 748 (2001)
PDF
USC Computer Science Technical Reports, no. 589 (1994)
PDF
USC Computer Science Technical Reports, no. 629 (1996)
Description
Shahram Ghandeharizadeh, Douglas J. Ierardi, Dongho Kim, and Roger Zimmermann. "Placement of data in multi-zone disk drives." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 625 (1996).
Asset Metadata
Creator
Ghandeharizadeh, Shahram
(author),
Ierardi, Douglas J.
(author),
Kim, Dongho
(author),
Zimmermann, Roger
(author)
Core Title
USC Computer Science Technical Reports, no. 625 (1996)
Alternative Title
Placement of data in multi-zone disk drives (
title
)
Publisher
Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA
(publisher)
Tag
OAI-PMH Harvest
Format
14 pages
(extent),
technical reports
(aat)
Language
English
Unique identifier
UC16270254
Identifier
96-625 Placement of Data in Multi-Zone Disk Drives (filename)
Legacy Identifier
usc-cstr-96-625
Format
14 pages (extent),technical reports (aat)
Rights
Department of Computer Science (University of Southern California) and the author(s).
Internet Media Type
application/pdf
Copyright
In copyright - Non-commercial use permitted (https://rightsstatements.org/vocab/InC-NC/1.0/
Source
20180426-rozan-cstechreports-shoaf
(batch),
Computer Science Technical Report Archive
(collection),
University of Southern California. Department of Computer Science. Technical Reports
(series)
Access Conditions
The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.
Repository Name
USC Viterbi School of Engineering Department of Computer Science
Repository Location
Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089
Repository Email
csdept@usc.edu
Inherited Values
Title
Computer Science Technical Report Archive
Coverage Temporal
1991/2017
Repository Email
csdept@usc.edu
Repository Name
USC Viterbi School of Engineering Department of Computer Science
Repository Location
Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089
Publisher
Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA
(publisher)
Copyright
In copyright - Non-commercial use permitted (https://rightsstatements.org/vocab/InC-NC/1.0/