Research On High Availability Of Cloud Computing For Video Surveillance Analysis

As the distributed computing system especially cloud computing system has been widely used, system scale continues to grow. The number of system components enlarges and the type of system components varies, and the interaction between them continues to increase. Failure occurrence becomes the norm rather than exception in today’s distributed computing system. As the cloud computing system for video surveillance analysis runs around the clock, failure occurrence or system maintenance could possibly lead to system crash, and therefore heavily impact system reliability and availability. To solove the above problems, studys the high availability of cloud platform maintenance, failure prediction and analysis mechanism, failure-aware virtual machine configuration and high available scheduling optimization strategy for the vedio surveillance analysis, can effectively improves system availability.For high availability of cloud platform maintenance problem, maintenance for vedio surveillance analysis system requires the non-stop of the ongoing system services, and the low cost and high availability throughout the maintenance process as the vedio surveillance system runs24/7. Dependencies between components are easy to cause the incompletion of the maintance or system error. To solove the issue, dependency-aware maintenance for cloud computing supports multiple granularities (service-, container-, and node-level) maintenance to reduce the effects from environment dependencies. In addition, dependency-aware maintenance mechanism manages a dependency map that is initially provided by automatic detector or system administrators. When a maintenance request arrives, dependency-aware maintenance mechanism recognizes the related dependencies using dependency map to generate an optimized maintenance solution which can reduce the dependency effects. Further more, dependency-aware maintenance mechanism provides the session management for maintenance to avoid the possible failures brought by hierarchal dependencies.The increasing number of nodes and the existence of component heterogeneity in vedio surveillance analysis cloud make the occurence of the system failure easier, failure on a single node may cause a chain reaction of its associate nodes or even the entire system, and thereafter increae the probability of system failure. For system which needs to run persistently and supply continues services, system crash caused by some reason may lead to serious consequences. To address this issue, analyzes the failure correlation to effectively predict failure occurrence, guide the job scheduling and virtual machine configuration based on predicition results are studied to improve the system availibility. For failure-aware virtual machine configuration, a proactive failure prediction framework which exploits the temporal and spatial correlations among failure events is proposed. By clustering signatures in the time and space domains, the proactive failure prediction framework explores the temporal and spatial correlations among failure occurrences. Node allocation information is utilized to refine the predicted correlations. Experimental results of offline and online prediction on production coalition systems present the feasibility of applying failure prediction to autonomic management for high-availability network computing.Based on failure precition, a proactive failure-aware mechanism is proposed to guide the virtual machine configuration. Failure prediction technique is introduced as a proactive methodology to facilitate the job scheduling and virtual machine reconfiguration and mitigate the potential failure impact on system reliability and productivity. The failure-aware mechanism takes the performance statues with reliability status into account when making the node selection decision. The experiments results show that the proposed strategy enhances system productivity and availability significantly.For specific application--vedio surveillance analysis, a heterogeneous terminal scheduling model automatically realizes effective optimize of intelligent analysis tasks from the infrastructure point of view under limited network bandwidth and computing power circumstance, to maximize the overall completion rate of video analysis tasks. Meanwhile, as a public resource, vedio surveillance platform needs to run multiple analysis tasks that usually complicate dependencies between them. Based on the understanding of the dependency between tasks, this model investigates the effect of dependency characristic on fault-tolerant intelligent terminal scheduling model and furthermore improves the scheduling mechanism. Experimental results show that, with respect to traditional task processing mechanism, heterogeneous terminal optimal scheduling model can effectively raise the completion rate of global analysis task and ensure the efficiency and reliability of large scale vedio surveillance network. Under different terminal processing ability and pressure, fault-tolerant scheduling mechanism exceeds15%-30%over traditional mechanisms, better adaptive to the change of system heterogeneity compared with traditional mechanisms.