As a seasoned IT backup administrator recently said on a Storage Swiss webinar, “Avoid hearing the footsteps…”. This IT pro was referring to the unpleasant situation when folks from the C level suite start milling around outside your office waiting for you to recover an application or data set that so that an application can return to operation. Now more than ever, end users expect business application services to be up all the time. And if they do go down, the expectation is they will be back near instantaneously. Such is life for the beleaguered IT admin in today’s “always on” Internet culture.
Forewarned is Forearmed
To be prepared for the unexpected, IT planners need to ensure that their backup environment can work hand-in-glove with the new realities of virtualized infrastructure. The fact is, as more virtual machines are packed on to hypervisor hosts, the increased backup I/O workloads begin to overwhelm most disk backup appliance systems.
What’s more, technologies like Change Block Tracking (CBT), which were designed to speed up backup and recovery times, ironically have increased the appetite for application owners to take even more frequent backups; resulting in still more I/O contention and performance bottlenecks on backup appliances.
To further complicate matters, some virtual server backup application solutions now allow virtual machines to boot directly off a backup image. Referred to as “recovery in place”, this has become a popular feature with virtual administrators. The only problem is, most backup appliances are configured and tuned to handle backup I/O workloads, not real-time application I/O workloads.
Disk Architecture Matters
Application quality-of-service (QoS) could fall through the floor if, for example, a performance sensitive application goes from running at solid-state disk speeds to the snail like performance that would be rendered by a backup appliance. Add the latency of deduplication workloads into the mix and doing a recovery in place on a backup appliance could be an absolute non-starter.
This begs the question, are disk backup appliances even a viable option for backing up large virtualized environments? As Storage Switzerland and ExaGrid discussed in a recent webinar, it’s all about the disk backup architecture.
Scale-Up to Scale-Out to GRID
Many backup appliances on the market utilize traditional “scale-up” storage architecture designs. These systems are primarily dependent on the CPU processors and memory in a single system to manage all the backup I/O workloads. These systems perform well when they are initially deployed but over time as workloads increase, they inevitably have to be replaced by bigger appliances with yet more CPU processor and memory capabilities.
Today’s increasingly virtualized data center can benefit from backup systems which can scale-out to support multiple independent units of backup processing power that collectively work together to meet the ongoing need for storage performance and capacity. In this manner, forklift upgrades can be completely avoided.
Interestingly, these storage grid architectures are already being utilized to meet the application I/O and storage capacity demands in primary storage environments today. But arguably, they are needed even more to handle the strenuous I/O requirements of backup workloads and data deduplication processes.
Hybrid Backup GRID
In fact, these grid based storage architectures can be used to even greater effect when they provide a hybrid configuration of deduplicated disk storage and non-deduplicated (or native) disk storage capacity. The deduplicated partition could be used to store multiple versions of backup images efficiently on disk while the native disk partition could hold a week’s worth of backup images to support fast local backups and restores.
By sending backup data streams directly to a native disk partition, backup jobs can run much faster since they are unhindered by the latency introduced by data deduplication processes. Likewise, data restore requests, particularly if an entire VM needs to be recovered, operate much faster from a native disk partition.
This hybrid approach also makes technologies like recovery in place a more viable option for backup and virtual administrators, since VMs could boot off a backup image on non-deduplicated storage and perform reasonably well.
To learn more, please go to the on-demand version of the Storage Switzerland and ExaGrid webcast “How Server Virtualization Is Breaking Disk Backup”. Because the only footsteps you want to hear, are your own walking briskly to your car at the end of the business day.
ExaGrid is a client of Storage Switzerland