Wednesday, November 11, 2009

VM Backup isn’t just ‘Backup’

The virtualized environment is very different than the traditional ‘brick and mortar’ server environment it was created to improve. Physical servers are consolidated into a few physical hosts to increase hardware utilization and improve flexibility. The number of virtual machines (VM) frequently ends up greater than the original physical server count it replaced. All the software, configuration information and data for each VM is typically  captured into a few image files, effectively encapsulating the server, applications and data.

The traditional backup process is based around the backup server. This device stores the information about which files are backed up, controls each backup job by contacting the client servers and physically pulls the backed up data to itself, before sending it to the storage device(s). It’s focused on the backup of specific data objects, and getting that done in a particular window of time. When a failure occurs, traditional backup simply replaces these lost data objects. It restores data and configuration files, but requires you to reassemble the server, application and then replace the data in the event of a full server loss

As the infrastructure has gotten larger and more complex, traditional backup has required specific software modules with a detailed awareness and understanding of applications and hardware in order to effectively access and move data from source to backup target. Examples of these are SQL, Exchange and NDMP modules. In addition, many applications are backed up using custom scripts and local utilities, outside of an enterprise backup system.

With its image-based architecture, virtual machines can be backed up without complex integration to applications or hardware. VMware developed the VMware Consolidated Backup (VCB) module to better enable various backup solutions to leverage this unique architecture. In practice it’s not been as widely adopted as one might think, due to some integration and operational challenges. While the concept was certainly sound, its implementation has generally proved to be more difficult.

The virtual environment has needs that may be better served by dedicated virtual machine backup solutions than by the traditional backup process. These dedicated solutions, like those from Vizioncore, focus on restarting the compute process, not just backing up and restoring specific files. In addition, they don’t require complex configurations and licensing, and don’t need modules for specific applications or hardware platforms in order to capture a backup. They don’t have software agents for each server instance and don’t pass all data through a backup server. What they do have is an understanding of the virtual environment and how to safely backup the host (ESX) server and the throttling required, based upon what hosts  the VMs are running on.

This reduction in complexity can also mean a reduction in cost. Less backup software is needed, which means fewer software licenses, less time spent implementing and configuring the backup system and lower software support fees. Also, reduced complexity usually means less time required to learn the system, expand it and upgrade it. Finally, a simpler system can also mean fewer problems keeping it running.

Virtualization technology has abstracted the server architecture, from a data perspective, to a few image files. This abstraction has enabled the entire server instance to be represented by files, which also enables the server to be restored at the same time its data is recovered. Traditional backup still requires a failed server be restored and ‘stood up’, prior to its data being reloaded, before it can be returned to service. Dedicated VM backup solutions take advantage of this abstraction and can restore a server instance and its data in one step.

Dedicated VM backup solutions aren’t disruptive to the applications that control much of the structured data. They don’t require a complex software module or even an understanding of these applications. These tools take care of quiescing the applications by using VSS technology in order to capture the transactions in flight prior to taking a snapshot of the VM’s image.

Traditional backups typically send all data through the backup server before going to the storage target. Dedicated VM backup solutions that utilize a direct to target architecture bypass the backup server altogether and send data directly from the ESX server to the storage hardware. This enables much faster backup jobs, consumes less bandwidth on the network and eliminates bottlenecks at the backup server. Also, the backup server needs less CPU, memory and storage resources when it’s not handling the entire data set.

Dedicated VM backup solutions can also have some other features that save time and improve the backup process. Multiple, simultaneous backup jobs can be run without being limited by physical storage targets or backup server throughput. Since these solutions don’t treat each VM as a physical server, they need only handle the few image files that now represent the VMs. And, they don’t have to process the thousands of discrete files that VMs actually contain.

The advent of virtual machine technology has been a big advantage for IT. It’s allowed for the physical consolidation of server infrastructures with the commensurate hardware cost savings. It’s also produced real reductions in data center footprint and power usage. The technology that enables this paradigm shift has produced an environment that’s much different than the data center just a few years ago.

A virtual server environment is much different from the physical environment it replaced. A server OS, its data, configuration and applications have been abstracted into a few files. Backup products that understand the virtual environment can use this abstraction to simplify backup, but also to enable fast and easy restores of virtual servers themselves. Dedicated VM backup products can provide real reductions in the time and effort required to restore a server, replace its data and get those processes running again.

Eric Slack, Senior Analyst