Professional Documents
Culture Documents
Information Systems
journal homepage: www.elsevier.com/locate/infosys
a r t i c l e i n f o
abstract
Article history:
Received 4 September 2012
Received in revised form
6 November 2012
Accepted 8 November 2012
Available online 28 November 2012
This paper presents the procedure for comparing costs of leasing IT resources in a
commercial computing cloud against those incurred in using on-premise resources.
The procedure starts with calculating the number of computers as depending on
parameters that describe applications features and execution conditions. By measuring
required execution time for different parameter values, we determined that this
dependence is a second-order polynomial. Polynomial coefcients were calculated by
processing the results of fractional factorial design. On that basis we calculated costs of
computing and storage resources required for the application to run. The same
calculation model can be applied to both a personal user and a cloud provider. The
results will differ because of different hardware exploitation levels and the economy of
scale effects. Such calculation enables cloud providers to determine marginal costs in
their services price, and allows users to calculate costs they would incur by executing
the same application using their own resources.
Leasing in cloud establishes a business relationship: buyer wants to reduce costs, and
cloud provider wants to generate prot. This relationship will be realized if the buyer and
the provider agree on a mutually acceptable fair price that can be determined by the
symmetric mediation plan.
All the steps in this procedure are integrated into CCCE method and represented as a
process model.
& 2012 Elsevier Ltd. All rights reserved.
Keywords:
Cloud computing
Cost effectiveness
Fractional factorial design
Symmetric mediation plan
1. Introduction
Although conceived as a technology a long time ago, it
is over the last few years that cloud computing has
established a more extensive commercial presence [3].
Most researchers [12,18,20], users and professional public
[4,7,8,10] dene commercial computer clouds as a set of
network services intended for providing IT services,
wherein their usage:
0306-4379/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.is.2012.11.002
496
2. Research overview
497
498
Table 1
CCCE method activities and description of the used labels.
Activity number, name and description
1
Select the application
2
Determine TA
3
Estimate r
4
Calculate CP
5
Calculate CS
6
Compare CP and CS
7
Determine b
8
Calculate DNPVD
9
Request the lease price
10
Agree a fair price
499
Table 2
Examined variables for calculation of function TA fT (p,a,u,b,q,r).
Label
(1)
p
a
u
b
q
Independent
variable
description (2)
Unit of measure in
the real domain (3)
Independent variable
variation range in the real
domain (5)
Power of used
computers
Application
complexity
integer
[1,4]
s/transaction
[1,10]
Volume of data to
be processed
Database size
KB
[5.4,540]
MB
[1.2,5.8]
Number of CRUD
actions per query
number/query
[10,200]
Number of used
computers
integer
[1,5]
Substitution for
reduction of the
real domain to
[ 1, 1] (6)
Measurement plan
(7)
x1 23 Up 53
x2 29 Ua 11
9
x3
2
545:4
534:6 Uu 534:6
x4
2
7
4:6 Ub 4:6
x5
2
210
190 Uq 190
x6 12 Ur 32
2
d g
X
dg i dg
500
Table 3
Measured TA according to measurement plan 26 2 and estimated regression coefcients.
Measurement
plan (1)
1
ab
cd
abcd
ace
bce
ade
bde
acf
bcf
adf
bdf
ef
abef
cdef
abcdef
0
24
119
103
198
232
343
134
295
73
124
18
52
149
164
205
256
152
161
156
Coefcient
signicance
(4)
155.6
38.3
91.99
2.1
0.27
4.3
1.17
66.7
278.70
3.9
0.97
1.8
0.21
6.4
2.60
25.4
40.55
19.4
23.68
0.6
0.02
1.9
0.24
3.3
0.69
6.3
2.50
36.2
82.07
0.2
0.00
The mean value is given for all parameters.
Measurement is repeated 3 times, and the results
will be used for the assessment of the adequacy of the mathematical model.
Estimated
effects (5)
B 3 3 5
D 3 5 3
BD AC 4 4
E 5 3 3
BE 4 AF 4
DE 4 4 CF
BDE 3 3 3
F 5 3 3
BF 4 AE 4
DF 4 4 CE
BDF 3 3 3
EF 6 AB
BEF 5 A 3
DEF 5 3 C
BDEF 4 AD BC
From expression (4) it is evident that sample application execution time does not depend on all the six
variables, as it was originally assumed. That time
increases with the relative complexity of application a
and the volume of data sent for processing as well as with
the increase in the number of CRUD operations q, while it
is reduced by the using a larger number of computers r.
501
502
CS
T0
CT
T
1 k
Y TC
DNPV
N
X
C T ET LT
T0
1 k
S
1 k
C
10
11
C T rHT 8760dP C PD dV T eO
12
503
Table 4
Parameters of the model for calculating DNPVD value.
Label (1) Parameter description (2)
d
O
r
g
C
HT
k
K
LT
PC
PD
VT
Price of electricity
[$/kWh]
Initial disk capacity
[GB]
Difference between the costs of using on-premise and leased disks [relative number]
Decrease in the disks market value
[relative number]
Price of a disk control unit
[$]
Annual staff costs
[$]
Interest rate on nancial assets
[$/$]
Price of disk capacity for T 0
[$/GB]
Annual cost of disk capacity lease
[$/GB]
Disk control unit power
[kW]
Power required by a disk unit
[kW]
Required annual additional disk capacity
[GB]
Operator for calculating the number of disks with capacity
O required for sufcing the required capacity V T
456[$]
DNPVD NPVP-NPVL (after 3 years)
0.05
500
0
0.2
500
0.08
0.30
1.8
0.50
0.01
100
0.05
3000
0.5
0.2
1000
50000
0.08
0.30
1.8
0.70
0.01
1000
18023[$]
50198[$]
0.05
5000
0.5
0.2
2000
75000
0.08
0.30
1.8
1.00
0.01
10000
13
CM Mr rp Mu u Md d Mq q Mb b Mrb rb
14
CG Gr r Gu u Gd d Gb b Ge e
15
504
Table 5
Price calculation parameters.
r
p
u
d
b
q
rb
os
e
a
Description
Amazon (A)
Microsoft (M)
{1,2, y}[unit]
Typical congurations []
{1,2, y}[GB]
{1,2, y}[GB]
{1,2, y}[GB m]
{10,200}[action/query]
{1,2, y}[h] or [GB m]
{Linux, Win, y}[]
{1,2, y}[]
Ar 0.085[$/h]
Mr 0.12[ $/h]
p 1 (1.7 GHz; 4 GB RAM) for A, M and G
Au 0.10[$/GB]
Mu 0.10[$/GB]
Ad 0.15[$/GB]
Md 0.15[$/GB]
Ab 0.25[$/GB m]
Mb 0.15[$/GB m]
Aq 0a
Mq 0.01[$/10000]
Arb {0.105, 0.155}[$/h]
Mrb (complex calc.)
Aos {1, 1.41}
Google (G)
Gr 0.10[$/h]
Gu 0.10[$/GB]
Gd 0.12[$/GB]
Gb 0.15[$/GB m]
Ge 0.01[$/100]
Table 6
Data and results of TCO calculation for a 3-year period.
1
2
3
4
5
6
7
8
9
10
11
App1
App2
App3
6
100
90
100000
1000
90
7
10
500
200
500000
5000
120
11
P
Amazon CS,A(3)[$]
P
Microsoft CS,M(3)[$]
P
Google CS,G(3)[$]
Amazon CA(3)[$]
1
10
10
10000
100
20
2
1.7 GHz; 4 GB RAM
12201
5385
5391
4494
6375
42704
18846
18870
15726
28746
67107
29616
29652
24711
79116
Microsoft CM(3)[$]
Google CG(3)[$]
6024
5112
25206
24711
61332
55671
Required features
of virtual applications
Calculated value of
virtual applications
505
Table 7
Symmetric mediation plan for trading IT resources
Vendor
(cloud provider)
Strong (0.5)
Weak (0.5)
CS 0.12[$/h]
CS 0.08[$/h]
Buyer
(cloud computing user)
Strong (0.5)
CP 0.10[$/h]
0, no transaction
q, 0.08 y[$/h]
Weak (0.5)
CP 0.14[$/h]
q, 0.14y[$/h]
1, 0.11[$/h]
506
monthly data uploadbytes
0:5U1012 bytes
200 kB
Table 8
ISVU computing resources (r).
Computers
(processor & OS)
Type (physical F/
virtual V)
Amount
Number of CPU
cores (r)
Purpose
SPARC/Solaris
x86/Linux
F
V
1
9
4
1
10%
40%
x86/Windows
40%
Informix DB server
Apache web server
ISVU app.
Active Directory
MS SQL
1.
2.
3.
4.
507
$
$
C ISVU 8760 h 9U0:34
1U0:46
h
h
p
$
$
1 20:568
1U1:315
h
h
$
GB
1U500
12 month 0:12
GB
month
GB
GB
1U500
9000
month
month
$
50000GB
12 month 0:011
GBmonth
C$68,330.28
Total yearly cost of ISVU-like system run on Amazon
would be (as calculated from CCCE) about $68,330.28.
Real yearly cost of ISVU with included value-added tax of
25% is about $96,000, so without VAT it would be about
$77,000. It is somewhat more expensive than running in
large commercial cloud would be, but there are additional
security and legal benets in having it physically under
domestic jurisdiction. These aspects are not covered by
CCCE and it is up to every user to analyse them for a given
type of application.
Capacity [GB]
6. Conclusion
105
25
150
50000
508