Author: Julie Herd, Director, Technical Marketing
A few months ago, Dell/EMC published benchmark figures against Pure, touting the strength of their Unity arrays against Pure’s all-flash systems. Those results (found here) have sparked several discussions online. We’ve enjoyed reading the recent articles and blogs, and have watched the competition between first generation all-flash arrays play out online.
For E8 Storage, it was like watching horse races when you are driving a formula one race car. The competition is exciting but it’s a completely different race from where we are competing. That’s because the first generation all-flash arrays are architected and built for SSDs that use legacy SATA / SAS interfaces that were originally designed for hard drives. With a max I/O speed of 12Gb/s, those protocols can’t truly unlock the full performance potential of the underlying NAND flash. Trying to compare our NVMe-based storage performance against Gen 1 arrays isn’t a fair comparison.
That doesn’t mean that we didn’t get curious to find out how E8 Storage performs on this test. While nothing is better than testing in a customer data center using their own data and test environment, standardized benchmark tests do provide a level playing field for vendors to showcase their wares. The VDBench test workload is a mix of block sizes, reads and writes, with a dataset sized to ensure that it can’t be served entirely from system memory. This gives results that are closer to ‘real world’ than a simulated run of 4K block IOPS with QD=1 latency.
The challenge was determining what sort of competitive comparison we should make. Using the published results for the Gen1 AFAs wasn’t a fair comparison, but thankfully Pure Storage announced its FlashArray //X70 next generation all-flash storage. Based on NVMe flash, Pure announced that the //X70 would be 2x – 4x faster than their //M70 system, which gave us enough information to extrapolate some performance results. (Note: Pure disputed the validity of the original VDBench test results mentioned above but have not published any updated performance figures)
So, how does E8 Storage perform on the same standardized workload? Using the same workload and proof points, our team ran those tests against our D24 appliance. Straight from the box, no fancy tricks or tuning knobs. Just 24 NVMe drives (RAID6 22+2) in a 2U dual controller appliance, connected to 12 host servers with E8 client software via 100GbE (25GbE to each host server).
We think the results speak for themselves. In fact, we had to slow down the 256K BW results to match the latency profiles of the original tests. When we ran that test unhindered, we achieved 39,500 MB/s at 6ms latency. We left the originally published //M70 results in the table to show how we extrapolated the //X70 performance, using 4x the performance of an //M70.
How did we achieve such great performance, you might ask? Simple, E8 Storage isn’t just another storage appliance nor is it just another all-flash array. It’s a new architecture built from the ground up to leverage a new interface designed for flash, NVMexpress (NVMe). At every point of our architecture, we’re addressing the bottlenecks that impact competitive systems without sacrificing availability or reliability.
Our architecture uses high speed 100GbE network interconnects to remove the network bottleneck. We offload and distribute data path operations to the hosts via our E8 client agent to remove the HA controller bottleneck. We’re using the highest performance NVMe drives to remove the media bottleneck. All while ensuring high availability and reliability in a dual controller HA appliance.
Our patented software architecture is why we can outperform the competition, even if they’re using NVMe drives. Their legacy dual controller architecture is still the bottleneck for performance. To be generous, even if the original Pure //M70 results were only 10% of their actual performance, when the Pure //X70 releases (in 2018) it will still be half the performance of E8 Storage that is available TODAY.
Customers need better performance for their mission-critical applications that drive their revenue. Our performance makes E8 Storage ideal for high-speed analytics, machine learning, transactional databases…anywhere that high throughput and low latency is essential.
It’s an exciting time in the storage industry, and E8 Storage is helping to create and drive new innovations in performance for how customers store and access their mission-critical data.