Cloud Computing - Semantic Scholar

68 International Journal of Cloud Applications and Computing, 2(3), 68-85, July-September 2012

Cloud Computing:

Locally Sub-Clouds instead of Globally One Cloud Nawsher Khan, Universiti Malaysia Pahang, Malaysia Noraziah Ahmad, University Ahmad Dahlan, Indonesia Tutut Herawan, University Ahmad Dahlan, Indonesia Zakira Inayat, University of Engineering & Technology, Peshawar, Pakistan

ABSTRACT Efficiency (in term of time consumption) and effectiveness in resources utilization are the desired quality attributes in cloud services provision. The main purpose of which is to execute jobs optimally, i.e., with minimum average waiting, turnaround and response time by using effective scheduling technique. Replication provides improved availability and scalability; decreases bandwidth use and increases fault tolerance. To speed up access, file can be replicated so a user can access a nearby replica. This paper proposes architecture to convert Globally One Cloud to Locally Many Clouds. By combining replication and scheduling, this architecture improves efficiency and easy accessibility. In the case of failure of one sub cloud or one cloud service, clients can start using another cloud under “failover” techniques. As a result, no one cloud service will go down. Keywords:

Cloud Computing, Effectiveness, Efficiency, Replication, Scheduling, Scheduling Techniques, Sub-Cloud

INTRODUCTION Efficiency, scalability and easy accessibility are the key factors and should be the key features of cloud computing. From end-user computing, data storage and data transferring requirements are growing. Users demand for more capacity, more reliability and the capability to access information from anywhere in the world. Cloud services (computing, storage and transferring) meet this demand by providing transparent,

DOI: 10.4018/ijcac.2012070103

easy and reliable solutions. Since late 2007 the concept of cloud computing was proposed (Weiss, 2007) and it has been utilized in many areas with some success (Brantner, Florescuy, Graf, Kossmann, & Kraska, 2008; Moretti, Bulosan, Thain, & Flynn, 2008). Cloud computing is deemed as the next generation of IT platforms that can deliver computing as a kind of utility (Buyya, Yeo, Venugopal, Broberg, & Brandic, 2009). Foster, Yong, Raicu, and Lu (2008) made a comprehensive comparison of grid computing and cloud computing. By a cloud, we mean an infrastructure that provides resources and/or services over

Copyright © 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

International Journal of Cloud Applications and Computing, 2(3), 68-85, July-September 2012 69

Figure 1. Simple architecture of cloud computing

the Internet. A storage cloud provides storage services (block or file based services); a data cloud provides data management services (record-based, column-based or object-based services); and a compute cloud provides computational services. Often these are layered (compute services over data services over storage service) to create a stack of cloud services that serves as a computing platform for developing cloud-based applications. Examples include Google’s Google File System (GFS), BigTable and MapReduce infrastructure (Dean & Ghemawat, 2004; Ghemawat, Gobioff, & Leung, 2003). Amazon’s S3 storage cloud, SimpleDB data cloud, EC2 compute cloud (Amazon, 2009); and the open source Hadoop system (Borthakur, 2007; Dean & Ghemawat, 2008). Figure 1 shows the simple architecture of cloud computing. For the majority of applications, databases are the preferred infrastructure for managing and archiving data sets, but as the size of the

data set begins to grow larger than a few hundred terabytes, current databases become less competitive with more specialized solutions, such as the storage services (e.g., Borthakur, 2007; Dean & Ghemawat, 2008) that are parts of data clouds. For example, Google’s GFS manages Petabytes of data (Hbase Development Team, 2009). Cloud architectures are middleware services for different purposes i.e., resource allocation management, job scheduling, security, authorization and data management etc. When a user requests a file, a large amount of bandwidth could be spent to send the file from the server to the client and the delay or response time involved could be high (Bsoul, Al-Khasawneh, EddienAbdallah, & Kilani, 2011; Ben Charrada, Ounelli, & Chettaoui, 2010a). Besides that, maintaining local copies of data on each accessing site are cost prohibitive while storing all data in a centralized manner is impractical due to remote access latency



(Shorfuzzaman, Graham, & Eskicioglu, 2010; Shorfuzzaman, Graham, & Eskicioglu, 2011; Sashi & Thanamani, 2010a, 2010b). This may lead the Internet turns to be the bottleneck in accessing the files in the Cloud Computing. Due to the high latency of the Wide Area Network (WAN), the main issue is to design the strategy for efficient data access and share around the world with considerably low time complexity in Data Grid and Cloud research (Zhao, Xu, Xiong, & Wang, 2008). Furthermore, in order to manage the data there are another several problems must be considered such as failures or malicious attacks during execution, fault tolerance, scalability of data and etc. These problems can be solving by using the replication techniques (Bsoul, Al-Khasawneh, EddienAbdallah, & Kilani, 2011; Naseera & Murthy, 2009; Ben Charrada, Ounelli, & Chettaoui, 2010a, 2010b; Shorfuzzaman, Graham, & Eskicioglu, 2010; Zhao, Xu, Xiong, & Wang, 2008; Al-Mistarihi & Yong, 2009; Shorfuzzaman, Graham, & Eskicioglu, 2011; Zhao, Xu, Wang, Zhang, & He, 2010). Globally, we are sending nearly 3 million emails per second, 20 hours of videos are uploaded to YouTube just in 60 seconds, Google processes 24 petabytes of information, publishing 50 million tweets per day, nearly 73 products are ordered on Amazon for every second (Swamkant, 2011). Symantec cloud provider process more than 6 billion emails and 1 billion web requests every day for businesses of every size, global corporations and Governments (Symantec Cloud, 2011) IDC cloud research report shows that worldwide revenue from public IT cloud services exceeded $37 billion in 2009 and is forecast to reach $55.5 billion in 2014 and in 2020 it will hit 900 billion, this growth representing a compound annual growth rate of 27.4%. This rapid growth rate is over five times the projected growth for traditional IT products (5%) (IDC, 2011). The estimated amount of data 1,200 Exabyte (1,200 Billion GB) has been generated only during 2010. The amount of digital information increased by 73 percent in 2008 to an estimated 487 billion GB, according to IDC, in this re-

port the population of world was 360,985,492 in 2000 which increased to 6,767,805,208 in 2009, and the number of Internet user in 2009 was 1,802,330,457, it means that the overall internet user growth is 399.3% with the increase of population. As a result, when the number of hosted applications and the complexity of the whole system grow, it is a serious challenge for PaaS providers to deliver satisfied services as they promised (Shao, Wang, & Mei, 2012). The migration of business to cloud computing can be translated into saving of the software license, maintenance, number of support labor, utilities and office space (Karadsheh & Alhawari, 2011). User of cloud computing can, and have, created a virtual server room on one desktop. 100 million virtual machines are being created per year or 273,972 per day or 11,375 per hour. The number of physical servers in the World today is 50 million. By 2013, approximately 60% of server workloads will be virtualized means will convert to virtual cloud (IDC, 2011). With the popularization and improvement of social and industrial IT development, the people put much higher expectations on the services of computing, communication and network (Sasikala & Chaturvedi, 2011; Khan, Noraziah, Ismail, & Deris, 2012). Hence, for the better management of this day by day increasing heavy storage data, we need efficient scheduling as well replication techniques. The rest of this paper is organized as follows. First, we explain the related works and elaborate the problems to be solved. Afterwards, we explain the proposed methodology. Then, we deliberate the results and discussion. Finally, we then conclude the paper.

RELATED WORK There are some recent and related works that address the problem of scheduling and/or replication as well as the combination between them. However, first time Ranganathan and Foster (2003) proposed the realization of importance of data locality in job scheduling problem. The authors presented a Data Grid



architecture base on three main components i.e., External Scheduler (ES), Local Scheduler (LS) and Dataset Scheduler (DS). ES receives submitted jobs from user, then depend on ES’s scheduler policies, it decide which job to send to which remote site. How to schedule all the jobs, LS of each site decide on its local resources. Keeping track of popularity for each dataset currently available and making data replicating decision. Nguyen and Lim (2007) and Tang, Lee, Tang, and Yeo (2006) have improved the older Ranganathan and Foster (2003) works by integrating the scheduling and replication strategy to improve the scheduling performance.

PROBLEM STATEMENTS i.

Queuing Architecture: Analyzing the works of Tang et al. (2006), Ranganathan and Foster (2003) and Nguyen and Lim (2007), the authors have proposed new scheduling architecture. For the integration of scheduling and replication, in this stage, the author has focused on Total Completion Time (TCT) for a job. In above works, the authors use the following formula for Total Completion Time for a job.

ii. Cloud Services Outage: Cloud computing provides many services (XaaS). Anything can be a XaaS, means anything can be a cloudservice, i.e., Architecture As A Service (AaaS), Framework As A Service (FaaS), Network As A Service (NaaS), Hardwar As A Service (HaaS), Voice As A Service (VaaS), Recovery-As-A-Service (RaaS), Data As A Service (DaaS) etc. Author’s previous work (Stallings, 2000) is describes more about cloud services. Generally “As a whole cloud will never go down,” but a service can (as mentioned in Table 1). When Gmail or some other hosted service has an interruption, it does not indicate a cloud failure, it’s a service failure. Table 1 gives more detail about the outage of various services of cloud, with duration.

Due to these outages cloud does offer unmatched efficiency, flexibility and reliability. As a result cloud computing fail to keep satisfied its user. In reaction going back to an in-house data center requires higher capital costs of hardware, maintenance, and in many cases would require hiring back more IT staff. The clients (end user) are facing outage problems because globally Cloud is one and TTk ,i = max QT(i ), DT f (k ), i + ETk ,i there is no other cloud for substitution to take (1) over (using failover techniques) in this critical situation. A secure cloud model must be capable of providing reliability to the customer on the ETTC ji = max DT ( f ( j ), i ),QT(i ) + EETji location of the data for customer. Hence here (2) our challenge is how we can improve data efficiency, easy accessibility, strong reliability and Where TT is Total Completion Time, QT always availability by keeping cloud service is Queuing Time, DT is Data Transfer time, available forever. ET is the job Execution Time and in equation 2, ETTC is Estimate The Time for Completion PROPOSED METHOD and EET is Estimate Execution Time.

{

)}

(

{

}

{ (

)

}

Here in max DT f ( j ), i ,QT(i ) , one

value DT either QT is ignoring, because only one of them will be maximum value, the other minimum value is ignoring, Even both are two different parameters, and have their importance separately.

For the improvement of data efficiency, easy accessibility, strong reliability and always availability, the current research proposes an approach in order to overcome these qualities in cloud services. We need to make six (let say) local sub clouds on the basis of six sub-



Table 1. Outage in cloud’s services Duration Days/Hours/Minutes

Service

Date

Facebook (Co, 2011)

1h

Aug 10, 2011

Amazon Web Services (2011)

11 h

Apr 21, 2011

Amazon (Hickey, 2011b)

52 m

Apr 21, 2011

Foursquare (http://status.foursquare.com/)

4h

Aug 9,10,25, 2011

Foursquare (http://status.foursquare.com/)

33 m

Oct 22, 2011

Amazon S3 (Hickey, 2011a)

30 m

Aug 16, 2011

Amazon EC2 (Butcher, 2011)

2.4h

Apr 21, 2011

Amazon (Miller, 2010)

8h

May 8, 2010

Microsoft Sidekick (Williams, 2010)

6d

Mar 13, 2009

Microsoft Azure (Williams, 2010)

22 h

Mar 13, 2009

Google Gmail (Williams, 2010)

30 h

Oct 16, 2008

Google GMail, Google Apps (Williams, 2010)

24 h

Aug 15, 2008

continents instead of one Global cloud. Each local sub cloud will have same anytime updated replica (copy) of each other. Due to the local cloud, data in a wisely manner will offer a faster access to files required by cloud client, hence increase the job execution’s performance. Due to local cloud, reliability and accessibility will increase. If one sub cloud goes down, any client from anywhere in the world can use another sub cloud under the “failover” techniques as depicted in Figure 2. Data availability and easy accessibility are paramount for cloud storage providers, as data loss or unavailability can be damaging both to the bottom line, by failing to hit targets set in service level agreements (Monash, 2011) and to business reputation outages often make the news (http://wiki.cloudcommunity.org). Data availability and easy accessibility are typically achieved through under-the-covers replication. Dealing with large amount of data makes the requirement for efficiency, in data access, more critical. A good scheduling strategy will allow shortest access to the required data; therefore reduce the data access time. Vice versa, replication strategy that allows place data in a wisely manner will offer a faster access to required files. The goal of replication is to

shorten the data access not only for user accesses but enhancing the job execution performance. For this approach we need an architecture, in which scheduling is compatible with replication.

REPLICATION TECHNIQUE In data replications architecture, the data will be replicated into present available sub clouds. If one of sub cloud failed, it will fail independently and will not affect the others node. Therefore, the data replication will improve the reliability, availability of the data and performance. Replication in distributed environment receives particular attention for providing efficient access to data, fault tolerance (Bsoul, Al-Khasawneh, EddienAbdallah, & Kilani, 2011) and enhance the performance of the system (Zhao, Xu, Wang, Zhang, & He, 2010; Noraziah, Deris, Norhayati, Rabiei, & Shuhadah, 2008). The replication strategy can minimize the time access to the file by creating many replicas and storing replicas in appropriate locations, which provide nearby data access. Furthermore, using replication is to reduce bandwidth consumption (Naseera & Murthy, 2009; Ben Charrada, Ounelli, & Chettaoui, 2010a, 2010b; Shorfuzzaman, Graham,



Figure 2. Simple sub-cloud structure for the globe

& Eskicioglu, 2010) to achieve efficient and dependable data access to improve access time (Monash, 2011; Noraziah, Deris, Norhayati, Rabiei, & Shuhadah, 2008) fault tolerance (Bsoul, Al-Khasawneh, EddienAbdallah, & Kilani, 2011) and load balancing (Zhao, Xu, Wang, Zhang, & He, 2010). There are three fundamental questions (Zhao, Xu, Wang, Zhang, & He, 2010; Ben Charrada, Ounelli, & Chettaoui, 2010a, 2010b; Zhao, Xu, Xiong, & Wang, 2008; Tang, Lee, Tang, & Yeo, 2006; Al-Mistarihi & Yong, 2009; Ranganathan & Foster, 2001) that must be answer in managing replica placement strategy in data grid/cloud: a. When should the replicas be created? b. Which files should be replicated? c. How many replicas should be created? Different replication strategies have been developed or design to answer these questions above. For the solution, the current research proposes an approach in order to improve data efficiency, easy accessibility, strong reliability

and forever availability; these qualities are the key needs of cloud services. We need to make six clouds on the basis of six cotenants instead of one cloud globally. Each local sub cloud will have same replica (copy) of each other anytime updated. Due to the local cloud, data in a wisely manner will offer a faster access to files require by cloud client, hence increase the job execution’s performance. Due to local cloud, data will be reliable, no risk for losing data in presence of other replicas. Accessibility will increase data locality, data will access from local near cloud instead of global faraway cloud. Quick Access of data due to shortest distance because the execution performance is greatly impacted by the data locality (Ranganathan & Foster, 2001; Nguyen & Lim, 2007; Khan, Noraziah, Deris, & Ismail, 2011). To fulfill this approach the ROWA (ReadOnce-Write-All) is the most suitable Model. ROWA is the simplest technique for managing replicated data. A read operation is allowed to read any copy of data, meanwhile, a write operation is required to write all copies of data to the target places. All of the replicas have the same



value when an update transaction commits. This technique has the lowest communication cost of read operation, because the read operation is using only once.

SCHEDULING ARCHITECTURE The scheduling architecture is encapsulated in three distinct modules, as shown in Figure 3. External Scheduler (ES): Each user is associated with an External Scheduler in the system and submits jobs to that External Scheduler. ES decides the remote site to which to send the job to depending on some scheduling algorithm. ES uses external information for taking decision as input such as load at a remote site or the location of a dataset. Local Scheduler (LS): Assigned jobs are managing by Local Scheduler to run at a particular site. The allocation of allocated jobs is also responsibility of the LS. LS decide about the priority and refusing to run jobs submitted by a certain user. Dataset Scheduler (DS): At each site DS keeps track the popularity of each locally available data set. By using external information such

as whether the data already exist at the concern site and loading to target remote site. With encapsulation of three schedulers ES, LS, and DS, the Time for Completion of a job is also encapsulation of three Times. Wq or t1 is that time, which a job passing in waiting after entrance and before starting execution. Time denoted by Ws or t2 is the job execution time after entrance from queue and before starting transferring process. Tt or t3 is the time after completion execution time and upto the completion of data transferring process. Means the Total Completion Time for a job is the sum of all these three times (wait in queue, execution time and transfer time).

QUEUING MODELS In queuing theory, a queuing model is used to approximate a real queuing situation or system, so the queuing behaviour can be analysed mathematically. These measures are important as issues or problems caused by queuing situations are often related to customer dissatisfaction with service or may be the root cause of economic losses in a business. Analysis of the relevant

Figure 3. Scheduling architecture



queuing models (Stallings, 2000; Blanc, 2011; Dombacher, 2009; Lipsky, 2009; Jain, Mohanty, & Bohm, 2007; Ng & Soong, 2008; Kobayashi, & Konheim, 1977; Philippe, 1998; Adan & Resing, 2008) allows the cause of queuing issues to be identified and the impact of proposed changes to be assessed. Queuing models can be represented using Kendall’s notation as in Dombacher (2009), Philippe (1998), and Adan and Resing (2008) where A/B/S/K/N/D A: Is the inter-arrival time distribution or arrival rate per unit time, denoted by λ. B: Is the service time distribution or service rate per unit time, denoted by μ. U: Is the server utilization, denoted by ρ. S: Is the number of servers, denoted by C. K: Is the queue system capacity, same notation K. N: Is the calling population or total number of jobs, denoted by P. D: Is the service discipline assumed, same notation D.

M/M/1: Single-Server Queue Model M/M/1/∞/∞ represents a single server that has unlimited queue capacity and infinite calling population, The mathematical nature of the exponential distribution, a number of quite simple relationships are able to be derived for several performance measures based on knowing the arrival rate and service rate. The arrival is Poisson process, meaning the statistical distribution of the inter-arrival times still follow the exponential distribution. The distribution of the service time may follow any general statistical distribution, not just exponential. Relationships are still able to be derived for a (limited) number of performance measures if one knows the arrival rate and the mean and variance of the service rate. Models briefly discussed in Dombacher (2009), Lipsky, (2009), Jain, Mohanty, and Bohm (2007), Ng and Soong (2008), and Little and Graves (2008).

M/M/C: Multiple-Servers Queue Multiple (identical)-servers queue situations are frequently encountered in telecommunications or a customer service environment. When modelling these situations care is needed to ensure that it is a multiple servers queue, not a network of single server queues, because results may differ depending on how the queuing model behaves (Dombacher, 2009; Lipsky, 2009; Jain, Mohanty, & Bohm, 2007).

M/M/∞: Infinitely Many Servers While never exactly encountered in reality, an infinite-servers (e.g., M/M/∞) model is a convenient theoretical model for situations that involve storage or delay, such as parking lots, warehouses and even atomic transitions. In these models there is no queue, such as, instead each arriving customer receives service. When viewed from the outside, the model appears to delay or store each customer for some time (Jain, Mohanty, & Bohm, 2007; Ng & Soong, 2008).

M/M/c/K/∞Model: Capacity Constraints By customizing the parameters for the load dependent model, capacity constraints may be introduced to a multiserver system M/M/c/K. The queue capacity limitation is the only difference between the limited and the unlimited model (Dombacher, 2009; Lipsky, 2009).

M/M/c/K/P Model: Finite jobs population The M/M/c/K/P queuing system is another variant of the M/M/c queuing system. In this system, there is a finite number P (General notation for population is M) of potential jobs, while there is room (queue) for K jobs, including the jobs in service (P ≥ K ≥ C ) . Note that this is in fact a closed system where each of the



P jobs is either inside the system (waiting or being served) or passive outside the system until its next visit to the system. In this case, the arrival rate depends on the number of passive customers so that the arrival process is not a Poisson process. It is assumed that each potential customer returns to the system after an exponentially distributed passive time with ratel. Such a system is stable for every positive value of l¸ and service rate m. Special case is the M/M/c/c/P loss systems which are often referred to as Engset loss systems. Stallings (2000) and Blanc (2011) have discussed and more detail in Dombacher (2009), Lipsky (2009), Jain, Mohanty, and Bohm (2007), Ng and Soong (2008), Kobayashi and Konheim (1977), and Adan and Resing (2008).

SCHEDULING ARCHITECTURE When receiving a job submission, the LS will estimate the time for completing executing (ETTC) a job in a site i, as mentioned by Ranganathan, K., & Foster (2001), Ranganathan and Foster (2003), and Nguyen and Lim (2007).

{ (

}

)

ETTC ji = max DT f (( j ), i ,QT(i ) + EETji

(3)

Where DT is the Data Transfer Time for job; QT is the Queuing Time in site; and EET is the Estimate Execution Time for a job. For Turnaround Time simply we use, Total Completion Time (TCT) for a job which is the sum of all these three Times.

(

)

TCTj ,i = QT(i ) + ETj, i + DT f ( j )

will be waiting in the queue (zero or more); assume the average number of items waiting is w, and the mean time that an item must wait is Tw. Tw is averaged over all incoming items, including those that do not wait at all. The server handles incoming items with an average service time Ts; this is the time interval between the dispatching of an item to the server and the departure of that item from the server. Finally, two parameters apply to the system as a whole. The average number of items resident in the system, including the item being served (if any) and suppose the items waiting (if any), is r; and the average time that an item spends in the system, waiting and being served, is T; we refer to this as the mean residence time. If we assume that the capacity of the queue is infinite, then no items are ever lost from the system; they are just delayed until they can be served. FIFO (First In First Out) is suitable principals to use (Figure 4). Assumptions: According to Little’s Law (Little & Graves, 2008) Tw = Wq = mean waiting time Ts = Ws = mean service (execution) time for each arrival; Tr = Wq+Ws = mean residence time (time spends of an item in system) Here, Tw = Wq = QT, Ts = ET = Ws and DT = Tt (5) Hence, Tr = Wq +Ws

(6)

In our Equation 2, the residence time is (4)

According to Little’s Law, Equation 4 illustrates some important parameters associated with a queuing model. Items arrive with some average rate (items arriving per second). At any given time, a certain number of items

Tr = Wq +Ws

(7)

Where Wq is the Wait in Queue for a job and Ws is the Wait in Server which is generally Execution time. As described in Figure 5.



Figure 4. A schematic representation for a single server queuing system

Putting Equations 4 in Equation 2, the Total Completion Time for a job TCT TCTj ,i = QTw + ETs +

(

)

DT f ( j ) = Wq +Ws + Tt

(8)

To evaluate this technique, M/M/c/K/P Model has used. By using this model, calculation has done to determine all other parameters, which we need for the calculation of TCT. The purpose of using this model is, because we need the job size in advance to calculate the total time for transferring.

Using of M/M/c/K/P Model: Finite Jobs Population If there are n jobs in the queue there are N - n jobs in the source. We assume that jobs wait in the source an exponentially distributed amount of time with average before returning to the queue and that they are independent of each other. If there are N - n jobs in the source then they create a Poisson input process to the queue with rate λ (Figure 6). In Finite Population Model (Ghemawat, Gobioff, & Leung, 2003; Amazon, 2009; Borthakur, 2007; Dean & Ghemawat, 2008).

Figure 5. Schematic view, flow of JOB



Figure 6. State transition diagram, finite population M/M/1

(M − n) λ  λn =    0   

for 0 ≤ n < M for n ≥ M

for 1 ≤ n < c for n ≥ c

For average queue size Lq and average waiting time in queue Wq calculation, we need probability of ‘0’ entity in the system P0, probability of ‘n’ entities being in the system Pn, average number of jobs in the system L and average time spent in the system Ls. Using Hickey (2011a, 2011b), Butcher (2011), Williams (2010), Foursquare (http://status.foursquare. com/), and IDC (2011) we can calculate as:   (M  pn =    (M 

,

 M!  ρn p 0 − n ) !n!  M!  n! ρn p 0 − n ) !n! (cn−cc!)

nomial coefficient. Applying ∑ n =0 pn = 1 and solving for p0 gives M

Here M denotes the total size of the population, means number of jobs. Assuming a system with C < M service units (Dean & Ghemawat, 2004; Ghemawat, Gobioff, & Leung, 2007; Amazon, 2009; Borthakur, 2007; Dean & Ghemawat, 2008), i.e. n µ µn =  cµ 

M  M!  where   = , denoting the bi n  (M − n ) !n!

for 0 ≤ n < c for c ≤ n ≤ M

(9)

 c −1     ∑   n =0 (M p0 =    M   ∑   n =c (M 

   ρn +     − n ) !n!     M!  n! ρn    − n ) !n!  cn-cc !  

−1

M!

(10)

Using the definition of the expected value, It is now easy to drive the average system size means number of jobs in the system (Hickey, 2011a, 2011b; Butcher, 2011; Williams, 2010; Foursquare, http://status.foursquare.com; IDC, 2011)    c −1   ∑ n   n =0 (M L =    M   ∑ n   n =c (M 

   ρn +     − n ) !n!  p  0  n ! M!  n   n-c ρ  − n ) !n!  c c !   M!

M

M

M

n =c

n =c

n =c

(11)

Lq = ∑ (n − c ) pn = ∑ npn − c ∑ pn



   M!  ρn = L − c + p0 ∑ (c − n )   !n! − M n ( n =0 )  (12) c −1

By using Little’s Law (Foster, Yong, Raicu, & Lu, 2008) and detail in Dean and Ghemawat (2004) and Ghemawat, Gobioff, and Leung (2003) with mean arrival rate λ can give

Lq

+ L −c + λ (M − L )   c −1  M!  ρn + p0 ∑ (c − n )  (M − n ) !n!  n =0   Sizef (k ) / BW(i,j )

TCTji =

(17)

RESULTS AND DISCUSSION

In order to evaluate the performance of proposed Ws = (13) architecture, author has removed the MAX function from equation 1 and 2. And add all λ (M − L ) three parameters as mentioned in equation 4 or in equation 8. There is great impact on accuracy and in calculation of Total Time of Completion for a job, which brings improvement in efficiency (in term of accuracy). Lq Wq = (14) Dealing with large amount of data makes λ (M − L ) the requirement for efficiency in data access more critical. Users of the world need more For Consuming Time (Execution Time) copy (replicas) of the Global one cloud. This one cloud should be replicate in many replicas by Processor on the base of sub-continents or may be on the base of population (number of user). Each local   c −1  M! n  sub cloud will have same anytime updated copy  ρ ET ( j ) = L − c + p0 ∑ (c − n )    !n! − M n  of each other. Due to the local cloud, data in a n =0 )  ( (15) wisely manner will offer a faster access to files required by cloud client, hence increase the job execution’s performance. Due to local cloud, If N p is the Number of Jobs in Processer reliability and accessibility will increase. In and Tp is the Total Time Consumed by proces- this case if one sub cloud goes down, any client sor for those jobs. Then Data Transfer Time from anywhere in the world can use another sub according to Ranganathan and Foster (2003) cloud under the “failover” techniques. Cloud service failure is not acceptable in this archiand Nguyen and Lim (2007) will be as tecture. In propose architecture, ROWA is the most suitable technique to combine replication DT f ( j ) = sizef (k ) / BW(i,j ) (16) with scheduling. For evaluating new architecture, by using formulas, first we calculate the Total consume Where Size f (k ) is the FileSize in bytes time just in transferring of data, excluding the and BW is the available Bandwidth between queue time and system time. As described in computing sites. Now Putting Equations 13, Figure 7. 14, and 16 in Equation 8, result will be as This section compares overall results, and calculates the variance between Total Time (TT) according to existing models and Total CompleL

(

)



Figure 7. Transfer time for data

tion Time (TCT) according to new proposed model. Here TT is actual and TCT is estimated value. The following general formula for error estimation has been used, to compare both time variations. After General Error and Accuracy Estimation, now we need to compare the change in Error Estimation with various bandwidths.

Error is decreasing with increasing population M, it means that new proposed model is more efficient to use for large data transferring as shown in Figure 8 by using C=1, λ=1, μ=1, M= 2 ~ 50, BW=56Kbps. Using C=1, λ=1, μ=1, M=2, BW=56Kbps, the accuracy is 85.500%, it is increasing upto 92.4639% by using M=50, and by taking C=1,

Figure 8. Error estimation with C = 1 and BW= 56Kbps



Table 2. Error and accuracy estimation with C = 1, BW = 512Kbps

λ=1, μ=1, M=2, BW=512Kbps as shown in Figure 8. Accuracy is 98.5000%, by using M=50, it will increase upto 99.1753% as mentioned in Table 2. Result shows that with increase in population (M), by using same bandwidth and same number of server, the gap is decreasing between the results of both techniques. Decreasing the gap, means accuracy is increasing by using new technique. When M>500, stability point (where accuracy is 100%) can be achieved. Hence new technique is more efficient when we need to transfer large amount of data.

ACKNOWLEDGMENT Appreciation conveyed to Ministry of Higher Education Malaysia for project financing under Exploratory Research Grant Scheme, RDU120608.

CONCLUSION This paper presents scheduling architecture for cloud computing to support efficient data access for the job. Cloud should be a “true service provider” which is the demand of each user. Proposed architecture fulfills this property, because there is no chance to go down all local clouds at a time. This scheduling technique gives

importance to each parameter while calculating Total Time of Completion for a job. This research evaluate the result of total completion time for jobs only using single server finite population model using 56~512kbps bandwidth. There is a great impact on accuracy by taking each parameter separately. Further we need to evaluate our new model by using others M/M/C, M/M/Inf, M/M/C/K and M/M/C/*/M queuing models. In future work, we plan to investigate more realistic scenarios and real user access patterns. Additionally, we want to propose a complete real time model by using CloudSim, with combination of replication and scheduling.

REFERENCES Adan, I., & Resing, J. (2008). Queuing theory. Eindhoven, The Netherlands: Eindhoven University. Al-Mistarihi, H. H. E., & Yong, C. H. (2009). On fairness, optimizing replica selection in data grids. IEEE Transactions on Parallel and Distributed Systems, 20(8), 1102–1111. doi:10.1109/TPDS.2008.264 Amazon. (2009). Amazon S3. Retrieved from http:// aws.amazon.com/s3 Amazon Web Services. (2011). Summary of the Amazon EC2 and Amazon RDS service disruption. Retrieved from http://aws.amazon.com/message/65648/



Artalejo, J. R., & Lopez-Herrerob, M. J. (2007). A simulation study of a discrete-time multiserver retrial queue with finite population. Journal of Statistical Planning and Inference, 137, 2536–2542. doi:10.1016/j.jspi.2006.04.018 Ben Charrada, F., Ounelli, H., & Chettaoui, H. (2010a). An efficient replication strategy for dynamic data grids. In Proceedings of the International Conference on P2P, Parallel, Grid, Cloud and Internet Computing (pp. 50-54). Ben Charrada, F., Ounelli, H., & Chettaoui, H. (2010b). Dynamic period vs. static period in data grid replication. In Proceedings of the International Conference on P2P, Parallel, Grid, Cloud and Internet Computing (pp. 565-568). Blanc, J. P. C. (2011). Queuing models: Analytical and numerical methods. West Tilburg, The Netherlands: Tilburg University. Borthakur, D. (2007). The hadoop distributed file system: Architecture and design. Retrieved from http://lucene.apache.org/hadoop Brantner, M., Florescuy, D., Graf, D., Kossmann, D., & Kraska, T. (2008). Building a database on S3. In Proceedings of the ACM SIGMOD International Conference on Management of Data (pp. 251-263). Bsoul, M., & Al-Khasawneh, A., EddienAbdallah, E., & Kilani, Y. (2011). Enhanced fast spread replication strategy for data grid. Journal of Network and Computer Applications, 34(2), 575–580. doi:10.1016/j. jnca.2010.12.006 Butcher, M. (2011). Amazon EC2 goes down, taking with it Reddit, Foursquare and Quora. Retrieved from http://eu.techcrunch.com/2011/04/21/amazonec2-goes-down-taking-with-it-reddit-foursquareand-quora/ Buyya, R., Yeo, C. S., Venugopal, S., Broberg, J., & Brandic, I. (2009). Cloud computing and emerging IT platforms: Vision, hype, and reality for delivering computing as the 5th utility. Future Generation Computer Systems, 25, 599–616. doi:10.1016/j. future.2008.12.001 Co, B. (2011). Why is Facebook down (August 10, 2011)? Hacked by anonymous? Retrieved from http:// manila-paper.net/why-is-facebook-down-august-102011-hacked-by-anonymous/8446 Dattatreya, G. R. (2008). Performance analysis of queuing and computer networks. Boca Raton, FL: Taylor & Francis. doi:10.1201/9781584889878

Dean, J., & Ghemawat, S. (2004). MapReduce: Simplified data processing on large clusters. In Proceedings of the Sixth Symposium on Operating System Design and Implementation (pp. 137-149). Dean, J., & Ghemawat, S. (2008). MapReduce: Simplified data processing on large clusters. Communications of the ACM, 51(1), 107–113. doi:10.1145/1327452.1327492 Dombacher, C. (2009). Stationary queuing models with aspects of customer impatience and retrial behaviour. Retrieved from http://www.telecomm. at/documents/Stationary_QM.pdf Foster, I., Yong, Z., Raicu, I., & Lu, S. (2008). Cloud computing and grid computing 360-degree compared. In Proceedings of the Grid Computing Environments Workshop (pp. 1-10). Ghemawat, S., Gobioff, H., & Leung, S.-T. (2003). The Google file system. In Proceedings of the Nineteenth ACM Symposium on Operating Systems Principles (pp. 29-43). Gupta, U. C., & Samanta, S. K. (2004). Computing queue length and waiting time distributions in finite-buffer discrete-time multiserver queue with late and early arrivals. Computers & Mathematics with Application, 48, 1557–1573. doi:10.1016/j. camwa.2004.05.010 Hbase Development Team. (2007). Hbase: Bigtablelike structured storage for Hadoop HDFS. Retrieved from http://wiki.apache.org/lucene-hadoop/Hbase Hickey, A. R. (2011a). Amazon offers explanations, apologies for dual cloud outages. Retrieved from http://www.crn.com/news/cloud/231500023/ amazon-offers-explanations-apologies-for-dualcloud-outages.htm?itc=refresh Hickey, A. R. (2011b). New Amazon cloud outage takes down Netflix, Foursquare. Retrieved from http://www.crn.com/news/cloud/231300459/ new-amazon-cloud-outage-takes-down-netflixfoursquare.htm?itc=refresh IDC. (2011). Analyse the Cloud. Retrieved from http://www.idc.com/prodserv/idc_cloud.jsp 1 Jain, J. L., Mohanty, S. G., & Bohm, W. (2007). A course on queuing models. Boca Raton, FL: Chapman & Hall/CRC/Taylor & Francis Group. Karadsheh, L., & Alhawari, S. (2011). Applying security policies in small business utilizing cloud computing technologies. International Journal of Cloud Applications and Computing, 1(2), 29–40. doi:10.4018/ijcac.2011040103



Khan, N., Noraziah, A., Deris, M. M., & Ismail, E. I. (2011). Cloud Computing: Comparison of various features. In Ariwa, E., & El-Qawasmeh, E. (Eds.), DEIS communications in computer and information science (Vol. 194, pp. 243–254). Berlin, Germany: Springer-Verlag. Khan, N., Noraziah, A., Ismail, E. I., & Deris, M. M. (2012). Cloud computing: Analysis of various platforms. International Journal of E-Entrepreneurship and Innovation, 3(2), 51–59. Kobayashi, H., & Konheim, A. G. (1977). Queuing models for computer communications system analysis. IEEE Transactions on Communications, 25(1), 2–29. doi:10.1109/TCOM.1977.1093702 Lipsky, L. (2009). Queuing theory: A linear algebraic approach (2nd ed.). New York, NY: Springer Science. Little, D. C., & Graves, S. C. (2008). Little’s law. Cambridge, MA: Massachusetts Institute of Technology. Mearian, L. (2011). Virtual Machine cloud replication and restoration to boom. Retrieved from http://www.networkworld.com/news/2011/110811virtual-machine-cloud-replication-and-252868. html?source=NWWNLE_nlt_virtualization_2011-11-14 Miller, R. (2010). Amazon addresses EC2 power outages. Retrieved from http://www.datacenterknowledge.com/archives/2010/05/10/amazonaddresses-ec2-power-outages/ Miller, R. (2011). Outages altering cloud perception, practice. Retrieved November 13, 2011, from http://www.datacenterknowledge.com/ archives/2011/10/07/outages-altering-cloud-perception-practice/ Monash, C. (2011). Oracle announces an Amazon cloud offering. Retrieved from http://www.dbms2. com.Oracle-announces-an-amazon-cloud-offering Moretti, C., Bulosan, J., Thain, D., & Flynn, P. J. (2008). All-Pairs: An abstraction for data-intensive cloud computing. In Proceedings of the IEEE International Symposium Parallel & Distributed (pp. 1-11). Naseera, S., & Murthy, K. V. M. (2009). Agent based replica placement in a data grid environment. In Proceedings of the First International Conference on Computational Intelligence, Communication Systems and Networks (pp. 426-430). Ng, C.-H., & Soong, B.-H. (2008). Queuing modelling fundamentals with application in communication networks (2nd ed.). Chichester, UK: John Wiley & Sons. doi:10.1002/9780470994672

Nguyen, D. N., & Lim, S. B. (2007). Combination of replication and scheduling in data grids. International Journal of Computer Science and Network Security, 7(3), 304–308. Noraziah, A., Deris, M. M., Norhayati, M. R., Rabiei, M., & Shuhadah, W. N. W. (2008). Managing transaction on grid-neighbour replication in distributed system. International Journal of Computer Mathematics, 86(9), 1–10. Pérez, J. M., Carballeira, F. G., Carretero, J., Calderón, A., & Fernández, J. (2010). Branch replication scheme: A new model for data replication in large scale data grids. Future Generation Computer Systems, 26(1), 12–20. doi:10.1016/j.future.2009.05.015 Philippe, N. (1998). Basic elements of queuing theory: Application to the modelling of computer systems. Le Chesney, France: INRIA. Ranganathan, K., & Foster, I. (2001). Identifying dynamic replication strategies for a high performance data grid. In C. A. Lee (Ed.), Proceedings of the Second International Conference on Grid Computing (LNCS 2242, pp. 75-86). Ranganathan, K., & Foster, I. (2003). Simulation studies of computation and data scheduling algorithms for data grids. Journal of Grid Computing, 1(1), 53–62. doi:10.1023/A:1024035627870 Sashi, K., & Thanamani, A. S. (2010a). A new replica creation and placement algorithm for data grid environment. In Proceedings of the International Conference on Data Storage and Data Engineering (pp. 265-269). Sashi, K., & Thanamani, A. S. (2010b). Dynamic replica management for data grid. International Journal of Engineering and Technology, 2(4), 329–333. Sasikala, P., & Chaturvedi, M. (2011). Cloud computing towards technological convergence. International Journal of Cloud Applications and Computing, 1(4), 44–59. doi:10.4018/ijcac.2011100104 Shao, J., Wang, Q., & Mei, H. (2012). Model based monitoring and controlling for Platform-as-a-Service (PaaS). International Journal of Cloud Applications and Computing, 2(1), 1–15. doi:10.4018/ ijcac.2012010101 Shorfuzzaman, M., Graham, P., & Eskicioglu, R. (2010). Distributed popularity based replica placement in data grid environments. In Proceedings of the International Conference on Parallel and Distributed Computing, Applications and Technologies (pp. 66-77).



Shorfuzzaman, M., Graham, P., & Eskicioglu, R. (2011). QoS-aware distributed replica placement in hierarchical data grids. In Proceedings of the IEEE International Conference on Advanced Information Networking and Applications (pp. 291-299). Stallings, W. (2000). Queuing analysis: A practical guide for computer scientists. Retrieved from http:// www.electronicsteacher.com/download/queuinganalysis.pdf Swamkant. (2011). The amount of data generated & consumed on the Internet. Retrieved from http:// www.yourdigitalspace.com/2010/10/the-amountof-data-generated-and-consumed-on-the-internet/ Symantec Cloud. (2011). Global infrastructure. Retrieved from http://www.symanteccloud.com/ aboutus/tech/global_infrastructure

Weiss, A. (2007). Computing in the cloud. ACM Networker, 11, 18–25. doi:10.1145/1268577.1268578 Williams, A. (2010). Top 5 cloud outages of the past two years. Retrieved from http://www.readwriteweb. com/cloud/2010/02/top-5-cloud-outages-of-the-pas. php Zhao, W., Xu, X., Wang, Z., Zhang, Y., & He, H. (2010). A dynamic optimal replication strategy in data grid environment. In Proceedings of the International Conference on Internet Technology and Applications (pp. 1-4). Zhao, W., Xu, X., Xiong, N., & Wang, Z. (2008). A weight-based dynamic replica replacement strategy in data grids. In Proceedings of the IEEE Asia-Pacific Services Computing Conference (pp. 1544-1549).

Tang, M., Lee, B. S., Tang, X., & Yeo, C. K. (2006). The impact on data replication on job scheduling performance in the data grid. Future Generation Computer Systems, 22, 254–268. doi:10.1016/j. future.2005.08.004

Nawsher Khan is a Doctoral Scholar at the University Malaysia Pahang (UMP) Malaysia. He received his bachelor degree in computer science from University of Peshawar, Pakistan and master degree in computer science from Hazara University, Pakistan. Since 1999, he has worked in various educational institutions. In 2005, he was appointed as CM in NADRA (National Database and Registration Authority) under the Interior Ministry of Pakistan. Currently he is a Research Assistant, in Faculty of Computer System and Software Engineering at University Malaysia Pahang (UMP) Malaysia and his research interest include Cloud Computing, Data Management, Distributed System, Scheduling and Replication. In addition he has published more than 25 articles in various International journals and conference proceedings. Noraziah Ahmad received PhD in Distributed Database from University Malaysia Terengganu (UMT) in 2007. She has published more than 130 papers in the journals and conference proceedings. Currently, she is the senior lecturer at Faculty of Computer Systems & Software Engineering, University Malaysia Pahang. In addition to serving as international program committee member and reviewers in many conferences, she is currently an editorial board members of the International Journal of Engineering and Technology (IJET), International Journal of Web Application (IJWA) and Journal of Emerging Technologies in Web Intelligence (JETWI); a member of IEEE Computer Society, International Association of Engineers (IAENG), World Academy of Science, Engineering and Technology (WASET), Malaysian National Computer Confederation (MNCC) and Senior member of International Association of Computer Science and Information Technology (IACSIT). She is also an advisory board of The Society of Digital Informatics and Wireless Communications (SDIWC). Her current research interests include distributed database systems, distributed systems, data grid, cloud computing and biofeedback.



Tutut Herawan received his BEd degree in year 2002 and MSc in year 2006 degree in Mathematics from Universitas Ahmad Dahlan and Universitas Gadjah Mada Yogyakarta Indonesia, respectively. Late, he obtained his PhD from Universiti Tun Hussein Onn Malaysia. Currently, he is a Senior Lecturer at Department of Computer Science and leader of Database and Knowledge Management (DBKM) research group at Faculty of Computer Systems and Software Engineering, Universiti Malaysia Pahang (UMP). He published more than 90 research papers in journals and conference proceedings. His research area includes Data Mining and Knowledge Discovery, Rough Set theory, Soft Set theory, and Mathematics Education. Zakira Inayat received her MS Degree from Institute of management Sciences (IMSciences) Peshawar, Pakistan in 2012, Master in Computer Science from Hazara University, Pakistan in 2006 and bachelor degree from University of Peshawar, Pakistan in 2003. Currently she is the senior lecturer at Department of Computer Science and Information Technology, University of Engineering and Technology, Peshawar, Khyber Pakhtunkhwa, Pakistan. Her research interests include Cloud Computing, Data Management and Distributed System. In addition she has published 5 articles in various International journals and conference proceedings.
