Performance and Cost Analysis of Modern Public Cloud Services

adfa, p. 1, 2011.
© Springer-VerlagBerlin Heidelberg2011
Performance and Cost Analysis of Modern Public Cloud Services.
Md. Saiedur Rahaman
Department of Information Technology, University of Stuttgart,
Stuttgart, Germany
rsaied (at) yahoo.com, rsaied876 (at) gmail.com
Supervisor: Vasilios Andrikopoulos
Abstract. Cloud computing or cloud servicing is a practical use of remote server by remote host via
internet where they store data, process data and manage data instead of personal computer or offline
compute. Architecture of cloud computing service has some challenge for significant difference from
traditional computer system. In personal computer or personal server we can use only when we are in
our local place but cannot use out of area. To setup several personal server it is more complex and cost
effective for maintenance its hardware and software. Many company found only for providing remote
services. For cloud service we consider mainly two issues. (1) How much cost for service. (2) How the
performance is. For these reason clients always wants best service with cheap cost. Public cloud
service provider spent a lot of time, money; efforts to build a high infrastructure for give best service
for their clients. According to clients requirements sometimes service provider cannot provide proper
service. Between service provider and client there is a Service Level Agreement (SLA). NICTA has
developed a service oriented performance modeling system for performance model scalability of
service oriented application platform for different requirements. At present many public service
provider such as Amazon EC2, Google App Engine, Microsoft Azure. My task target is to represent a
relationship between performance and cost of public cloud service.
Keywords:
Cloud computing, comparison, performance, cost.

Introduction:
Problem: The traditional computing system (In-house computing, LAN, Other
networking) is familiar to all for many decades. This traditional system has some
limitations as like one cannot access his data from outsides of his network or his computer.
If someone needs to access his data from other computer he needs an external storage
device. It is need extra cost and waste of time. If someone wants to transfer huge data it is
also need external storage and many procedures. Sometimes for a single time for a
special operation need special device (which may not need in future) that is cost effective.
Why cloud computing: The concept of cloud computing has introduce within a few years.
Today message from different organizations around the world is clear and loud. Publicly
available service such as Platform as a Service (PaaS), Software as a Service (SaaS), and
Infrastructure as a Service (IaaS) is main stream, essential part of IT armory. They try it,
they like it, they have more demand. Every company has a personal cause why they use
cloud computing service. The common purpose to use the cloud service, it enables end user
to manage the process and store data efficiently at very high speeds at reasonable cost.
Reduce cost: cloud computing enables to the organization to pay only they used service
from public cloud service. User no needs to think about infrastructure for run application.
Just login VMs and all other things are managed by service provider according to SLA, no
need to manage huge employee, save office space, save power consumption, no need to
maintenance server and other infrastructure as well. Organizations are not worried
about demand of future; even they are growing day by day.
Greater business agility (speedup): Agility is about responsibility. Cloud computing
gives you rapid action for business opportunities and challenge. If suddenly business area
is becoming very large it may be a problem for organizations to tackle. They have to think
about scaling infrastructure by requisition, justification, purchase, deployment, testing,
retesting, and then operation. But under cloud service it is easy to manage.
Flexibility: with flexibility you have a choice to introduce new VM or cancel existing VM
according to organization demand. Without cloud computing it is not possible to reduce or
increase instance instantly.

Motivation:
Cloud Computing Framework: Over all structure of cloud architecture [3]. The parts of
cloud architecture include. User, Application, Cloud Broker, Monitoring, Service Category
Performance measurement of cloud computing:
Service Response time: The efficiency of a service can be measured by response time. For example
how faster can response on request of a client. Service provider take time to response depends on
variety of factors such as turn on VM, task initialization, determine application and start application.
This time can divided in sub factor
Figure 2: Response time of different task in different cloud Providers [21]
Vertical axis represents time in millisecond.
Average response time: Average response time is determined by ∑ Ti /n. Where Ti is the time
between request and response and n is number of requests for IaaS service.
Maximum response time: Maximum response time is that which is assured by service provider
to clients for get service.
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Get Put Query
Cloud 1
Cloud 2
Cloud 3
Applications
Business,Banking, Social Network, Personal Data, University, Research center
Users
User 1, User 2, User 3,………………………………………….User n
Cloud Broker
SLA management SMI Calculator Ranking System
Monitoring
Qualitative Measure Quantitative Measure Service Filter
Service Category
Cloud provider 1, Cloud Provider 2,…… ,……………… Cloud Provider N

Failure response time: Failure response time means when service provider cannot keep their
SLAs promise. That is response time is more than promised.
Sustainability: Sustainability is the matter of environmental impact of the cloud service used.
For instance, average carbon footprint, or energy efficiency of the cloud service. Carbon
footprint may complex and depends on many facts. We can calculate carbon footprint by carbon
calculator such as Power Usage Efficiency (PUE) [4]. This also include IT equipment efficiency
and IT equipment utilization
Suitability: Suitability is the customer requirement. When customer wants to find some service
providers, he tries justifying the quality of provider by filtering. If customer find some service
provider according to his demand it is suitable for customer. If not matching the requirements it
is not suitable.
Accuracy: Accuracy of the service functionality measured the degree of proximity to the users
actual values when using a service compared to the expected values. For computational
resources such as Virtual Machines, accuracy`s first indicator is the number of times of the cloud
provider deviated from a promised SLA.
Adaptability: This is the ability to service provider to adjust changes of customer demand. It is
defined by time to taken to update the old system into new system (e.g. upgrading from small
Amazon VM to medium VM [5])
Elasticity: Elasticity is defined by how quickly cloud service provider can expand service area
during peak hour. This attribute can explain with two terms, maximum time take to expand
service capacity and maximum capacity of service.
Figure3: Show elastic capacity of cloud providers [23]
Vertical axis represents numbers of instances of cloud providers.
We can justify from figure that cloud 2 keeps 10 instances reserved for clients to serve in failure
of service and for new demand. Other hand cloud 3 keeps only two instances reserved.
Obviously cloud 2 will get prefer to select if we consider elastic performance.
Throughput: Throughput is an important factor to measure the performance of cloud service.
Throughput is the total number of task complete in a single period of time. For example n task
perform by VM, and completion time is Tc, T0 is all overhead. We can write throughput of
cloud service. α = n/ Tc + T0.
Scaling Latency: Scaling latency is measurement of time to taken by cloud system for executing
a given task. We can compare scaling latency of various system providers. Latency is important
to minimize the cost. We can consider the provider with different operating system Linux and
Windows. We can consider a particular task, (e.g. Booting latency).
0
10
20
30
40
50
60
Cloud 1 Cloud 2 Cloud 3
Used Instances
Reserted Instances
Total Intances

Figure 1: Scaling latency for cloud with different OS [23], Vertical axis is time in MS
We can see for booting instance and execution Linux is faster than Windows. So client always
try to select Linux for faster operation or faster execution.
Internet Bandwidth: For cloud service most common and primary requirement is internet. For
starting SaaS, Iaas, PaaS Internet speed is first priority. When client select a cloud service
provider, client must see the performance overview service provider internet bandwidth.
Bandwidth speed is different in uplink and downlink. Clients who frequently need to download
more data on every day, he always looking for more downloading speed. Other hand the clients
who just need to upload data every day, he looks more uploading speed.
Cloud name Upload Download Time
C1 128Mbps 512 Mbps Off
128 Mbps 256 Mbps Peak
C2 256Mbps 512 Mbps Off
128 Mbps 512 Mbps Peak
C3 256Mbps 256 Mbps off
256Mbps 256 Mbps Peak
Table 1: Upload and download speed of different cloud provide [24]
Selecting providers: According to customer review and performance overview we can chose
popular cloud service providers. Amazon EC2 and Rackspace cloud server, because it has large
number of web service [6]. Microsoft Azure is one nice cloud service reason is that it is phase
in cloud computing market that offers full spectrum of computation and storage services [7]. We
chose Google AppEngine. It is a unique PaaS provider [8].
Identifying common services: Despite of the complexity of various service providers there is a
common set of functionality.
Elastic compute cluster: This cluster includes a set of virtual instances that run application
code.
Persistent storage: Storage service provide service of keeping data store and keep a state of data
and application that can be access by application instances by API.
Intra cloud network: Intra network setup a network among application instances.
Wide area network: By the help of wide area network, the system make total network among
application program, instance and end users.
0
0.5
1
1.5
2
2.5
Linux Task1 Windows Task1 Linux Task2 Windows Task2
Execution latency Booting Latency Total Latency

Server performance comparison: Majority of cloud platform use virtualization to make sure
that applications are isolated from each other to make sure a minimum dedicated allocation of
resources to each virtual core. Consider EC2, Azure and GAE. Among these three cloud EC2
and Azure use virtualization. VM on the same server compete for other resources as like
memory and bandwidth. This model is good in performance. Other hand in house platform use
scheduler for different request where performance overhead is costly.
LAN latency and bandwidth: In cloud architecture all application deployed in VM instance.
This is feasible for clients in all situations. Clients no need to think about hardware, just tension
free execution, no more latency. But in house platform there are many network controller,
switches, routers and cabling matters which obviously cause of performance overhead. When
work station are not in same region that time many packet loss and typical inter instance latency
is between 1 MS to 1s. This latency may increase on time due to LAN age. For faster transfer
needs to kill some process.
Scalability: Different cloud infrastructure has different performance. We can say processor
speed of Intel processor and AMD processor has different performance. How many CPU is in a
core it is also a performance measurement it can be represent with a table.
Fig: maximum instances of EC2 [23], Vertical line represents number of instances.
According to a latest report from Synergy Research, AWS has hit the five years high market
share despite increasing competition from Microsoft, Google, and IBM. The research that
includes SaaS, IaaS and PaaS, AWS has achieved 28% of total cloud market, followed by
Microsoft (10%), IBM (7%), Google (5%), Salesforce (4%) and Rackspace (3%) [27]
Discussion on Cost
The term cost is related to pay for service used from service provider. The organization who wants to
use some specific service from cloud provider, it is important to know about the costing model of the
0
50
100
150
200
Small core cpu Medium core
cpu
Large core cpu Extra large core
Intel processor
AMD processor

service. If organization has some information on how to pay for service as like monthly, hourly, daily
it will be helpful for client. Sometime clients use some services and/or resources to solve some
problem of organization. I will try to discuss some issue to metering and billing or costing for cloud
service provider of upcoming technology.
Billing policy for cloud service provider: every cloud providers has its own cost model. Each
provide calculate cost or determine cost for service depending on their infrastructure cost,
application cost, maintenance cost, employee cost, accommodation cost, investment cost and many
factors as well. But all cloud service providers try to keep low cost for customers as they can.
Costing is a critical factor for organizations offering service of products [10]. Therefore determine
appropriate cost model will help to achieve high profit. To determine cost of service and product must
be considered manufacturing and maintenance cost. Service providers use many costing
techniques. Such as typical costing technique is “pay for limited usage”. This is rigid approach and
does not consider many other factors which affect costing, e.g. the age of resource and fairness of cost
[11]. Other popular cloud provider uses such AWS, GAE apply “pay per use fixed costing”. It
charges to user according to total consumption of resources. “Pay per resource” is another process of
costing that applies for charge of a particular resource as like storage or bandwidth. Subscription is
another costing technique that use where customer subscribes a certain service provider for a fixed cost
per unit for long time. Also dynamic costing is one which dynamically changes cost. All methods
have an agreement with both parties where include SLA.
The following factors are most important that impact of cloud costing for client [12].
Initial cost: initial cost is the money that spends service provider to setup all components and
resources for particular purpose.
Lease period: This is the time period the customer will lease resource from service provider.
Provider offer different options for customers. Provider always tries to get more clients in less cost.
Normally provider offers low cost for longer periods.
QoS: This option defines the service quality. QoS is a set of techniques and technologies that offer by
cloud providers to the clients to enhance the customer experience as like as data security, data privacy,
service availability, and others quality factors. Sometime quality factor define cost factor for service.
Age of resources: Normally life time of resources has a limit of use. If the age of resources deployed
by the service provider is older, the cost of service may be low. Cause of this is the resources
performance will reduce on time.
Cost of maintenance: Maintenance cost is another factor to define cost of service. Provider has to
perform best service, so they need to maintenance all resource timely which bears large cost of an
organization. Service providers calculate this cost in annually.
Data security and risk: customers keep their important data in the cloud. So, data security and risk of
data lost is a factor. If some provider will lost the data of client and third party can see the data it will
be punishment for provider. For this purpose provider and consumer make an agreement by a written
form. Above factors for service providers, but customer can have some questions. Is costing fair? Is it
right cost for this service? Same cost for all providers? An influence of costing is supply and demand.
Demand means what is want by customer for certain cost. According to demand law when the cost is
increase of a goods or service, demand will reduces. Other hands if supply is increase cost will reduce.
So demand and supply is a nonlinear relationship on costing. Customers can compare of cost explicit
or implicit for service. Explicit cost comparison means that the compare to another cost or a range of
cost (e.g. I paid more than another customer for same resource). Here customer to customer varies the
cost. Implicit comparison is that the service fixed cost is not perfect. Says I paid more than I use [13].

Aspects costing model: Client may evaluates a prospective service provider based on three main
parameters
1. Costing approach: Costing approach is the method of determine cost of service. Costing approach
can be [14].
1. Fixed cost regardless volume, 2 Fixed cost regardless volume, 3 Fixed cost plus per unit rate,
4 Assured purchase volume plus per unit cost rate (pay for certain quantity, if exceed need to pay more
extra utilization),5 Per unit rate with ceiling (customer pay for certain limit of use), 6 Per unit cost
(customer pay different cost per unit).
2. Quality of Service: QoS describes the requirements for the service providers what service should
provide to a customer. If QoS is ensure by the provider, it will increase the number of customer and
loyalty of the service offered.
3. Utilization period: Utilization periods reflect that a customer how long time can be used the service
according to SLAs between customer and service provider.
Aspect of cloud costing can be represent by a figure [15]
Costing model for cloud computing: Different service provider offer different schemes and model
for costing. Here I will show some market leading cloud computing organizations costing scheme.
Amazon considers the market leader in cloud computing [16]. Amazon uses a fixed cost model for
each hour of VM usage. GAE and Windows Azure is also leader in cloud market use “pay as you go”
and “pay for resources” cost model. Subscriptions model where customer pays in advance for the
services he is going to receive for a predefine period of time. The lower boundary of cost determine by
initial cost of cloud provider which should charge to customer. But upper boundary determine by
proposed compounded Moore’s law. The law presented by authors combines Moore’s law [17] with
the compounded formula. The author says that if the cost set between these two boundaries, it would
be beneficial for both cloud provider and customer. It is nice procedure, but it did not take into
consideration of maintenance cost. Here says initial cost for client and provider will be same. But
provider will get discount for buy large amount of assets. Also we need to consider the cost of
electricity consumptions, employments, marketing, time investment and accommodations etc.
Wang et al proposed an algorithmic [18] solution for optimized data center net profit with deadline
dependent scheduling by jointly maximizes revenue and electricity cost. They developed two
algorithms. (1) Net Profit Optimization for Divisible Job (NPOD). Divisible Jobs means the job can
interrupt. (2) Net Profit Optimization for Indivisible Jobs (NPOI). Indivisible job means the job cannot

be interrupted. By an experiment it is clear that, if largest job scheduled first it become optimized for
this algorithm. Here consider only static job arrivals and departures.
Macias and Guitart [19] proposed a genetic model of cost in cloud computing. To choose a good
model of cost via their algorithm follows three steps. Define a chromosome (type), evaluate it and
finally select a best pairs of chromosome for reproduction and discarding those with worst results.
The result of simulation illustrated that generic costing gain highest revenue in most scenarios.
Service provider deploys generic algorithm gained revenue up to 100% greater than other dynamic
cost strategies and 1000% greater than fixed cost strategies. The proposed generic model is easy to
implements, flexible, easy to adapt to a set of various parameters which influences costing.
Mihailescu and Teo [20] introduce a dynamic model for cost of federated cloud service that shared
resources among different cloud services.it uses for increasing reliability and scalability for clients and
service providers. Federated cloud users are able to buy and sell their resources. In case of high market
demand, fixed cost would minimize seller profit because cannot charge more than fixed charge.
Similarly when demand is low client utility would be minimize because he cannot reduce the cost as
like as present market value. Therefore dynamic costing is best in such situation, because dynamic
costing will be set according to levels of market supply and demand. By a simulation it has found that
dynamic costing is achieved 200% better performance comparing to fixed cost model.
With the help of a table [15] I will represent cost comparison models in terms of fairness, pros and
cons. Costing models in main two types static and dynamic. Cost is unchanged in static model but cost
is changing on different factors.
Pay as you go
Costing
approach set by
fairness pros cons
Cost is set by the
service provider
and cost remain
unchanged
Unfair for client,
he might pay for
more time than
need
Clients is aware
of the exact cost
to be paid
And resources are
reserved for
clients for the
time periods he
paid
1.service provider may reserve
the resources for longer time
than clients utilized
2. Service provider cannot raise
cost when demand is high; and
client pays high when demand is
low than market value.
subscription
Costing based on
the period of
subscription(stati
c)
Clients may
sometime overpay
or underpay
Clients may
underpay for the
resources when he
use extremely
Clients may overpay for
resources when do not more use.
A novel financial economic model
Usage based
on dynamic
Fair for both
provider and
client, because
cost is set
between upper
and lower
boundaries
Help the service
provider recover
its initial costs
and provides a
high level of QoS
for client.
Does not consider the
maintenance costs and assumes
the cost may charge for asset of
the clients
Pay for resource model
Cost based
(static)
Fair for both
provider and
client
Offers maximum
utilization of the
service provider’s
Hard to implement

resources
Costing algorithm for cloud computing resources
Real time costing
(dynamic)
Fair for provider
because it reduce
cost and
maximized
revenue
reduce cost and
maximized
revenue
Model is almost fixed and
cannot adapt to rapid change
between supply and demand of
market.
Dynamic resource costing on federated cost
Action based
costing
(dynamic)
Fair for both
client and
provider, because
value set
according to
levels of supply
and demand
Better than
average
performance with
increasing buyer
welfare and
number of
successful request
up to 200%
Less scalability of high demand
in the market than fixed costing
Generic model for costing in the cloud computing model
Real time
costing(dynamic)
Biased toward the
serviced provider,
algorithm
considers increase
revenue.
Achieve very high
revenue, stable
against noise,
flexible, easy to
implement
In very high and very low
demand it is in under
performance
Data center net profit optimization with individual job deadlines
Based on job
scheduling
(dynamic)
Biased toward the
serviced provider,
algorithm
increase revenue
reduce cost
Maximize
revenue and
minimized
electricity cost
Consider only static job arrival and
departure , hard to implement
Value based costing
Cost set
according to
value perceived
by the
client(dynamic)
Fair
to
provider
High revenue on
each service
Difficult to obtain and interpret
data from clients competitors to
evaluate client perceived value
Based initial cost
Cost set by
adding a profit
element on top of
the cost
(dynamic)
Not fair
to client
Simple to
calculate cost
Try to ignore client opinion
Competition based costing
Cost set
according to
Fair to client Easy
to implement
Does not take consider into

competitors
offers(dynamic)
client
Client based costing
Cost set
according
to client wish to
pay
(dynamic)
Fair to client Client taken into
account
Client rarely indicate to seller
what they willing to pay
Hybrid costing
Value of service
changed
according to the
job queue
waiting times
(static/
dynamic)
Fair to client, it
automatically
adjust cost within
static limit
Simple and low
computational
overhead
Need to reach an agreement
Fundamental component (storage, computation, data transfer) cost depends on
Computation:
1. Pay for computation time by hour
2. Multiple combinations of OS, RAM, local storage.
3. On demand, dedicated, Reserved and spot instances
4. Volume discounts
Storage:
1. Pay by GB
2. Cheaper storage at lower redundancy
3. Pay for performance option for block storage
4. Tired costing
Data Transfer:
1. Pay for GB of data out, no charge for inbound data.
2. Minimal charge for data transfer within the same region
3. Bandwidth agreement across all service
4. Tired costing
Relationship between service provider and client cost:
Infrastructure cost and Infrastructure cost and Quality of service and
Consumer cost is linear quality of service is linear. Consumer rice is linear
N N
N
0 N 0 N 0 N
State of art:
Service measurement index (SMI): SMI attributes designed based on International Standard
Organization (ISO) standard by the CSMIC (cloud service measurement initiative consortium)

consortium [2]. It is a set of business related Key performance Indicators (KPI) which provides
standard method for measuring and comparing business service. SMI provides a holistic view of QoS
that needed by the clients to selecting a cloud service provider according to Accountability, Agility,
assurance of service, cost, performance, security and privacy, usability. These attributes are
described briefly here [25]
Accountability: This attributes is used to measure the characteristic of a cloud service provider.
Because this attribute is most important for gain clients faith. No organization will agree without
account of their stored data. Client store their secure data for more security reason. If service
provider cannot ensure the proper accountability it may great loss for clients as well as service
provider. Accountability is critical when measuring and scoring services, include auditability,
compliance, data ownership, sustainability, provider ethicality which determine by SMI.
Agility: Organization can easily and quickly change without expenditure. Agility is rate of change
matric, which show how rapidly new capabilities are integrated into IT as needed by the business.
When considering a cloud service’s agility, organization wants to know whether service is flexible,
adaptable, portable and elastic.
Cost: Cost is the first priority to the organization whether it goes to cloud computing or not, the cost
will effective or not. For this reason, cost is vital attribute of organization when using cloud service
whether it use private or public cloud. Costing can be single and most quantifiable matric today. It is
also important for service provider to clarify the cost of all services.
Performance: There are several solutions are offered by service provider for same problem
addressing to IT needs of different clients. Every solution has different performance in term of
functionality, accuracy, service response time. Clients need to understand how their application will
perform in different cloud platform and where there demand will be fulfilled or not.
Assurance: This attribute is about that service provider are giving proper service as SLA? Every
organization has a tendency to increase their business area or makes more customers. According to
promised service agreement client try to get best output and service provider also try to keep satisfy
the customer with reliability, and stable service.
Security and privacy: Data protection and privacy are most important concern to nearly all
organization. Keeping data under other organization is always critical issues which required strictly
security policy applied by cloud service provider. Every client wants to ensure that data consistency
and data integrity. Security and privacy is multidimensional in nature and include many attributes such
as protecting confidentiality and privacy, data integrity and availability.
Usability: For rapid access of data kept in cloud is plays an important role to spread-out of popularity
of cloud computing. It is easy to access and can access from anywhere from internet and it is faster.
Usability of cloud service can depend on many factors such as accessibility; install ability,
learnability and operability.
Elastic compute cluster: The cluster includes a variable number of virtual instances that run
application code.
Persistent storage: The storage service keeps the state and data of an application and can be accessed
by application instances through API calls
Wide area network: The content of an application is delivered to end users through the wide-area
network from multiple data centers (DCs) at different geographical locations [23].
Comparison & Discussion:
SMI Cloud Approach: which helps Cloud customers to find the most suitable cloud provider and
therefore can initiate SLAs? The SMI Cloud framework provides features such as service selection
based on QoS requirements and ranking of services based on previous user Experience and
performance of services. It is a decision making tool, designed to provide assessment of Cloud services
in terms Of KPI. [3]
Performance analysis of public cloud: Measurement parameters, Service response time,
Sustainability, Accuracy, Adaptability, Elasticity, Throughput, Scaling latency, Internet bandwidth.

Cloud CMP Compare Approach: Compared to traditional computing model that uses in-house
infrastructure, cloud computing offers advantages of cost and reliability like no need infrastructure and
maintenance cost [23].
Goals and approaches: Guided a client to select a cloud provider, Relevant to the cloud provider
service, Fair, Thoroughness vs. measurement cost, Coverage vs. development cost.
To move or not to move Approach: Cloud computing give some advantages as easy of management,
infrastructure saving, salary saves. Although many advantages some client will not get advantages
from cloud providers if they have already infrastructure and administrative base [26].
Identify set of cost factor: Net present value, Workload intensity and growth, Data transfer, Storage
capacity, Software license, Workload variance and cloud elasticity, Cost components, Application
hosting choice.
Performance & cost assessment Approach: NICTA has developed service oriented performance
modeling (SOPM), a technology of service oriented architecture (SOA) application for performance
and scalability. SOA has a sub task called cloud architecture. SOPM are parameterized by workload,
performance [24].
Resource and cost: In house hosting, bare hardware, In-house hosting, virtualization, Server
performance variability, LAN latency and bandwidth, WAN latency and bandwidth.
Performance analysis of cloud computing Approach: There are several solutions are offered by
service provider for same problem addressing to IT needs of different clients. All solution has
different performance in term of functionality, accuracy, service response time. Clients need to
understand how their application will perform in different cloud platform and where demand will be
fulfilled [23].
Cloud computing services for scientific computing: (Job structure & source, Bottleneck resources,
Job parallelism)
Four SelectedClouds: (GAE, AWS, Azure, Rackspace)
MTC presence in scientific computing workloads
Cloud performance evaluation: (Multi machine benchmark, single machine benchmark, resource
book & release, performance stability)
Clouds vs. others scientific computing infrastructure : (Performance vs. cost, security vs. cost).
Comparison of approaches:

Comparison of different cloud providers:
Platform
Elastic
cluster
Storage Wide Area
network
Others
Amazon
AWS
Xen VM SimpleDB(table),
S3(blob),
SQS(queue)
3 data center
(2USA,1Europ)
Flexible, proper
load balance,
more functions
Microsoft
Azure
Azure
VM
Xstore
(Table,blob,queue)
6 data center
(2-USA,
2-Europ,2 Asia)
Running only
MS service &
technology
Google
AppEngine
Proprietary
Sandbox
DataStore
(table)
Unpublished
Number of Google
data center
Better network,
secure, faster
Rackspace
CloudServer
Xen VM CloudFiles
(blob)
2 data center(USA) Better
ferformance,
cost,security
Discussion:
Performance: Many service providers are acting. But there is no unique rules and prerequisite for
them. For this reason they have lack of proper services on Security (Authentication, Authorization,
Access control), Privacy, Workload, Processor power, Latency (waiting time), Scalability, Buffer
overflow (Request reject), Storage, Availability.

Cost: Due to lack of standard for costing model a service provider gain more revenue. But they may
not provide proper service according to SLAs. Other hand clients are not getting proper service but
paying more comparing the service.
In my work I have reviewed key concept cost and performance of public cloud. The cost model is not
feasible for clients but provider. Costing model should be regarding to end user. Provider should be
aware to give more importance to clients in all aspects (security, cost etc.). Provider need to be more
concern to their QoS. Provider need to reduce the risk of cloud computing which is harmful for user.
Client need to choose the provider according to over view of provider service. In the event of provide
outage, it is necessary to ensure data availability. There need to have a unique cost and performance
metrics for public cloud.
Conclusion: In my work I try to review the cloud computing costing key concepts and attributes and
provide a thorough background of costing for cloud provider as well as client. I have included the
comparison between various costing models in cloud market. Many costing model is implement in
cloud market but all models are not beneficial for client. Most model focus providers benefit. Most of
them are used for gain more revenue of providers and reduce of cost. A good costing model should
consider of the end user such as user satisfaction, QoS, end user utility and many more. If a customer
is satisfied with service provider he always uses the service thus the provider can gain more revenue.
The client choose cloud provider which is more convenient for the client. Many modeling approach of
the cloud platform give good visibility into operational cost of application and can show various
resource types. Different cloud service model and costing models allows to choice the hosting options
among them. This add both flexible and complexity. Modeling is potentially powerful tool to
understand and compare performance, scalability, cost, benefits, risk of various platforms, hosting
and deployment options. These models of service and costing are not enough. For more betterment of
cloud service need more analysis and research. If elastic infrastructure able to scale rapidly to load
spikes, it will more beneficial for end user. For future generation of cloud computing we need
enhance modeling of service performance and simple, effective costing technique.
References:
1. I used following references.
2. State of the Cloud, May 2010. http://www.jackofallclouds.com/2010/05/state-of-thecloud-may-
2010
3. Microsoft Windows Azure. http://www.microsoft.com/windowsazure.
4. Google AppEngine. http://code.google.com/appengine.
5. Amazon EC2, http://aws.amazon.com/ec2/
6. S. Dutta, M. Zbaracki and M. Bergen, “Pricing Process as a Capability: A Resource-Based
Perspective”,Strategic Management Journal, vol. 27, no. 7, (2003).
7. S. Maxwell, “The Price is Wrong: Understanding What Makes a Price Seem Fair and the True Cost
of Unfair Pricing”, Wiley, (2008).
8. B. Sharma, R. K. Thulasiram, P. Thulasiraman, S. K. Garg and R. Buyya, “Pricing Cloud Compute
Commodities: A Novel Financial Economic Model”, Proc. of IEEE/ACM Int. Symp. on Cluster,
Cloud and Grid Computing, (2012).
9. L. Xia, K. B. Monroe and J. L. Cox, “The Price Is Unfair! A Conceptual Framework of Price
Fairness Perceptions”,J. of Marketing, vol. 68, (2004).
10. E. Iveroth, A. Westelius, C. Petri, N. Olve, M. Coster and F. Nilsson, “How to Differentiate by
Price: Proposal for a Five-Dimensional Model”, European Management Journal, (2012).
11. International Journal of Grid and Distributed Computing Vol.6, No.5 (2013),
http://dx.doi.org/10.14257/ijgdc.2013.6.5.09
12. Amazon Web Services, http://aws.amazon.com/.
13. G. Moore, “Cramming More Components onto Integrated Circuits”, Electronics, vol. 38, no. 8,
(1965).
14. W. Wang, P. Zhang, T. Lan and V. Aggarwal, “Datacenter Net Profit Optimization with Individual
Job Deadlines”, Proc. Conference on Inform. Sciences and Systems, (2012).

15. M. Macias and J. Guitart, “A Genetic Model for Pricing in Cloud Computing Markets”,Proc. 26th
Symp. of Applied Computing, (2011).
16. M. Mihailescu and Y. M. Teo, “Dynamic Resource Pricing on Federated Clouds”, Proc. 10th
IEEE/ACM Int. Symp. on Cluster. Cloud and Grid Computing, (2010).
17. A framework for ranking of cloud computing services, journal homepage:
www.elsevier.com/locate/fgcs
18. On a Catalogue of Metrics for Evaluating Commercial Cloud Services/ 2012 ACM/IEEE 13th
International Conference on Grid Computing
19. Amazon EC2 Cost Comparison Calculator
20. Cloud Computing Pricing Models A Survey
21. Cost breakdown of Public Cloud Computing and Pricing Strategy for Cloud Computing Services
22. CloudCmp Comparing Public Cloud Providers
23. Performance Analysis of Cloud Computing
24. Performance and Cost Assessment of Cloud Services
25. SMICloud A Framework for Comparing and Ranking Cloud Services
26. To Move or Not to Move The Economics of Cloud Computing
27. http://dazeinfo.com

Performance and Cost Analysis of Modern Public Cloud Services

More Related Content

What's hot (20)

Viewers also liked (16)

Similar to Performance and Cost Analysis of Modern Public Cloud Services (20)

Performance and Cost Analysis of Modern Public Cloud Services