Вы находитесь на странице: 1из 6

Real-time Disk Scheduling Algorithms Based on the Two-way SCAN Technique

Myung-Sub Lee
Div. oI Computer Technology
Yeungnam College oI Science & Technology
Daegu, Korea
skydreamync.ac.kr

Kwang-Jung Kim Chang-Hyeon Park
Dept. oI Computer Engineering
Yeungnam University
Gyeongsan, Gyongbuk, Korea
kim7463ynu.ac.kr parkyu.ac.kr
Abstract-Recently various real-time scheduling algorithms
are being studied in order to maximize throughput per unit
time in real-time disk scheduling. Particularly in many
algorithms, SCAN technique is applied to EDF(Earliest
Deadline First), which represents real-time disk scheduling
algorithms. However, proposed algorithms have many
restrictions because they have to consider I/O requests that
continue to be made while SCAN groups are being created. In
addition, the fixed service direction in the SCAN technique
lowers the efficiency of the disk. In order to solve the problem,
this paper proposed a new real-time disk scheduling algorithm
based on the insertion technique and the two-way SCAN
technique and proved the superiority of the proposed system in
terms of throughput per unit time and serviceable I/O requests
through an experiment.
Keywords-component; disk scheduling; EDF; insertion
technique; two-way SCAA technique; hard real-time system
I. INTRODUCTION
A disk scheduling algorithm is a technique oI setting the
order oI disk I/O requests aiming at improving the
perIormance oI disk system. To improve the perIormance oI
disk system, we must consider three aspects: Iirst, throughput
per unit time that measures how many disk I/O requests are
served during speciIic unit time; second, average response
time that measures how Iast each disk I/O request is
processed; and third, the predictability oI response time used
in the variance oI response time as an element to measure
predictability. Considering these aspects, various disk
scheduling algorithms have been proposed|1|. However,
because these algorithms are Iocused on disk eIIiciency but
do not consider the ending time oI real-time disk I/O requests,
they are not adequate Ior real-time systems. In response to
the problem, a number oI real-time disk scheduling
algorithms have been proposed.
To use the disk eIIiciently while processing disk I/O
request within the ending time limit, a real-time disk
scheduling algorithm decides the order oI disk I/O requests.
Existing real-time disk scheduling algorithms include EDF
(Earliest Deadline First) |5, 6|, P-SCAN (Priority SCAN)|7|
and SCAN-EDF|8| that added the SCAN technique to EDF,
SSEDO/SSEDV (Shortest Seek Time and Earliest Deadline
by Ordering/Value)|9| that uses queues, UG-SSTF (Urgent
Group and Shortest Seek Time First)|10| that is based on the
urgency oI disk I/O requests, MSG (Maximum Scannable
Group)|11| that expanded SCAN groups, and DMS
(Deadline ModiIication SCAN)|12| that modiIied ending
time using MSG. These real-time disk scheduling algorithms,
however, lower disk eIIiciency as they process disk I/O
requests in order oI ending time. To solve this problem, disk
eIIiciency was improved with scheduling algorithms using
the SCAN technique but they have a limitation in that they
can Iorm SCAN groups only with consecutive requests.
What is more, the Iixed service direction oI the SCAN
technique lowers disk eIIiciency.
In order to solve the problem, the present research
proposed a new real-time disk scheduling algorithm based on
the insertion technique that can eIIectively insert
inconsecutive real-time requests into SCAN groups, the
SCAN merge technique that can merge consecutive SCAN
groups, and the two-way SCAN technique that can improve
the perIormance oI services using the SCAN technique.
II. RELATED WORK
When deciding the order oI services, existing non-real-
time disk scheduling algorithms such as SCAN |2| and
SSTF(Shortest Seek Time First)|3, 4| consider only the
physical location oI data demanded by each I/O request, so
they are not adequate Ior processing real-time disk I/O
requests that must be completed within the ending time limit.
In order to solve the problem, a number oI real-time disk
scheduling algorithms have been proposed and a lot oI
relevant research results have been reported. Representative
algorithms are real-time disk scheduling algorithms as
presented below.
A. EDF(Earliest Deadline First)
The EDF is the most well-known real-time disk
scheduling algorithm, which processes the request closet to
the ending time Iirst. That is, the request oI the shortest
ending time has the top priority and is processed Iirst. EDF is

known to be optimal when the service time oI each disk I/O
request is known in advance. However, because EDF
algorithm does not consider a method oI reducing the seek
time oI the disk head in processing disk I/O requests, disk
eIIiciency is quite low.
B. P-SCAN(Prioritv SCAN)
The P-SCAN is based on the strategy that all I/O requests
in the disk queue are classiIied into a number oI priority
levels. In addition, each priority level processes I/O requests

using SCAN. AIter I/O requests belonging to the highest
priority level are all processed, those oI the second highest
priority are processed. AIter the service oI each I/O request,
the disk scheduler checks iI there are I/O requests oI higher
2009 Eighth IEEE International Conference on Embedded Computing; IEEE International Conference on Scalable Computing and Communications
978-0-7695-3825-9/09 $26.00 2009 IEEE
DOI 10.1109/EmbeddedCom-ScalCom.2009.33
137
2009 Eighth IEEE International Conference on Embedded Computing; IEEE International Conference on Scalable Computing and Communications
978-0-7695-3825-9/09 $26.00 2009 IEEE
DOI 10.1109/EmbeddedCom-ScalCom.2009.33
137
2009 Eighth IEEE International Conference on Embedded Computing; IEEE International Conference on Scalable Computing and Communications
978-0-7695-3825-9/09 $26.00 2009 IEEE
DOI 10.1109/EmbeddedCom-ScalCom.2009.33
137
International Conference on Scalable Computing and Communications; The Eighth International Conference on Embedded Computing
978-0-7695-3825-9/09 $26.00 2009 IEEE
DOI 10.1109/EmbeddedCom-ScalCom.2009.33
137
priority waiting Ior service and, iI there are, the scheduler
processes them by raising the priority level. A disadvantage
oI the algorithm is that, because the acceptance control
strategy is not adopted, requests oI low priority are not
processed Ior a long time and as a result the ending time
limit may be exceeded.
C. SCAN-EDF
The SCAN-EDF is an algorithm with reduced seek time
as EDF processes requests with similar ending time limit in
the SCAN method. II disk I/O requests arrive, they are lined
up in the queue in order oI ending time limit and, as in EDF,
requests oI short ending time limit are processed Iirst. II
there are I/O requests oI the same ending time limit, tracks
where requested data are positioned are searched in sequence
in order oI SCAN service. However, it is quite rare Ior disk
I/O requests to have the same ending time lime. Accordingly,
SCAN-EDF mostly works like EDF, so its disk eIIiciency is
still as low as that oI EDF.
D. SSEDO/SSEDJ(Shortest Seek Time and Earliest
Deadline bv Ordering/Jalue)
The SSEDO/SSEDV algorithms proposed by Chen, line
up I/O requests in the disk queue in order oI ending time
limit and schedules in a window deIined with m top I/O
requests in order to reduce the execution time oI the
algorithms. The two algorithms give priority to I/O requests
in the window using diIIerent methods, and process requests
Irom those with the least value oI priority. When
i
d is the
ending time limit oI the i th request among I/O requests in
the queue and
i
l is the seek distance Irom the current
position oI the head, priority
i
P is deIined as Iollows.
SSEDO:
. ,..., 2 , 1 ), 1 (
,..., 2 , 1 ,
1
m i i where
m i d w P
i
i i i
= > =
= =

| |

(1)
SSEDV:
. ,..., 2 , 1
), 1 0 ( ) 1 (
m i
l d P
i i i
=
s s + + = o o o


I/O requests in the window are given priority by equation
1 and processed Irom the request oI the least value oI priority.
That is, requests with short seek distance and ending time
limit has higher priority. Because SSEDO/SSEDV can
reduce the average excess oI ending time limit only when
even I/O requests exceeding ending time are processed
quickly, they are inadequate Ior hard real-time systems that
do not need to process I/O requests exceeding ending time
limit.
E. UG-SSTF(Urgent Group and Shortest Seek Time First)
The UG-SSTF is an algorithm that divides disk I/O
requests into three groups according to urgency, and
processes I/O requests in the same group Irom those with
short seek time. I/O requests exceeding the ending time limit
belong to MG (Miss Group), urgent ones to UG (Urgent
Group) and those with a long time to the ending time limit to
DG(DeIerable Group). As this algorithm processes I/O
requests divided into three groups, it lowers the rate oI
lateness and provides a mechanism that controls the balance
between the rate oI lateness and the excess oI ending time
limit, which are key Iactors oI perIormance in soIt real-time
systems. Like SSEDO/SSEDV as described above, however,
UG-SSTF is also inadequate Ior hard real-time systems.
F. MSG(Maximum Scannable Group)
The MSG deIines extended SCAN groups that resolve
the problem oI SCAN-EDF that processes only requests with
similar ending time limit in the SCAN method. SCAN group
means a group oI requests that can be processed with the
SCAN technique within the ending time limit. When there is
a set oI disk requests scheduled with EDF, iI its extended
SCAN groups MSGs are created, the i th extended SCAN
group
i
MSG is deIined as equation 2. MSGs have a
restriction that only consecutive disk requests can be put
together into a SCAN group.

i k for
a d f satisfies which
R R R MSG
f i k
m i EDF i EDF i EDF k
=
s
=
+ +
:
,..., ,
) ( ) 1 ( ) (

(2)
G. DMS(Deadline Modification SCAN) first
The DMS changes the ending time limit oI requests, the
ending time oI which is longer than the ending time oI MSG,
among request included in MSG to the ending time oI MSG.
In addition, it compares the ending time limit oI requests and
maintains requests processed earlier to have ending time
limit shorter than or equal to that oI requests processed later.
As a result, a larger number oI disk requests can be
processed. However, DMS algorithm also has the
disadvantage oI MSG that only consecutive requests can be
put together into scan groups. Moreover, due to the Iixed
direction between SCAN groups, there can be time loss.
In the real-time disk scheduling algorithm as examined
above, iI requests are processed using SCAN-EDF, MSG,
DMS, etc. all requests can be processed with the ending time
limit and the response time oI requests can be signiIicantly
reduced. However, there are still problems in deciding
SCAN groups and in the service direction oI SCAN as
mentioned above.
The present research proposed a new real-time disk
scheduling algorithm based on the insertion technique and
the two-way SCAN technique, which can increase the
number oI real-time I/O requests processed and disk
throughput per unit time by solving problems in deciding
SCAN groups and the service direction oI SCAN.
III. PROPOSED ALGORITHM
The real-time disk scheduling algorithm proposed in this
research is composed oI insertion technique that can insert
inconsecutive requests into proper SCAN groups when
deciding SCAN groups and SCAN group merge technique
that can merge consecutive SCAN groups, and two-way
SCAN technique that can decide the direction oI SCAN in an
eIIective way.
138 138 138 138
A. Definition of Problem
In real-time disk request
i i i i i
b a d r R , , , = , the location oI
data track a is related to seek time and data capacity
i
b to
data transIer time. TransIer time is deIined as
i
T
(data
TransIer time), maximum rotational latency time as
(Maximum Rotational latency time), actual seek time as
i
S ,
and maximum seek time as MS (Maximum Seek time). In
addition, the sum oI MR and
i
T is deIined as
i
MRT . In
Iigure 1, a set oI real-time disk requests
2 1
, ,
+ +
=
i i i
R R R R
is the
result oI scheduling in the EDF technique. Here,
i
MSG is
2 1
, ,
+ +
=
i i i
R R r R
.

M: Ml
i
M: Ml
i1
M: Ml
i2
l
i
l
i1
l
i2
d
i
d
i1
d
i2
imc LDF
Ml
i
= Ml Muximum louionul lucn imc)

i
runsIcr imc)
M: Ml
i
M: Ml
i
M: Ml
i1
M: Ml
i1
M: Ml
i2
M: Ml
i2
l
i
l
i1
l
i2
d
i
d
i1
d
i2
imc LDF
Ml
i
= Ml Muximum louionul lucn imc)

i
runsIcr imc)

Figure 1. Results oI EDF scheduling
As shown in Fig. 1,
i
T in cannot be reduced, and even iI
MR is reduced we cannot expect signiIicant improvement in
disk perIormance. Consequently, because MS exists in each
disk request and is independent Irom the request, we can
improve disk perIormance by decreasing the size and the
number oI MS . Table 1 shows parameters used in this
research.
TABLE I. PARAMETERS RELATED TO REAL-TIME SCHEDULING
Variable Description
R
A set oI requests
n
R R R R ,..., ,
1 0
=
i
R A real-time disk requests,
i i i i i
b a d r R , , , =

i
r Release time oI
i
R
i
d Deadline oI
i
R
i
a The track location oI
i
R
i
b The data capacity oI
i
R
fi
c
The service time oI
i
R ,
when it`s severed aIter
f
R
Z
R
The schedule result oI R by the algorithm Z ,
) ( ) 1 ( ) 0 (
,..., ,
n Z Z Z Z
R R R R =
) (n Z
f The schedule IulIill-time oI
Z
R
i
MSG

The maximum-SCANable-group stated Irom
i
R
i
m The count oI request oI
i
MSG

i
T
The data transIer time oI
i
R
i
S
The seek time oI
i
R
MS The maximum seek time
MR The maximum rotational latency time
i
MRT

MR
i
T

o
The sum oI time Ior stop the disk head and accelerate the
disk head
B. Two-wav SCAN technique
This research proposed the insertion technique that can
insert inconsecutive appropriate request within the limit oI
the ending time oI SCAN groups when deciding SCAN
groups. Each real-time disk requests is independent, the i th
MSG is
n i i
R R MSG ,..., = and when n i s , the maximum seek
time oI MSG is
) (
i
m MS + o
. Here, o is sum oI time to
stop the disk head and time to accelerate the head.
i
m
is the
number oI requests belonging to
i
MSG . In addition, because
MSG is a request group arranged in the SCAN method,
n i
a a <
and the data tracks oI all requests belonging to
i
MSG
are positioned between
i
a
and
n
a
. Because MS is the
maximum seek time oI the disk head, the sum oI the seek
time( S ) oI all requests belonging to
i
MSG is smaller than or
at best equal to MS as in equation 3.

=
>
n
i k
k
S MS

(3)
As Ior actual seek time, however, because the time oI
o

exists Ior each requests, equation 3 is modiIied to equation 4.

=
> +
n
i k
k i
S m MS ) (o

(4)
The actual seek time oI
i
MSG is ) (
i
m MS + o , but the
seek time oI MSG considered in EDF is
MS m
i

. Thus, aIter
processing all requests oI
i
MSG , there is a spare time as long
as the diIIerence between
MS m
i

and ) (
i
m MS + o . That is,
even iI requests with MRT are inserted during the spare time,
any request belonging to MSG does not violate the ending
time. However, because there is time Ior the head to move
between MSG and MSG , there must be MS given one more
time. In addition, because the value oI o is extremely small,
i
m o is smaller than MS and consequently ) (
i
m MS + o in
equation 4 is changed to
MS 3
. MRT oI an insertable
request is the product oI
MS
and
3
i
m
as in equation 5.

n i and R R MSG where
m MS
MS m MS request insert of Time
n i i
i
i
< = =
=
=
,...
) 3 (
3

(5)

As shown in equation 5, the larger the value oI
i
m
is the
higher the perIormance may be. That is disk perIormance
may be enhanced using an algorithm that can merge diIIerent
MSGs. In this research, iI the track location oI the last
request oI a SCAN group produced by the insertion
technique and MSG is larger than the track location oI the
139 139 139 139
Iirst request oI the next SCAN group, the two groups were
merged. That is, iI the track location
f
a
oI the last
request(
f
R
) oI
i
MSG is smaller than the track location
k
a
oI
the Iirst request(
k
R
) oI
k
MSG the two MSGs can be merged.
When
n i i
R R MSG ,..., = , the entire execution
time(
) (n MSG
f ) is the same whether requests are processed
Irom
i
R
to
f
R
or reversely Irom
f
R
to
i
R
. When the MSG
next to
i
MSG is
k k k
R R MSG ,..., = , the Iixed SCAN technique
processes requests, moving the head Irom
i
a
to
f
a
,
k
a
and
n
a . Here, iI the distance between
n
a
and
f
a
is shorter than
that between
k
a and
f
a , it is more eIIective to move Irom
f
a to
n
a rather than to
k
a . That is, the changeable two-way
SCAN technique can show higher disk perIormance than the
Iixed SCAN technique.
C. Execution procedure of the proposed
Figure 2 shows the entire structure oI the proposed
algorithm, and the execution procedure is as Iollows. First, iI
a real-time disk request comes in, acceptance control is
applied. An insertion index is created Ior each request that
has passed through acceptance control, MSG is composed
Ior scheduled requests. MSGs are merged, iI possible, with
one another through the SCAN merge technique, and
modiIied MSG is composed in the insertion technique using
then insertion index. The ending time oI requests in the
modiIied MSG is changed using the DMS technique, and
lastly the direction oI SCAN is decided using the two-way
SCAN technique. This procedure can be described in
detailed steps as Iollows.

Q Insert into ready list: Insert all oI arriving I/O requests
(real-time I/O requests) into the ready list.
Acceptance control: Insert only requests that satisIy
acceptance control to the EDF list. Initially the number oI
requests accepted is similar to that oI EDF but a larger
number oI request may be accepted in the acceptance
control oI the next time because the DMS technique is
applied, which modiIies insertion real-time disk
scheduling and ending time.
Create insertion index: II the set oI requests belonging to
the EDF list is
2 1
, ,
+ +
=
i i i
R R R R
and i is n i s s 0 ,
insertion indexes are created according to the multiple oI
MS by comparing
i
MRT oI request(
i
R
) with MS . An
insertion index is composed oI a linked list Ior easy
insertion and deletion and the internal list indicates the
location oI the EDF list oI request(
i
R
). Insertion indexes
are coded as threads, so they are created continuously iI
requests come even during execution. In addition, when a
request is included in MSG , the index oI the request is
deleted.
Q EDF scheduling: Requests in the EDF list are scheduled
into EDF Irom the request with the shortest ending time
limit. Here, the EDF list contains requests that have
passed through only acceptance control. Scheduling into
EDF is done in order oI ending time limit.

M:C
i
M:C
i1
M:C
i2
lcud
Lis
LDF
Lis
o.
!nscrion !ndcx
4. LDF
slcdulin_
5. M:C
2. udmission
onrol
7. inscr
b. M:C mcr_cr
10. :C.N dircion dcision
& scrvic
. u hol dircion
:C.N ul_orilms
:C.N _rou
mcr_cr ul_orilms
inscrion
ul_orilms
lcucs
1. inscr
o. DM:
M:C
i
M:C
i1
M:C
i2
lcud
Lis
LDF
Lis
o.
!nscrion !ndcx
4. LDF
slcdulin_
5. M:C
2. udmission
onrol
7. inscr
b. M:C mcr_cr
10. :C.N dircion dcision
& scrvic
. u hol dircion
:C.N ul_orilms
:C.N _rou
mcr_cr ul_orilms
inscrion
ul_orilms
lcucs
1. inscr
o. DM:

Figure 2. Execution procedure oI the proposed algorithm
Create SCAN group( MSG ): MSG is created with
requests scheduled into EDF. When a request is included
in MSG , it is deleted Irom the index oI the insertion list
in order to prevent the request Irom being inserted again.
MSG is created successively, and iI two or more MSG s
are created the SCAN group merge technique is applied.
SCAN group merge technique: II two or more MSG s are
created, they are merged, iI possible, using the SCAN
group merge technique.
Q Insertion technique: AIter MSG s have been merged,
appropriate requests Irom the insertion index are inserted
using the insertion technique. They are inserted Irom that
with the highest possibility. Here, because the size oI
MRT oI an executable task is
) 3 (
i
m MS
to the
maximum, iI the number oI requests belonging to MSG
is 3 or less, the insertion technique is not applied. Lastly,
the index oI the inserted request is removed Irom the
insertion index.
DMS: EDF scheduling order is maintained by applying
the DMS technique to the merged and inserted MSG .
Two-way SCAN technique: The direction oI SCAN is
decided Ior consecutive MSGs, the closing time oI which
has been modiIied in the DMS technique.
Q Service: Lastly, MSG is processed in the direction decided
by the two-way SCAN technique, and iI there are still
requests in the EDF list, this procedure is repeated Irom
Q.
140 140 140 140
IV. PREPARE YOUR PAPER BEFORE STYLING
The present research used CSIM18 simulation engine
and MFC 6.0 as simulation tools Ior evaluating the
perIormance oI the proposed algorithm. The experiment
assumes that real-time I/O requests are not cyclic and they
request access to disk blocks at random. In addition, it
assumes that each real-time I/O request demands data oI
8KByte per time. Environment Ior the experiment |13, 14|
was set as in Table 2.
TABLE II. ENVIRONMENT SETTING FOR THE EXPERIMENTS
Parameter Value
Number oI cylinders 6176
Number oI tracks per cylinder 4
Number oI sectors per track 126
Rotation time 8.3 10
-3

Maximum seek time 15 10
-3

Average seek time 8 10
-3

Disk driver transIer speed 4 MByte/sec

One oI important Iactors in evaluating the perIormance
oI a real-time disk scheduling algorithm is the number oI I/O
requests that can be processed per unit time within the limit
oI ending time. First, data transIer time Ior request
transIerring is calculated by equation (6).
ms
MBvets
KBvtes
time transfer data T 2
sec / 4
8
) ( = =

(6)
In order to evaluate disk eIIiciency when the algorithm
proposed in this research was used, we measured the number
oI I/O requests processed within a speciIic length oI time.
Figure 3 shows the number oI requests processed at diIIerent
numbers oI I/O requests.


Figure 3. Numbers oI I/O requests processed within a speciIic length oI
time
In Fig. 3, the number oI I/O requests processed in the
section Irom "A" to "B" was similar in all oI the three
algorithms. This is because all oI the three algorithms can
process input requests properly when the number oI I/O
requests is small. From "B", the number oI requests
processed by EDF is Iixed and increases slightly because the
seek time oI I/O requests are more likely to the maximum
seek time and this lowers disk eIIiciency. On the contrary,
SCAN-EDF or the proposed IT-SCAN(insertion technique-
based and two-way SCAN technique-based real-time disk
scheduling) puts together appropriate requests into SCAN
groups Ior processing, so the number oI I/O requests
processed increases with the increase oI the number oI
incoming requests. The increase rate is higher in IT-SCAN
than in SCAN-EDF because the Iormer can create larger
SCAN groups by merging SCAN groups using the insertion
technique even iI requests are not consecutive. In addition, it
is because disk eIIiciency is enhanced through the two-way
SCAN technique. As shown in Figure 3, when the number oI
incoming I/O requests within a given length oI time is small,
the number oI requests processed is similar among EDF,
SCAN-EDF and IT-SCAN, but with the increase oI I/O
requests, the number oI requests processed increases more
rapidly in IT-SCAN than in EDF and SCAN-EDF. This
suggests that IT-SCAN uses the disk more eIIiciently.
Figure 4 shows the total execution time when real-time
I/O requests are processed using EDF, SCAN-EDF and IT-
SCAN. Here, the total execution time is time to process real-
time I/O requests when the closing time oI real-time I/O
requests is set to be extremely long.


Figure 4. Total execution time
As shown in Fig. 4, the total execution time increases in
proportion to the number oI real-time I/O requests in all the
three algorithms. However, it is shorter in SCAN-EDF than
in EDF because in SCAN-EDF, with the increase oI the
number oI requests, the number oI requests with similar
ending time limit increase and, resultantly, the number oI
requests processed in the SCAN technique increases. The
reason that the total execution time is shorter in IT-SCAN
than in SCAN-EDF is that larger SCAN groups can be
created and the processing direction between SCAN groups
can be decided eIIectively using the two-way SCAN
technique in IT-SCAN. This suggests that IT-SCAN is
suIIiciently eIIective even Ior the disk scheduling oI non-
real-time I/O requests because I/O requests with quite large
ending time limit have the characteristic oI non-real-time
requests.
Fig. 5 and Fig. 6 show the result oI an experiment that
compared the response time oI non-real-time I/O requests
among EDF, SCAN-EDF and IT-SCAN. In IT-SCAN, non-
real-time requests are inserted Iirst to SCAN groups.
Because EDF and SCAN-EDF process real-time I/O requests
Iirst and then process non-real-time I/O requests, the
response time is much longer than that in IT-SCAN, which
141 141 141 141
can insert non-real-time requests. As shown in Figure 6, iI
the number oI real-time I/O requests is increased up to 30,
the diIIerence becomes more remarkable. This result
suggests that the diIIerence in the response time oI non-real-
time I/O request is small when the number oI real-time I/O
requests is small, but iI the number oI real-time I/O requests
increases the diIIerence in the response time oI non-real-time
I/O requests grows bigger.


Figure 5. Comparison oI the response time oI non-real-time I/O requests

Figure 6. Comparison oI the response time oI non-real-time I/O requests
When real-time disk scheduling algorithms based on the
insertion technique and the two-way SCAN technique are
compared with EDF or SCAN-EDF, a matter oI concern is
the complexity oI the algorithms. The proposed algorithm
needs to make additional calculation whenever a new real-
time I/O request comes in or is processed. Considering that
the development oI processors is relatively Iaster than that oI
disks, however, the lowering oI perIormance due to the
complexity oI algorithms is not a matter oI prime importance.
V. CONCLUSION
In order to solve problems in real-time disk scheduling
algorithms applicable to existing hard real-time systems, this
research proposed a new real-time disk scheduling algorithm
based on the insertion technique and the two-way SCAN
technique and evaluated its perIormance by comparing it
with EDF and SCAN-EDF, which are representative hard
real-time disk scheduling algorithms. According to the
results oI experiment, the insertion algorithm and the two-
way SCAN algorithm show much higher perIormance than
EDF and SCAN-EDF in terms oI the number oI I/O requests
that can be processed within unit time. In addition, they are
superior to EDF and SCAN-EDF with regard to the total
execution time oI real-time I/O requests. Not only Ior the
average response time oI real-time I/O requests but also Ior
that oI non-real-time I/O requests, they show exceedingly
high perIormance compared to other algorithms, and the
diIIerence grows even bigger when the number oI requests is
large. It is because the insertion technique creates SCAN
groups bigger than those in SCAN-EDF and the two-way
SCAN technique decides the processing direction between
SCAN groups eIIectively.
REFERENCES
|1| P. J. Denning, 'EIIect oI Scheduling on File Memory Operations, In
Proceedings oI AFIPS SJCC, volume 30, pages 9-21, 1967.
|2| T. J. Teorey, 'Properties oI Disk Scheduling Policies in Multi
programmed Computer Systems, In Proceedings oI AFIPS SJCC,
volume 41, pages 1-11, 1972.
|3| N. C. Wilhelm, 'An Anomaly in Disk Scheduling: a Comparison oI
FCFS and SSTF Seek Scheduling Using an Empirical Model Ior Disk
Accesses, CACM, 19(1), pages 13-17, 1972.
|4| M. HoIri, 'Disk Scheduling: FCFS vs. SSTF Revisited, CACM,
23(11) pages 9-21, 1967.
|5| B. Kao and R. Cheng, " Disk Scheduling," In REAL-TIME
DATABASE SYSTEMS, pages 97-107, Kluwer, 2002.
|6| C. L. Liu and J. W. Layland, 'Scheduling Algorithms Ior
Multiprogramming in a Hard Real-Time Environment, Journal oI
ACM, pages 46-61, 1973.
|7| M. J. Carey, R. Jauhari, and M. Livny, 'Priority in DBMS Resource
Scheduling, In Proceedings oI the 15th VLDB ConI, pages 397-410,
1989.
|8| A. L. N. Reddy and J. Wyllie, 'Disk Scheduling in a Multimedia I/O
System, In Proc. oI the Iirst ACM International ConIerence on
Multimedia, pages 225-233, Anheim, CA, 1993.
|9| S. Chen, J. A. Stankovic, J. F. Kurose and D. Towsley, 'PerIormance
Evaluation oI Two New Disk Scheduling Algorithms Ior Real-Time
Systems, The Journal oI Real-Time Systems, 3(3), 1991.
|10| K. Hwang and H. Shin, 'Real-time disk scheduling based on urgent
group and shortest seek time Iirst, in Proceedings oI 5th
EUROMICRO Workshop on Real-Time Systems, page 124-130,
1993.
|11| H. P. Chang, R. I. Chang, W. K. Shih and R. C. Chang, 'Enlarged-
Maximum- Scannable-Groups Ior Real-Time Disk Scheduling in
Multimedia System, IEEE COMPSAC, pages 383-388, 2000.
|12| R. I. Chang, W. K. Shih and R. C. Chang, 'Deadline-ModiIication-
SCAN with Maximum-Scannable-Groups Ior Multimedia Real-Time
Disk Scheduling, Proc. IEEE RTSS, pages 40-49, 1998.
|13| C. Ruemmler and J. Wilkes, "An introduction to disk drive
modeling," IEEE Computers, pages 16-28, 1994.
|14| R. P. King, "Disk arm movement in anticipation oI Iuture requests,"
ACM Trans. Computer Systems, volume 8, no. 3, pages 214-229,
1990.

142 142 142 142

Вам также может понравиться