Ian Masters examines how protecting multiple production applications with one disaster recovery server provides an enterprise-class high availability and disaster recovery solution.
For most businesses, ‘protecting the data’ – getting it out of the building – is the lynchpin that all other disaster recovery aspects rely on. But often, other measures need to be considered in order to ameliorate data replication and provide a robust business continuity solution. One of these is to transform the existing IT infrastructure into a virtual infrastructure, which provides organisations with the key to greater flexibility as their business needs change.
Virtual infrastructure works by providing a layer of abstraction between computing, storage and networking hardware, and the software that it runs on. Within this environment, users see resources as dedicated solely to them while the administrator is able to manage and optimise those resources globally across the enterprise. IT organisations are then able to provision new services and change the amount of resources dedicated to a software service, treating the data centre as a single pool of processing, storage and networking power.
The end result is achieved through using server virtualisation technology, which allows one physical platform to run multiple virtual machines simultaneously. Each virtual machine has its own processor(s), memory, disks and network interfaces, and the operating system running in each virtual machine is referred to as a ‘guest’. Functionally, each virtual machine is autonomous and unaware that the hardware is being shared, enabling multiple servers (even of different operating systems or versions) to run on just one hardware platform.
This means, for example, that a single machine is able to support three different virtual machines such as:
- Windows Server 2003 and Exchange Server 2003
- Windows Server 2000 and SQL 2000
- A domain controller (DC)
The third machine in the example above, a virtualised DC, is particularly popular with large distributed branch offices that require a local DC but do not wish to pay for the physical asset. It is also useful for disaster recovery sites for the same reason. The key to these solutions is in the complete autonomy between virtual machines and their respective operating systems and applications.
One of the challenges, however, of protecting virtual machines, versus physical servers, is the storage method used within virtual machines – namely virtual disks and raw disks. Where a group of related virtual disk files exists on the server, it presents itself as storage devices to the virtual machine operating system. This causes encapsulation of the entire file set, which provides benefits in its autonomy, but also complexity in its protection. Alternatively, the raw disk method allows for virtual machines to access a physical partition on the host server directly.
The most common customer need is to protect unique and sometimes conflicting applications on multiple servers to a single recovery server. While the solution must support a ‘many-to-one’ model for data replication, certain applications cannot co-exist on the same server due to conflicts or operational requirements. By leveraging virtual machines on the target platform, each failed source can be protected and made available by an independent virtual machine target.
Some leading solutions enable replication within a guest OS or on the host OS. Replicating for a guest OS is just like replicating from a traditional physical server, so no special considerations are necessary. This offers the flexibility of providing file level selection for replication of individual files with the virtual disks belonging to a virtual machine. Replicating from the guest means that disk-writes can be captured at the host’s file system layer, while all other applications operate above the actual file system layer. This allows transparent and reliable data replication to a separate target server running another copy, sending only real-time byte-level changes across any IP connection.
With the addition of virtual servers such as VMware GSX, users have the choice of replicating data from within the guest operating systems or from the host operating system. Used in conjunction with byte-level data replication technology (as above), this offers the capability to replicate non-Windows OS virtual machines as well as providing complete failover capability of the guest OS.
In order to provide fault-tolerance across any IP network (LAN or WAN), back-up solutions should be able to replicate the bytes of data that change to another windows server somewhere in the enterprise. There is a number of ways of combining the technology to achieve these outcomes:
1. Installed within the guest OS, it enables replications on any part of the files or directories within the virtual machine disks, while the target server has no idea that the source server is a virtual machine and not ‘real hardware’. Upon failover, the target server would assume the name, IP, shares and services of the failed guest OS.
2. The most common solution is, however, installation on the host OS to maintain replicas of all the virtual disk files on the target server. The target server would monitor the guest OS for failure, but instead of assuming its properties it would simply invoke a script that would initialise the virtual server on the target server. The virtual server would then use the replicated virtual disk files that hold the actual configuration and data from the guest OS.
3. Alternatively, while the guest OS system partition is a virtual disk, the actual data drives may be allocated partitions of the host OS. In this case, the back-up would be installed on the host OS of the source platform and replicate both data from the allocated partitions as well as any virtual disk files being used by the guest OS(s). In the event of a failure on the source machine, the target server would invoke the virtual server on the target. Once this server starts, the virtual machine could be started using the replicated virtual disk file(s) and data replicated from the allocated partitions on the source server.
As the target server might be used for the failover of multiple servers, care would have to be taken to prevent compatibility issues. During failover, the target server would need to be configured to start virtual machines corresponding to each source server.
Protecting multiple production applications (or clients) using one disaster recovery server is possible through a combination of real-time data replication with virtual infrastructure architecture. These technologies combine to offer leading-edge disaster recovery solutions, with proven flexibility to deliver enterprise-class protection anytime, anyplace, anywhere.
Ian Master is sales director at Double-Take Software.