Storage Performance, Failure & Data Loss, and Diagnosability – Unitrends Recovery-943: The Soul of a New Backup Appliance (Part 7)

Unitrends recently introduced the Unitrends Recovery-943. This appliance is the next generation flagship of the Unitrends Recovery-series of backup appliances. We began this series of blog posts by discussing why backup appliances are experiencing such rapid demand and adoption, and continued the discussion by addressing the challenges facing enterprise-class data protection backup appliance vendors. The following blog post explored the Recovery-943 architecture and the most recent blog post addressed the specific responses that the designers of the Recovery-943 had to each of the challenges discussed.

Storage Performance: Industry’s First 3-Tiered Backup Appliance

Creating insanely fast storage or highly fault resilient storage or very affordable storage isn’t very difficult; however, creating a storage architecture that takes the best characteristics of each is profoundly challenging. And yet that’s what our designers and engineers have done with the storage architecture of the Recovery-943.

The storage architecture of the Recovery-943 consists of the following primary storage components:

• A pair of mirrored RAID-1 solid state drives that consist of the hardened, performance-optimized, and space-optimized RecoveryOS™ Linux-based operating system and a small amount of metadata that is optimized for random I/O performance.
• A pair of mirrored RAID-1 deeply cached high-speed rotational drives that consist of the core metadata associated with backup, archive, instant recovery, and disaster recovery operations. This storage pool is optimized for random I/O performance.
• 36 affordable and high-speed rotational drives that contain the raw backup information. These 36 drives are arranged as a pair of striped RAID-6 arrays consisting of 16 data drives and 2 parity drives. This configuration is commonly referred to as a RAID-60 and is highly fault resilient. The creation of two logical spindles that are striped means that this configuration is not only more fault resilient than a single large RAID-6 but has better read and write performance characteristics as well. See the next section for more information concerning RAID-60s.

This independent three-tier architecture allows significantly less seek time and rotational latency for the raw backup data write and read operations due to not only the striped RAID-6’s in the RAID-60 but also because random I/O is independently distributed among the deeply cached high-speed rotational drives and the solid state devices.

Failure and Data Loss: RAID-60

Unitrends has historically used a RAID-6 for the core backup array of drives in its 2U and 3U Recovery-series appliances. As our designers and engineers were developing the Recovery-943, it became clear than for the 36 drives that constituted the backup array that we needed a higher degree of fault tolerance than a RAID-6 provided. In other words, on a device with this many disk drives and of this class, it became apparent that we needed to drive MTTDL (Mean Time To Data Loss) much higher than we could on a simpler RAID-6 architecture and implementation.

Unitrends used in its Recovery-943 a tiered RAID-1/RAID-1/RAID-60 storage architecture in order to minimize MTTDL (Mean Time To Data Loss.) While we use a RAID-1/RAID-6 storage architecture in our 2U and 3U based systems, the 36 drives that are used primarily for backup is large enough that we felt it was important to divide the 36 drives into two RAID-6 stripes. This had the secondary effect of improving performance as well.

The simplest way to view the defining characteristic of RAID-6 versus RAID-60 is to look at how many drives can fail at the same time in a RAID-6 versus a RAID-60 without losing data. A RAID-6 can lose up to 2 drives without data loss; however, a RAID-60 can lose twice that number without data loss.

Diagnosability: Proactive Diagnostics and Vertical Integration

One advantage to vertical integration is enhanced diagnosability. As we mentioned in the vertical integration section, the Recovery-943 continues the tradition of all Unitrends Recovery-series appliances with respect to a strongly interconnected stack consisting of the

• Hardware platform consisting of an integrated server, networking, and storage complex;
• Software platform consisting of our RecoveryOS™ Linuxbased operating system and Unitrends Simply Scalable™ backup, archiving, instant recovery, and disaster recovery software;
• Support platform consisting of our proactive monitoring coupled with our world-class support organization with a 99% NPS (Net Promoter Score) customer satisfaction rating.

Typically when technologists talk of a “stack” they are referring to components that work well together through a set of well-defined APIs that allow abstraction among the elements of that stack. Unitrends uses this concept to create its hardware, software, and support stack – but it extends that concept such that the components have intimate knowledge of each other. This enables, for example, the proactive monitoring such that each of our backup appliances can “phone home” via an SNMP trap each day with collected information concerning not just the software but the underlying hardware as well. Unitrends support then uses that information to monitor events and notify customers proactively. This enables a complete “virtuous cycle” of monitoring, analysis, and action for the entire vertically integrated solution.

IPMI 2.0 is also supported on the Recovery-943 for deeper hardware and firmware diagnostics.

I’d love your feedback on what we’ve covered so far. Leave me a comment to let me know what you think.