So Should Your Replication Solution Be
Server virtualization changed the way the server infrastructure is deployed, managed and maintained. It also changed the way servers were acquired. No longer was the user locked into a particular set of server hardware for their data center to maintain consistency, the hypervisor became that consistency. Server manufacturers are now forced to create compelling server solutions based on performance, reliability and price.
In the same way that the hypervisor is revolutionizing the way servers are acquired, the latest releases of VMware and Citrix may be signaling the end of the proprietary array. Server virtualization platforms are now able to provide much of the storage services functionality that only the storage vendors used to provide. This includes capabilities like thin provisioning, snapshots, clones and even automated movement between different storage tiers.
The result is that now, thanks to these advancements in the hypervisor, both server selection and storage selection can be based on what is the ‘best of breed’ solution for the data center, or even the specific task at hand. The concept of a mixed, open storage environment all driven by the hypervisor is now reality. The next step in this metamorphosis is leveraging these different storage platforms to deliver a consistent disaster recovery strategy.
Replication Puts The Lock Back On
The challenge to making this work is replicating the data, either between storage systems or between data centers. While most storage systems have a replication feature, almost all require that the replication be to another storage system from the same manufacturer and even to the same model of storage system.
This restriction runs counter to the mixed vendor, best of breed storage strategy that’s being driven by the hypervisor. It requires that each storage system managed locally by the hypervisor would also be required at the disaster recovery site. There is also the challenge of managing multiple replication software products from multiple vendors at the same time. Monitoring this critical task would become the very time consuming process of checking in with each console to make sure that replication jobs are working correctly.
There’s also the problem of buying the secondary hardware. What could be considered best of breed for primary storage is not always the best solution for DR. In many cases the organization may want the ability to replicate all critical data to a single cost effective, high capacity but moderatly performing storage system in the DR site, regardless of the brand of the source disk.
Another challenge with storage system based replication is that it’s often not granular replication below a volume level. This is particularly problematic when it comes to the virtualized server environment, since servers are stored as disk images which are individual files, often on a single LUN. Not all virtual machines are created equal and you may want to select specific VMs (files) to be replicated.
The Requirement for Hardware Independent Replication
The move to a hypervisor that controls the storage infrastructure and allows best of breed storage hardware selection is also going to drive the need for software vendors to address certain weaknesses within the hypervisor storage capabilities or even add storage functionality that’s not present in the hypervisor at all. An excellent example of this is software based replication like that offered by Vision Solutions.
Software based replication provides a hardware independent replication capability to the hypervisor. By using software within the virtual infrastructure only a single replication product needs to be learned and managed which reduces cost. More importantly it allows a mixed vendor replication strategy where multiple storage systems from different vendors can replicate to a single (or multiple) disaster recovery target(s). It also enables specific “high value” VMs to be replicated while others are not, since the software has a more granular understanding of what it’s replicating.
Including the Physical Into Virtual Infrastructure DR
Another important capability of software based disaster recovery solutions is that they can operate on physical, stand-alone servers that have not yet been virtualized. Their target can be a virtual machine which can be replicated to the virtual infrastructure at the DR site. Not only does this further simplify and reduce the cost of the DR site, it also provides a safe process for testing the physical server in a virtual world and potentially a final migration path to a virtualized infrastructure.
Data Copy is Only One Part
Another advantage of software based replication solutions is that they reside in the virtual machine itself. This allows them to have a full understanding of what is actually happening inside the guest and its application. Storage based replication only sees blocks of data, nothing more. It has no understanding of the VM, its operating system or the application it’s running.
Being inside the VM lets the software based solution take action on VM failure or, more importantly, an application failure. Hosts can of course stay up and running even when a VM fails and a VM can be online even though an application has hung. Storage replication has no or limited visibility into the VM. By being inside the VM the software based replication tool can trigger a failure to a standby VM almost instantly, no matter which part of the infrastructure stack has failed.
Replication can be used in a disaster recovery strategy, but it can also be used locally. One of the functions unavailable from the hypervisors yet today is the ability to mirror a VM image to different storage systems. While almost all storage systems offer mirroring as a protection strategy they can only do that mirroring between their own storage systems and certainly not to a storage system from another manufacturer.
The problem with this internal storage system mirroring is that it does not protect the VM from a complete storage system failure. While most systems have redundancy to protect from this sort of error it can still happen, especially during a controller software upgrade. The ramifications in a virtual server environment of all the VMs on a storage system failing at the same time can be catastrophic and may actually cause an unwanted site disaster recovery declaration forcing an unneeded move to the secondary site.
Software based replication can provide a near-mirror via asynchronous replication to a secondary disk target. This allows for VMs to be protected on a second independent storage system without adversely impacting performance. The asynchronous nature of the replication also means that the secondary disk target does not need to have the same performance capabilities as the primary, again saving money. It also means less impact on the performance of the VM itself.
Hardware independent software replication completes the puzzle for users looking to break the chains of storage vendor lock in. It allows the customer to select the best of breed storage system for their data center, for the specific task at hand and even the location of the system. It also enables physical server inclusion into a virtual DR strategy. Finally, it provides more than just data movement. HA versions of replication software provide operating system- and application-awareness that storage hardware based solutions simply do not have.