Research On Key Technologies Of Scientific Workflow Security In Clouds

Cloud computing provides an efficient,convenient,flexible,and inexpensive computing mode,which has become a hot topic in academics and industry in the fields of computing,networking,and storage in the recent decade.Therefore,with the rapid development of cloud computing,more and more scientific computing tasks are executed in clouds.Since scientific computing tasks usually consist of many sub-tasks and intermediate data,scientific workflows are often adopted to model them so as to properly orchestrate,schedule,execute,and trace in a distributed computing environment.The flexibility of resource management mechanism in clouds enable the implementation of scientific workflows more flexible and economical,yet,the coexistence of multi-tenant in clouds brings serious security risks,such as interrupting the workflow execution,tampering with the workflow results and stealing the intermediate data.Existing researches on scientific workflow security mainly consider the abnormality caused by resource failures,ignoring the threats from malicious attacks to scientific workflows.Compared with traditional cloud business workflows,the security issues of cloud scientific workflows are particularly serious.First,the characteristics of large-scale computing of cloud scientific workflows require multiple VMs.In clouds,the more VMs,the larger the attack surface.Second,the long execution time of scientific workflows provides sufficient scanning and penetration time for attackers.Third,the scientific workflows usually involve some important scientific fields.Once the intermediate data is stolen or the results are tampered with,it will cause enormous losses.Therefore,in order to effectively improve the ability of the cloud scientific workflow system deterring malicious attacks,based on the active cyber defense(ACD),we study the key technologies of scientific workflow security in clouds.We propose to terminate attacks from three aspects based on Cyber Kill Chain: anti-reconnaissance,intrusion tolerance and status recovery,ensuring the reliability and credibility of scientific workflow execution.The main research includes:(1)CLOSURE,a CLoud scientific wOrkflow SchedUling algoRithm based on attack-defensE game model is proposed to prevent attackers from reconnoitering and penetrating the cloud scientific workflow execution environment.In CLOSURE,the attacker's strategy set includes various attacks exploiting on different operating system vulnerabilities,while different operating system distributions in virtual machine clusters performing scientific workflows are regarded as the defender's strategy set.This is an incomplete information game,where the attacker can obtain the information about the defender's strategies through network scan while the defender cannot obtain the information about the attacker's strategies.Therefore,we propose to dynamically change the defense strategy during the workflow execution to weaken the network scan.For both rational attackers and workflow users,the goal is to maximize the benefits,so it can be modeled as an attack-defense game problem.After that,the Nash Equilibrium in the attack-defense game model is calculated to obtain the probability distribution of the optimal hybrid strategies.In reference to the results,diverse virtual machines would be deployed to execute workflows.Furthermore,a task scheduling algorithm based on dynamic HEFT(Heterogeneous Earliest Finish Time)is proposed to accelerate the switching of defense strategies and improve the workflow execution efficiency.The experimental results show that compared with the existing algorithms,CLOSURE can reduce the attacker's benefits by around 15.23% and save the defender's time cost by around 7.86%.(2)MCSW,a Mimic Cloud Scientific Workflow system is proposed to guarantee the intrusion tolerance of cloud scientific workflow task.We propose MCSW to ensure the correct execution of the scientific workflow when attackers have successfully penetrated into the cloud environment.Mimic defense theory mainly focuses on heterogeneity,redundancy and dynamics.For heterogeneity,diversified virtual machines are adopted to build robust system architectures and are quantified with the number of common vulnerabilities among different operating systems.For redundancy,each workflow sub-task is executed by multiple virtual machines simultaneously to enhance the reliability.A lagged decision mechanism is proposed to check the intermediate data without interrupting the workflow execution and evaluate its confidence.For dynamics,we propose to periodically recycle and generate new virtual machines,eliminating latent threats and cleaning the workflow execution environment.In addition,a confidence-based intermediate data backup mechanism is proposed to store intermediate data with the confidence of 1.When there are idle resources in the system,the stored intermediate data can be used to re-execute the sub-tasks with low confidence.The experiment first uses Matlab to perform simulation test for the system security,then uses WorkflowSim to test the system performance.Finally,OpenStack is used to build a small prototype system and conduct actual network attack for security test.The experimental results show that MCSW can effectively prevent the attacker from interrupting and tampering with the scientific workflow.(3)ACISO,Availability Confidentiality and Integrity Strategy Optimization method is proposed to achieve intrusion tolerance of cloud scientific workflow data.A scientific workflow consists of multiple sub-tasks,each sub-task produces intermediate data as the input for subsequent sub-task execution.The correct execution of scientific workflows relies on the security of the intermediate data that is frequently transferred between virtual machines during the workflow execution.In the multi-tenant cloud,the intermediate data has three attributes: availability,confidentiality and integrity.If the intermediate data is lost,stolen or maliciously tampered with,these attributes are destroyed,resulting in workflow interruptions,secret information leakage and incorrect workflow execution results.To address this issue,ACISO is presented.In ACISO,using erasure codes with different parameters,different encryption algorithms and hash functions to construct availability,confidentiality and integrity strategy pool,respectively.Then,we present a security strategy optimal allocation model SSOA,aiming at maximizing the overall intermediate data security strength under the constraints of workflow makespan and storage overhead.Normally,a scientific workflow contains lots of intermediate data,so solving this model is an NP hard problem.Therefore,we propose a heuristic algorithm to solve SSOA.The simulation results show that ACISO can effectively prevent the attacker from destroying,stealing and tampering with the intermediate data of the scientific workflow.(4)MSTI,a Multi-Strategy cloud scientific workflow protecting method based on Task Importance is proposed to rapidly resume cloud scientific workflows from abnormal states.Resource failures and network attacks lead to the abnormality of cloud scientific workflow.In order to rapidly resume cloud scientific workflows from abnormal state,we propose MSTI,which combines the advantages of both task redundancy and checkpoint backtracking,respectively.MSTI first analyzes the topology of the workflow and proves that the importance of different sub-tasks to the workflow makespan varies.Therefore,we present the importance ordering method of workflow sub-tasks and divide workflow sub-tasks into three categories: sub-tasks with high importance,medium importance and low importance.For the sub-tasks with high importance,we use task redundancy to generate multiple replicas and execute them in different virtual machines.For the sub-tasks with medium importance,all input data is stored as a checkpoint.When these sub-tasks are in abnormal states,checkpoint backtracking will activate the resuming process.Last,we take no protection for the sub-tasks with low importance.Moreover,in order to further improve the workflow execution efficiency,a virtual machine allocation algorithm based on improved HEFT is proposed,which fully considers the diverse task dependencies.The experimental results demonstrate that MSTI can achieve rapid workflow abnormal state resuming and reduce the workflow makespan under abnormal conditions.This paper relies on the National Natural Science Foundation Project “research on the basic theory of cyberspace mimic defense”.The research results will provide support for the key technologies of the mimic defense and expand the application of mimic defense technology in the field of cloud scientific workflows.