Techniques for Reducing the Down Time during Virtual Machine Migration: Resource Failure & Live Migration Circumstances

Virtual Machines need migration from the present host to an another appropriate one due to error in the present hardware or presently loaded scenario causing slowness of execution. Most of the situations fall into second category leading to a solution called live migration. Here, in live migration the virtual machine runs for sometime on the present host while its required data gets transferred to the destination. The prevalent techniques under live migration are pre copy and post copy having their relative merits and demerits. In this paper, we are presenting our work in terms of proposing several techniques for better designs, namingly, Remote DMA, parallel buses, NUMA, server consolidation and blade server and in case of OS virtualization virtual file system implementation is proposed. In case of hardware failure, provisioning of redundant hardware component is included in this paper. All these techniques are there to enhance the data transfer and sharing capacities beyond what is offered by pre and post and other prevalent designed techniques so as to reduce the downtime. INTRODUCTION: There are two types of virtualizations hardware virtualization and OS virtualization. In hardware virtualization, multiple virtual servers called as virtual machines are created; they perform the application execution on their virtual servers utilizing hypervisor software to make this happen on real hardware. Increasing the number of virtual machines on a host server leads to better utilization of server resources but sometime may cause overload and slowness of execution. In such situations any particular virtual machine identified as slow running one needs live migration to another host [1]. Virtual machine migration may also be required if any critical hardware component is diagnosed as faulty. In this case, migration in the form of shutting down the virtual machine on the present server and restarting it on the destination host takes place. Live migration has a necessity to transfer data in a faster way because the migration is gradual, for sometime the VM runs on the present host and data is transferred to destination host where it is moved eventually. There are two techniques to do live migration pre-copy where data belonging to the VM is transferred to the destination as a whole and then VM is started on destination. If data is large the migration is slow. In postcopy the most necessary data to start the VM is transferred, VM is started on host and rest of the data is transferred on demand basis. For faster data transfer as needed in either case more in pre-copy a modified technique of I/O transfer which is RDMA (remote direct memory access) is included in this paper, which is like DMA (direct memory access) but it is applicable when communication has to take place between two separate computers, one supplying the data and another receiving it. DMA or RDMA are much faster then the conventional I/O techniques of ‘Polling’ and ‘Interrupt driven I/O’ as they don’t involve CPU [2]. Use of parallel buses as multiple buses instead of a single bus as commonly used in RDMA is also included. Another technique for better data communication is the NUMA (non – uniform memory access) here different computers have different memory access times for accessing there local and other computers memory which is termed as remote access through some interconnection network. Consolidation is to substantially increase the efficient use of server resources which is advantageous but it may also increase the complexity of the configuration of data, applications, resources and servers that can be confusing for a normal user to share with. To overcome this problem, we have made use of server virtualization approach or blade server. In blade server, number of servers is placed in proximity with some common resources. In case of OS virtualization, for implementing virtual machines use of file system virtualization to handle the problem of heterogeneity of file system belonging to different guest OS is included in this paper. Rest of the paper is organized as: 1. Reasons of VM migration 2. Migration due to fault in hardware and remedies. 3. Live migration. 4. RDMA and server consolidation as proposed for live migration 5. Sharing of memory addressed with ccNUMA 6. File system differences in source and destination host and their solution with file system virtualization accomplished with pod (PrOcess Domain). Reasons of VM migration VM migration is majorly caused due to two main reasonswhen the hardware components prominently CPU goes faulty, secondly when CPU is slowing down due to load and not likely to fulfill the service level agreement of response time or other parameters pertaining to limitations on processing time or volume of processing. When hardware is erroneous or faulty How to detect the faulty state of hardware; The symptoms of a faulty CPU could be wrong result, different result for same data and instruction when computed repeatedly. Abhay Kothari et al, / (IJCSIT) International Journal of Computer Science and Information Technologies, Vol. 5 (5) , 2014, 6208-6212