Вы находитесь на странице: 1из 6

Towards a Compromise between the Customer

Objectives and the Provider Profit in cloud Systems


through Data Replication
Abdenour Lazeb1, Riad Mokadem2 and Ghalem Belalem3
1 Université Oran1, Ahmed Ben Bella , Algérie
lazeb.abdenour@edu.univ-oran1.dz
2 Institut de Recherche en Informatique de Toulouse (IRIT), Paul Sabatier University, Tou-
louse, France
Riad.Mokadem@irit.fr
3 Université Oran1, Ahmed Ben Bella , Algérie
ghalem1dz@gmail.com

Abstract. Cloud Computing has emerged in recent years as a major paradigm


for using IT infrastructures. This one meets growing needs and demands in
terms of availability and flexibility. It is critical for a cloud tenant to minimize
its payment cost to cloud providers while providing SLA-guaranteed services.
In this work, an adaptive dynamic replication is presented. The case study
shows that our approach can significantly decrease the total cost of data storage
for the provider without neglecting the user satisfaction.

Keywords: SLA, Provider, Data Management, Cloud Systems, Data Replica-


tion, Cost Model, Business Model, Performance.

1 Introduction
Applications in the field of video conferencing, open systems administration,
voice union additionally the gaming part fundamentally depends on the effectiveness
in superior correspondence in server farms. Improving these in a superior manner
with high dependable administrations and with low inactivity sort of information ad-
ministration which will be incorporated. Information resources will be taken nearer
with the physical information systems for server farm that has been proposed. This
procedure will improve the framework transfer speed additionally the information
accessibility among the foundation at geologically has been circulated among the
server farm. Dynamic, cost-mindful, enhanced information replication system is rec-
ommended. It distinguishes the minimum measure of reproduced pieces of infor-
mation to ensure that there is information accessibility. The idea of the improved and
proficient Knapsack [1] has been utilized to lessen the expense in the zones of repli-
cation with presenting the replicate idea and at the expansion of expense in the server,
farm to minimize the cost less without blocking the accessibility of information.
Many endeavors have been made on how to provide performance SLAs ensuring
least level of performance[2], but benefit suppliers still got to confront to a problem,
2

which is what they ought to do when the predefined performance SLAs are abused
beneath a few specific circumstance [3].
In this paper, we will propose an algorithm for an efficient replication management.
We propose an algorithm that decides the replication degree of each data object. It
minimizes the ask failure probability, consistency support cost and capacity cost.
This paper is organized as follows: Section 2 tackles related work. Section 3 ex-
plains our approach aspects. Finally, the last section contains the conclusions and
future work.

2 Related Work

Numerous strategies have been proposed to deal with non-connected or corre-


sponded machine disappointments. Zhong et al. [4] accepted autonomous machines
disappointments and proposed a model that accomplishes high expected administra-
tion accessibility. Be that as it may, this model does not consider connected machine
disappointments thus can't deal with such disappointments. Cidon et al. proposed
Copyset Replication [5] and Tiered Replication [6] to diminish the recurrence of in-
formation misfortune brought about by related machine disappointments by constrain-
ing the reproduction hubs of numerous pieces to a solitary copyset. These strategies
for related machine disappointments don't consider information prevalence to limit
information misfortune likelihood in corresponded and non-associated machine dis-
appointments. Not at all like Copyset Replication and Tiered Replication, our pro-
posed MRR decreases likelihood of information misfortune brought about by both
associated and non-related machine disappointments with thinking about information
ubiquity, and it infers various replication degrees for information objects with various
popularities and in this way builds the general information object accessibility by
making more copies for famous information objects and less for disliked information
objects. In addition, MRR reproduces information objects with considering lessening
consistence support cost and capacity cost, which is basic for cloud suppliers to boost
the income. Thomson et al. [7] displayed CalvinFS, a versatile conveyed document
framework, which on a level plane parcels and duplicates record framework metadata
over a common nothing group of servers. In any case, CalvinFS can't deal with both
associated and non-related machine disappointments while limiting expense (e.g.,
consistency support cost and capacity cost) brought about by replication. The above
recently proposed techniques can't deal with both connected machine disappointments
and non-related machine disappointments while using the constrained asset to build
information accessibility. They disregard information prevalence existing in current
distributed storage framework and along these lines can't completely use the asset to
build information accessibility. Likewise, they don't simultaneously consider limiting
the consistency upkeep cost and capacity cost brought about by replication.
3

3 The Proposed Approach

We propose a replication system which contains needy and autonomous modules


in the design appeared in Fig 1. Every module assumes a job in the Work Process.
.

Fig. 1. The proposed system architecture.

3.1 Model System

For our model, we have a Master, Global System Module and data-centers. Master
contains five modules: DC Mapping, Solver, Springy, Repair and SLA-Violation.
Global System Module composed by data matrices and Min Cost Access that are
utilized by all the framework.

We have a set of data-centers where: DCi  DC1 , DC2 ,,DCN  Each DCi
have two modules Leader Election and Node-Mapping and many Group where:
Gi.h  Gi.1 , Gi.2 ,, Gi.H  Each Group have Leader Li.h So

Li.h  Li.1 , Li.2 , Li.3 , Li.4 , Li.H  Without overlooking that each cluster contain
Lot of Node Ni.h.k  Ni.h.1 , Ni.h.2 , Ni.h.3 , Ni.h.4 , Ni.h.K  .
Finally, we have File for each Node in Group for Data center.
We mentioned that Fi.h,k , j   Fi.h.k .1 , Fi.h.k .2 , Fi.h.k .3 , Fi.h.k .M  .

3.2 Description
Our architecture consists of three module described as follows:
4

Global System Module: it's a component contains three elements that are used glob-
ally throughout the system
Min Access Cost: for all exchanges, it is fundamental that the solicitations pursue
the shortest path to touch base at the ideal goal (Bellman–Ford algorithm) without
dismissing the expense of access for every Node crossed.
destination destination
Min j 1
Cacc j  
i 1
Ctransferi
(1)

Cacc j : cost of access Node j. Ctransferi : Cost of transfer by the link i.

Data Matrices: it contains three matrices:


Matrix of Popularity𝑃𝑖ℎ𝑘𝑗 , for the access frequency for each replica.
Matrix of Capacity-Node Sihk refers to the storage of Host (Node).
Matrix of Size-Dataset Vj for storage space of Dataset.
Matrix of Threshold Td is compromise between quality of service, maximum
u, j
budget and minimum response time for each user.

Master Module: We can call it the brain since it has several components we start
with:
Solver: Since our concern is to decide the number of fundamental copies, for example,
the objectives of the occupant will be fulfilled while guaranteeing a benefit for the
cloud supplier we consider this Mathematical statement of the issue:
n m
Minimize  C
i 1 j 1
ij xij

m
subject to v
j 1
j xij  Capi , i  1, n

a
i 1
j xij  Td u , j , j  1, m, u  1, r

xij  N , i  1, n, j  1, m
(2)

Where:
Cij : replication cost and allocate Dataset j space in the datacenter i (Fi.h.k.j) //
whatever the node or the cluster
Xij: number replicas of dataset j in the datacenter i (Fi.h.k.j)
Vj: storage space (file size) of Dataset j (Fi.h.k.j) (file size)
Capi : Storage capacity Of DCi where
K H
Capi   Sihk (3)
k 1 h 1
5

aj: the Coefficient importance of Dataset j (Fi.h.k.j)


(4)
1
aj  n K H

 P
i 1 k 1 h 1
ihkj

Tdu,j: Threshold SLA fixed by provider and consumer u for the Dataset j (Fi.h.k.j) such
as the resolution is done by the simplex. Toward the end, we duplicate and erase to
achieve the ideal number of replications.
Repair: We need the framework to contemplate consistency and adaptation to
internal failure so it is desirable over purpose them by the rule of majority we begin to
dispatch a check demand for every one of the copies to acquire the right an incentive
by the lion's share and we make the update for every one of the information with
blunders and false esteem.
Springy. A flexibility of the assets it tends to be an expansion or lessening as
per the prominence of every datum, it implies copy a file 𝐹𝑖ℎ𝑘𝑗 if its popularity (recur-
rence of access to this information) is more prominent than a given edge Treplication and
delete a data 𝐹𝑖ℎ𝑘𝑗 if its popularity (recurrence of access to this information) is lower
than a given limit Terasure.
The popularity of each file is calculated by the following simple formula:
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑞𝑢𝑒𝑠𝑡 𝑓𝑜𝑟 𝑡ℎ𝑒 𝐹𝑖ℎ𝑘𝑗
𝑃𝑖ℎ𝑘𝑗 = (5)
𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑟𝑒𝑞𝑢𝑒𝑠𝑡

This last parameter is stored in the popularity matrix found in Global System Module.
SLA Violation: SLA Violation Utilize quality of service, most extreme spending
plan for every client and least reaction time as a limitation in a cost minimization
based calculation. in instances of severe infringement, the Replication instrument
triggers an expansion in assets.
DC Mapping: a module made to orientate the solicitation to the Concerned data
center since it contains a data structure pretty much all data centers (fast indexing).

Data-Center Module: We see that the data centers composed by cluster each cluster
contains nodes and a leader as well as two module leader election and Node mapping.
Leader Election: Once in a while the consummation of an assignment requires the
contribution of different examples of a similar cloud administration. In the event that
the administration purchaser summoning the cloud administration occurrences does
not have the important rationale to facilitate them, runtime special cases can happen
prompting information debasement and inability to finish the assignment.
One of the conjured cloud administration occurrences is assigned as the pioneer
hub, in charge of totalling the other cloud administration cases in an organized exer-
tion to finish the undertaking. These are the most accessible hubs in each Group in
Data Center by following Algorithms of choice (arranged rundown of the accessibility
of the hubs and takes the first).
Node Mapping: a module made to orientate the solicitation to the Concerned Node
since it contains a data structure pretty much all Node in the Data Center (quick order-
ing). It is connected with the Leaders.
6

4 Conclusion and Perspectives

In this paper, in arrange to increase data accessibility and decrease cost caused by
replication, we define an issue that decides the replication degree of each data object
so that the ask failure probability, consistency support cost and capacity cost are min-
imized in a cloud capacity using many of the different protocols and tools like sim-
plex and quorum. It takes into consideration a number of very important parameters.
For the perspectives, we attempt to apply this strategy in a simulation environment
such as Cloudsim [] to better flatten the results of the proposed approach.

References

[1] Assi, M., & Haraty, R. A.. A Survey of the Knapsack Problem. International Arab
Conference on Information Technology (ACIT), 2018.

[2] Abdenour Lazeb, Riad Mokadem, Ghalem Belalem: Economic Data Replication
Management in the Cloud. JERI 2019

[3] Limam, S., Mokadem, R., Belalem, G.:«Data replication strategy with satisfaction
of availability, performance and tenant budget requirements». Cluster Compu-
ting: The Journal of Networks, Software Tools and Applications, January 2019.

[4] M. Zhong, K. Shen, and J. Seiferas. Replication degree customization for high
availability. In Proc. of EuroSys, Glasgow, 2008.

[5] A. Cidon, S. Rumble, R. Stutsman, S. Katti, J. Ousterhout, M. Rosenblum. Copysets:


Reducing the frequency of data loss in cloud storage. Proc. of USEN!X ATC, 2013.

[6] A. Cidon, R. Escriva, S. Katti, M. Rosenblum, E. G. Sirer. Tiered replication: A cost-


effective alternative to full cluster gee-replication. Proc. of ATC, pp 31-43, 2015.

[7] A. Thomson and D. Abadi. Calvinfs: Consistent wan replication and scalable
metadata management for distributed file systems. Proc. of FAST, Santa Clara,
2015.

[8] R.N. Calheiros, R. Ranjan, A. Beloglazov, C. De Rose, R. Buyya. CloudSim: a Toolkit


for Modeling and Simulation of Cloud Computing Env. & Evaluation of Resource
Provisioning Algorithms. Software: Practice & Experience. Vol. 41, Issue 1, pp. 23-
50. (2010)

Вам также может понравиться