At the #SFD16 event, we had the opportunity to delve into myriad of storage technologies. It was quite a firehose couple of days. I’ve already written on NetApp, Nasuni, and SNIA, and wanted to talk about another very interesting presentation we received, this time from Infinidat. As a company, in storage metrics, they’ve grown 10x in the last three years, to an excess of 3.4Exabytes currently in use of Infinidat storage. I’d say that’s some pretty impressive statistics.
So, about the technology: The classic model of Data Tiering seems to be the goal, with Solid State disc leveraged as the Hot Data tier, but in a very different approach than most tiered storage stories. The SSD for read and write cache sits at the very top of the architecture in a dedicated node or nodes, depending on size. The more caching required, the more SSD required. Then, cold data sits in spinning disc, what appears as classic JBOD, but is in fact far more elegant. This extremely dense, nearline SAS involves massive parallelism, allowing for huge scalability.
The key here is a table called TRI Data. In this table, is every piece of data that has ever been written to the storage infrastructure, as well as the relationship of that data to other data that had been written to that infrastructure at the time. There is also a sequence adjuster or ignorer classifier. And, an “Activity Vector” which is then used to predict future Input/Outputs. These predictions are then transmitted to a “Sequence Handler, in an effort to pre-fetch sections, giving them an elevated predictability and pushing those most reliably up to hot data based on this identifier. To me, it appears as if the modality of predictive data use/analytics that transcends those I’ve seen previously, which essentially revolve around a data heatmap, based on previous use.
In addition, the folks from Infinidat have launched another proximate data architecture, called Neutrix, which provides an Infinidat based storage model, SAAS based, which feels similar to offerings by HPE, previously discussed, and mentioned in my vBrownbag presentation at Discover. CloudVolumes (originally a Nimble product, now HPE as part of the acquisition), and CloudBank, now a similar approach in place with 3Par. These solutions will mitigate a lot of the problems inherent in storing data on the major cloud providers, wherein the storage costs themselves are quite painful, with the inclusion of data egress charges. In these cases, the magic of the native storage protocol is leveraged, but with the Operational Expense model, and the benefit of only being charged (sort of on a lease base) for what is in use. My Video here.
I also like the InfiniGuard model, that Infinidat has released, which does storage like the Infinibox machine, but more to the point, a dedicated disc target for backup with deduplication built in to save on space, and time. It’s a highly secure, and highly protected machine with all kinds of durability, and environmental protection built-in.