[Preamble: I was a delegate of Storage Field Day 14 from Nov 8-10, 2017. My expenses, travel and accommodation were paid for by GestaltIT, the organizer and I was not obligated to blog or promote the technologies presented at this event. The content of this blog is of my own opinions and views]
E8 Storage technology update at Storage Field Day 14 was impressive. Out of the several next generation NVMe storage technologies I have explored so far, E8 came out as the most complete. It was no surprise that they won the “Best of Show” in the Flash Memory Summits for the “Most Innovative Flash Memory Technology” in 2016 and “Most Innovative Flash Memory Enterprise Business Application” for 2017.
Who is E8 Storage?
They came out of stealth in August 2016 and have been making waves with very impressive stats. When E8 was announced, their numbers were more than 10 million IOPS, with 100µsecs for reads and 40µsecs for writes. And in the SFD14 demo, they reached and past the 10 million IOPS numbers.
The design philosophy of E8 Storage is different than the traditional dual controller scale-up storage architecture design or the multi-node scale-out cluster design. In fact, from a 30,000 feet view, it is quite similar to a “SAN-client” design advocated by Lustre, leveraging a very high throughput, low latency network.
Here is another view of how the E8 Storage network design looks like:
The very high throughput, low latency network is of course, the RDMA fabric running 25Gb/sec, 40Gb/sec or 100Gb/sec ROCEv2 or Infiniband. In order to get an extremely fast, line-rate network speed, the E8 Storage deploys 4 or 8 100GbE NICs to support 24 x 2.5″ NVMe drives in a 2U chassis.
Each “controller” in the 2 cannisters 2U box houses 2 CPU+memory+PCIe switch as show below:
The 2U E8 Storage is just an ODM (original design manufacturer) chassis, and is central to aggregate and translate PCIe to Ethernet. In additional, another very important piece is the E8 Storage Metadata Server (MDS) software, which works in tandem with all the E8 Host Agents in each of the compute servers drawing the extreme high performance from the E8 Storage.
The E8 Metadata Server is the mapper and the intelligence to the data blocks in the NVMe SSDs. And the MDS-design in E8 Storage cleverly assigns and redirects the I/O data path between the E8 Host agents and the NVMe SSDs, once it has established the metadata required for each I/O.
This also means that E8 Storage is able to scale linearly without the burden and bottlenecks of storage controllers. The control path and the data path are separated using E8 Storage patented technology, which they designed and built from ground zero. The description of this separation is shown below:
But it is not just about speed. I have seen a few next generation all-Flash storage technologies. Speed was always the feature they touted, but other data management features were not quite ready yet. E8 Storage, as I said in the beginning, is more complete. They have a shared volume technology (patented) via smart MDS coordination, support for scaling multi-enclosures via federation, clustering and many more that I cannot fit into this blog.
For the full deep dive, check out this video:
As SFD14 drew a close on my trip on Nov 10th, I was pleased to have the pleasure of E8 Storage sharing their technology with the delegates. It was refreshing to listen to their confidence and pride of their technology, and I enjoyed it thoroughly.
The digital transformation revolution has introduced a slew new applications ranging from Deep Learning, Machine Learning, Edge Computing, Exascale data analytics and more. As the all-Flash platforms evolved, Generation 1.0 All-Flash cannot rip out their existing design and architecture to meet more extreme requirements of these new generation of applications.