We ship the B1200i with 3 SSDs, leaving the remainder of the drive bays available for HDDs. Because the B1200i is in “dual disk redundancy” mode by default, those SSDs are in a triple mirror, allowing you to lose two of them and still have access to your data. While both the transactional tier (SSDs) and the bulk tier (HDDs) are in the same common pool of storage, they do have to be protected individually. Because there are 9 drive bays available for HDDs, more capacity-efficient protection mechanisms can be used, like stripes with parity.
In Drobo 5D and Drobo Mini, there are many fewer drive bays than in the B1200i, so if we used the same tiering setup, overall capacity would be significantly compromised. Taking thef Drobo Mini as an example, if you put in 2 SSDs and 2 HDDs – with the tiering setup in B1200i, the SSDs would be mirrored for transactional data and the HDDs would be mirrored for bulk data. Since bulk data is what’s used for capacity, you would effectively have the storage capacity of a single drive, even though you purchased and installed four drives.
In these new Drobos, the first thing we did was add an additional drive bay for an SSD. To ensure we did not disrupt Drobo’s compact form factor, we utilized the industry-standard mSATA size. Then, we enhanced Drobo’s tiering technology to utilize it for transactional data, leaving all of the standard drive bays for high-capacity HDDs. With the mSATA SSD installed, the capacity of that SSD is paired up with equal space on the HDDs (only a sliver of the HDD capacity). That space on the HDDs is mirrored for performance and redundancy (learn more about BeyondRAID). When transactional writes come in, they go to the SSD and the HDD mirror. No parity data is created, so this is fast. When reads are issued to the “hot” data that is on the SSDs, it is only read off of the SSD, so no seek penalty that you get from HDDs.
The result is a desktop (5D) and portable storage array (Mini) that’s not only fast in standard benchmarks (like other Thunderbolt arrays), but is also fast in real-world workloads.