Enabling the Path to the All-Flash Data Center with K2

It is a very exciting day for Kaminario, as today we announced K2 v5.5, changing the way flash storage is consumed in the data center. Our announcement unveils:

  • A new, disruptive, industry-leading average price point below $1/GB for usable flash storage.

  • The use of 3D TLC SSDs, which doubles our capacity.

  • Completion of K2’s comprehensive and robust data-protection suite with native array-based asynchronous replication.

  • K2: The Perpetual Array™ – allowing customers to take control of their storage future.

Our new architecture allows customers to gain the unparalleled benefits of a scalable all-flash array at a lower cost than any hybrid array.

Let’s see what’s unique about the K2 architecture that enables us to push flash adoption in the modern data center.

The flash industry is evolving faster than anticipated. Flash capacity and density improve faster than Moore’s law. In practice, they improve at a rate that is at least twice as fast as that of HDDs, DRAM and CPUs. Such rapid technology progress is mainly driven by two new technologies:

  • 3D NAND (or V-NAND depending on the brand name) – a technology that effectively create layers of cells.

  • Triple-level-cell (TLC).

These technologies create a great opportunity to build a highly cost-efficient all-flash array that is more efficient than any hybrid array. But there are several technological challenges to overcome:

  • How to effectively deploy SSD drives that are much denser and will continue to improve density at a very rapid rate?

  • 3D NAND and TLC technologies come with lower flash endurance than the previous generation.

The way to tackle the rapid increase in density is to build an architecture that does not assume constant proportion between the DRAM size and the size of the attached SSDs. Some products put all metadata in DRAM, and by doing so, limit the maximum amount of capacity that can be attached to each controller. This means that such products will not be able to deploy the new, high-density SSDs without a significant increase in the amount of DRAM per controller. This is extremely expensive, and in practice, becomes impossible with the number of DIMM slots inside typical x86 servers. The K2 was architected differently and does not assume that all metadata resides in the memory. With Kaminario, metadata is managed in a highly optimized way that includes the use of an adaptive block size algorithm. The metadata footprint is four to eight times smaller compared to other architectures that do not use adaptive block size algorithm.  Moreover, with Kaminario, there is no assumption that all metadata resides in the memory. This means that the product is ready to deploy high density SSDs. Today, we deploy multiple shelves with 1.92TB SSDs per controller and we will continue to be fast in deploying new high-density SSDs (last week at Flash Memory Summit, Samsung announced 16TB SSD, which will be available in 2016!).

The second challenge is endurance. 3D TLC SSDs come with lower endurance.  The datacenter version of these SSDs that Kaminario is deploying comes with endurance of one write per day (of the entire capacity) for five years. The Kaminario HealthShield™ has been collecting endurance statistics for the past few years, and from analyzing the data we see that 97% of customers are writing less than a single write per day of the entire capacity. As such, for the remaining 3% and in order to be safe, we implemented a number of critical features to significantly improve endurance:

  • Inline Compression and Deduplication – note that always being inline is critical, as a post-processing algorithm writes six to eight times more than an inline algorithm.

  • Fine grained load-balancing of writes evenly across all SSDs. This effectively prevents any possible write hotspot.

  • Efficient RAID implementation that always writes full RAID stripes; a partial stripe write performs a read modify write operation that triples the number of writes in a typical dual parity implementation.

  • A Log Structure Array (LSA) data layout that writes large blocks in a way that is optimized for the flash media.

These optimizations improve flash endurance by five to ten times and, in practice, convert a drive that supports a single write per day to support five to ten writes per day. Shrinking a host write into a 0.1-0.2 write ensures that the SSDs can survive much longer. This means that with Kaminario’s flash optimizations, a typical customer will have enough endurance for dozens of years on a Kaminario deployment with 3D TLC SSDs! We are also continuing with our seven-year flash endurance guarantee under any workload.

Overprovisioning resources is always expensive. Customers should not overprovision capacity, performance or endurance as it significantly increases the system cost without delivering any real business value. Not all storage arrays were built in a way that is optimized for capacity, performance and endurance. Kaminario’s SPEAR™ architecture is the only one that allows customers to scale up and scale out, and is well-optimized for flash endurance. Combining the three benefits creates the most scalable and cost efficient all-flash array in the market.

I believe that after understanding the architecture tradeoffs Kaminario made, it is no longer a surprise that we are the only vendor that can deliver the benefits of flash at a cost that is lower than any hybrid array. Kaminario has made the all-flash data center a reality!

New Call-to-action