Existing Fault Tolerance Techniques In Cloud Computing

Published Date: 02 Nov 2017

Abstract

Cloud computing is emerging as a new paradigm of large scale distributed computing embraces cyber infrastructure and builds upon on the concept in virtualization, grid computing, utility computing, networking, web services and software services to implement a service oriented architecture for reducing information technology overhead for the end-user for provide great flexibility and reduced total cost of ownership and all above on-demand services to a shared pool of computing resources.

The ability of a system to respond gracefully to an unexpected hardware or software failure is known as fault tolerance.so fault tolerance is an important aspect in cloud computing. This paper aims to provide a better understanding of fault tolerance techniques used for fault tolerance in cloud environments along with some existing model and further compare them on various parameters.

Introduction:

"Cloud computing"-the result of evolution of on demand service in computing paradigms of large scale distributed computing embraces cyber infrastructure and builds upon on the concept in virtualization, grid computing, utility computing, networking, web services and software services to implement a service oriented architecture for reducing information technology overhead for the end-user for provide great flexibility and reduced total cost of ownership to a shared pool of computing resources. It has the capacity to yoke the internet and wide area network to use the resources that are available remotely there by to provide cost efficient solution on pay per use basis [1][2].Due to the rapid exponential growth of cloud computing the need of fault tolerance in cloud is an key factor for concern.

Fault tolerance bear-on with all the inevitably techniques to enable robustness and dependability .The main benefits of implementing fault tolerance in cloud computing include failure recovery, lower cost, improved performance metrics [3]. Robustness leads to the property to providing of a correct service in an adverse situation arising due to an uncertain system environment [4]. Dependability is related to some QOS aspects provided by the system, it includes the attributes like reliability and availability [5].

The motivation of the survey of existing fault tolerance techniques and models in cloud computing is to encourage researcher to contribute in developing more efficient algorithm. This paper is organized as follows: Section II discusses about various aspect of faults and the need of fault tolerance in cloud computing, Section III shows the study and analysis of the existing fault tolerance techniques in cloud computing there implementation on various models and comparison among the models, Section IV identifies the metrics considered in the existing fault tolerance models and carries out the comparison between them based on those identified metrics and Section V concludes the paper.

Fault taxonomy and need of fault tolerance in cloud computing:

The main aim for any system is to achieve dependability. Fault tolerance helps us to do so. Based on fault tolerance policies and techniques we can classify this technique into 2 types proactive and reactive. The Proactive fault tolerance policy is to avoid recovery from fault, errors and failure by predicting them and proactively replace the suspected component means detect the problem before it actually come. Reactive fault tolerance policies reduce the effort of failures when the failure effectively occurs. This technique provides robustness to a system. For the both techniques of fault tolerance have two sub-techniques error processing and fault treatment. Error processing aims at removing errors from the computational state. Fault treatment aims at preventing faults from being re-activated [4] [5].

Now for getting dependability the system bespeak for the combined utilization of a set of methods which are classified into fault prevention, fault removal, fault tolerance, and fault forecasting. Fault prevention and fault removal both encompassed in the notion of fault avoidance which aims to produce a system containing as few faults as possible. It encompasses Fault prevention which deals with the prevention of faults and fault removal deals with to reduce the number and severity of faults. Fault-forecasting aims to estimate the present number, the future indication and the likely consequences of faults [4] [6].

Fault tolerance is carried out by error processing which have two constituent phases. The phases are "effective error processing" which aimed at bringing the effective error back to a latent state, if possible before occurrence of a failure and "latent error processing" aimed at ensuring that the error does not become effective again. Effective error processing may take two forms. Error -recovery and error-compensation. Error recovery can be made on two forms backward error recovery and forward error recovery. Error processing generally completed by maintenance which can be put into two classes corrective maintenance and preventive maintenance [6].

Existing Fault tolerance Techniques in Cloud Computing:

Following fault tolerance techniques are currently prevalent in clouds [3] [4] [6] [7] [8]:-

Check pointing– It is an efficient task level fault tolerance technique for long running and big applications .In this scenario after doing every change in system a check pointing is done. When a task fails, rather than from the beginning it is allowed to be restarted that job from the recently checked pointed state.

Job Migration –Some time it happened that due to some reason a job can- not be completely executed on a particular machine. At the time of failure of any task, task can be migrated to another machine. Using HA-Proxy job migration can be implemented.

Replication-Replication means copy. Various tasks are replicated and they are run on different resources, for the successful execution and for getting the desired result. Using tools like HA-Proxy, Hadoop and AmazonEc2 etc replication can be implemented.

Self- Healing- A big task can divided into parts .This Multiplication is done for better performance. When various instances of an application are running on various virtual machines, it automatically handles failure of application instances.

Safety-bag checks: In this case the blocking of commands is done which are not meeting the safety properties [4].

S-Guard- It is less turbulent to normal stream processing. S-Guard is based on rollback recovery. S-Guard can be implemented in HADOOP, Amazon EC2.

Retry- In this case we implement a task again and gain. It is the simplest technique that retries the failed task on the same resource.

Task Resubmission- A job may fail now whenever a failed task is detected, In this case at runtime the task is resubmitted either to the same or to a different resource for execution.

Timing check: This is done by watch dog. This is a supervision technique with time of critical function [4].

Rescue workflow- This technique allows the workflow to persist until it becomes unimaginable to move forward without catering the failed task.

Software Rejuvenation-It is a technique that designs the system for periodic reboots. It restarts the system with clean state and helps to fresh start.

Preemptive Migration- Preemptive Migration count on a feedback-loop control mechanism. The application is constantly monitored and analyzed.  

Masking: After employment of error recovery the new state needs to be identified as a transformed state. Now if this process applied systematically even in the absence of effective error provide the user error masking [6].

Reconfiguration: In this procedure we eliminate the faulty component from the system.

Resource Co-allocation: This is the process of allocating resources for further execution of task.

User specific (defined) exception handling- In this case user define the particular treatment of a task failure.

Now there are several models are implemented based on these types of techniques. They are

"AFTRC" a fault tolerance model for real time cloud computing based on the fact that a real time system can take advantage the computing capacity, and scalable virtualized environment of cloud computing for better implement of real time application. In this proposed model the system tolerates the fault proactively and makes the diction on the basis of reliability of the processing nodes [9].

"LLFT" is a propose model which contains a low latency fault tolerance (LLFT) middleware for providing fault tolerance for distributed applications deployed with in the cloud computing environment as a service offered by the owners of the cloud. This model is based on the fact that one of the main challenges of cloud computing is to ensure that the application which are running on the cloud without a hiatus in the service they provided to the user. This middleware replicates application by the using of semi-active replication or semi-passive replication process to protect the application against various types of faults [10].

"FTWS" is a proposed model which contains a fault tolerant work flow scheduling algorithm for providing fault tolerance by using replication and resubmission of tasks based on the priority of the tasks in a heuristic matric. This model is based on the fact that work flow is a set of tasks processed in some order based on data and control dependency. Scheduling the workflow included with the task failure consideration in a cloud environment is very challenging. FTWS replicates and schedule the tasks to meet the deadline [11].

"FTM" is a proposed model to overcome the limitation of existing methodologies of the on-demand service. To achieve the reliability and resilience they propose an innovative perspective on creating and managing fault tolerance .By this particular methodology user can specify and apply the desire level of fault tolerance without requiring any knowledge about its implementation. FTM architecture this can primarily be viewed as an assemblage of several web services components, each with a specific functionality [12].

"Candy" is a component base availability modeling frame work, which constructs a comprehensive availability model semi automatically from system specification describe by systems modeling language. This model is based on the fact that high availability assurance of cloud service is one of the main characteristic of cloud service and also one of the main critical and challenging issues for cloud service provider [13].

"Vega-warden" is a uniform user management system which supplies a global user space for different virtual infrastructure and application services in cloud computing environment. This model is constructed for virtual cluster base cloud computing environment to overcome the 2 problems: usability and security arise from sharing of infrastructure [14].

"FT-Cloud" is a component ranking based frame work and its architecture for building cloud application. FT-Cloud employs the component invocation structure and frequency for identify the component. There is an algorithm to automatically determine fault tolerance stately [15].

"Magi-Cube" a high reliable and low redundancy storage architecture for cloud computing. The build the system on the top of HDFS and use it as a storage system for file read /write and metadata management. They also built a file scripting and repair component to work in the back ground independently. This model based on the fact that high reliability and performance and low cost (space) are the 3 conflicting component of storage system. To provide these facilities to a particular model Magi cube is proposed [16].

Comparison among various models based on protection against the type of fault, and procedure

Model no

Model name

Protection against Type of fault

Applied procedure for tolerate the fault

M1

AFTRC

Reliability

1.Delete node depending on their reliability

2.back word recovery with the help of check pointing

M2

LLFT

Crash-cost, trimming fault

Replication.

M3

FTWS

Dead line of work flow

Replication and resubmission of jobs

M4

FTM

Reliability, availability,

on demand service

Replication users application and in the case of replica failure use algorithm like gossip based protocol.

M5

CANDY

availability

1.It assemble the model components generated from IBD and STM according to allocation notation.

2.Then activity SNR is synchronized to system SRN by identifying the relationship between action in activity SNR and state transition in system SRN.

M6

VEGA-WARDEN

Usability, security, scaling

1. Two layer authentication and standard technical solution for the application.

M7

FT-CLOUD

Reliability, crash and value fault

1. Significant component is determined based on the ranking.

2. Optimal ft technique is determined.

M8

MAGI-CUBE

Performance, reliability,

low storage cost

1. Source file is encoded in then splits to save as a cluster.

2. File recovery procedure is triggered is the original file is lost.

Metrics for fault tolerance in cloud:

The existing fault tolerance technique in cloud computing consider various parameter. The parameters are like there type of fault tolerance (proactive, reactive and adaptive), performance, response-time, scalability, throughput, reliability, availability, usability, security and associated over-head.

Proactive fault tolerance: The Proactive fault tolerance policy is to avoid recovery from fault, errors and failure by predicting them and proactively replace the suspected component means detect the problem before it actually come.

Reactive fault tolerance: Reactive fault tolerance policies reduce the effort of failures when the failure effectively occurs. This technique provides robustness to a system.

Adaptive: All the procedure done automatically according to the situation.

Performance– This is used to check the efficiency of the system. It has to be improved at a reasonable cost e.g. reduce response time while keeping acceptable delays.

Response Time - is the amount of time taken to respond by a particular algorithm. This parameter should be minimized.

Scalability – This is the ability of an algorithm to perform fault tolerance for a system with any finite number of nodes. This metric should be improved.

Throughput –This is used to calculate the no. of tasks whose execution has been completed. It should be high to improve the performance of the system.

Reliability: This aspect aims to give correct or acceptable result within a time bounded environment.

Availability: The probability that an item will operate satisfactorily at a given point with in time used under stated conditions. Availability of a system is typically measured as a factor of its reliability as reliability increases, so does availability.

Usability: The extent to which a product can be used by a user to achieve goals with effectiveness, efficiency, and satisfaction.

Overhead Associated: determines the amount of overhead involved while implementing a fault tolerance algorithm. It is composed of overhead due to movement of tasks, inter-processor and inter-process communication. This should be minimized so that a fault tolerance technique can work efficiently

Cost effectiveness: Here the cost is only defined as a monitorial cost.

Comparison among various models based on the metrics element:

Model number

M1

M2

M3

M4

M5

M6

M7

M8

Proactive(y/n)

y

n

n

n

n

y

y

n

reactive(y/n)

n

y

y

y

y

y

n

y

adaptive(y/n)

y

n

n

n

y

n

y

y

Performance(h/l/a)

h

h

a

a

a

h

h

h

Response time(h/l/a)

a

a

a

a

a

h

a

a

Scalability(h/l/a)

h

h

l

l

h

h

h

h

Through put(h/l/a)

h

a

l

a

h

a

a

h

Reliability(h/l/a)

h

h

a

a

h

h

h

h

Availability(h/l/a)

h

h

a

h

h

h

a

a

Usability(h/l/a)

h

a

a

a

a

h

h

h

Overhead Associated(h/l/a)

a

l

h

l

l

h

h

a

Cost effectiveness(h/l/a)

a

l

h

l

l

l

h

h

(y=yes, n=no, h=high, l=low, a=average)

Conclusion:

Fault tolerance methods come into play the moment a fault enters the system boundaries. So theoretically fault tolerance techniques are used to predict these failures and take an appropriate action before failures actually occur .This paper discuss about the fault taxonomy and need of fault tolerance covering with its various techniques for implementing fault tolerance. Various proposed models for fault tolerance are discussed and compared on the basis of Metrics for fault tolerance in cloud.

In the present scenario, there are number of fault tolerance models which provides different fault tolerance mechanisms to enhance the system. But still there are number of challenges which need some concern for every frame work or model. There are some drawback.no one of them can full fill the all aspects of faults. So there is a possibility overcome the drawbacks of all previous models and try to make a compact model which will cover maximum fault tolerance aspect.

Acknowledgement:

In all humility and with much fervor, I owe my deep and sincere gratitude to my revered guide Assistance Prof. Harsh Ahluwalia of CSE, LPU, Jalandhar India for his enlightened guidance, continuous encouragement, estimated supervision and paternal affection throughout the period of this research. Key improvements in the proposed research work would not be possible without the valuable suggestion and feedback of my guide.