I was President of a Little League Baseball organization for several years. Every player under the age of 9 would get a “participation trophy” for showing up and trying. You see the same practice in most youth sports programs these days.
I thought of this when Pure Storage announced their support for DirectFlash™ Fabric this week. The new feature allows them to connect their FlashArray//X to RDMA-capable Ethernet hosts using NVMe-oF. Just like Pavilion, they are targeting ‘modern stack’ applications that require massive parallelism, low latency, and high bandwidth. These applications rarely use shared storage since traditional SANs add too much latency, too much cost, and not enough performance to satisfy requirements.
By using NVMe-oF it is now possible to deploy shared storage in these environments – if you have the right product. In addition to throughput and low latency, you need a choice of NVMe SSDs, enterprise reliability, data management features, and standard host software to minimize complexity. Pavilion has been shipping a product that delivers on these requirements for over two years now.
However, Pure falls way short for these applications. Buried in all of the slick messaging, some key requirements have been left unsatisfied, but they should still get credit for trying.
First, erudite customers know that you must deliver enough performance to replace all of the local SSDs in racks of servers without degrading scale-out application performance.
Pure strikes out here. They have not removed the real bottleneck, an antiquated dual-controller architecture. With only two controllers, bandwidth and performance are limited. The design is analogous to putting a lawnmower engine in a Formula 1 car. From the outside, it may look fast, but it has a serious design flaw that makes appearances very deceiving.
|’New Stack’ Application Requirement||Pavilion Data||FlashArray //X90||Pavilion Data Advantage|
|# of servers supported for local SSD performance replacement from a single chassis*||20+||<2||10X +|
|Single System Bandwidth||120 GB/s||12 GB/s||10X|
|Allow customer-supplied NVMe SSDs||Yes||No||Business Agility & Time-to-Market Advantage|
|Multiple Tiers of SSDs in a Single System||Yes||No||Workload Tiering & Reduced Infrastructure Footprint|
*Assumptions: Pure //X90 with Direct Flash Fabric 8K Read IOPS = 700,000; A single locally-attached NVMe SSD can deliver 400,000 8K Read IOPS; A single Pavilion array delivers 10 million 8K Read IOPS
The second requirement is choice in storage media. Most organizations deploying large-scale storage programs have volume purchase agreements with storage drive makers. They deploy the latest SSDs immediately after it is released. They redeploy existing drives across clusters as they see fit. This is a huge IT agility advantage.
Pavilion Data believes in an OPENCHOICE of SSD procurement. Negotiate the best possible price for the optimal SSD matching your workloads directly with the drive maker. Drop it into our NVMe-oF Storage Platform. Redeploy existing DAS drives into our array. Run different workloads (read-intensive volumes, heavy write IO volumes) within the same platform. When the latest NVMe technology is available, use it immediately – don’t wait for the “next generation” array with planned obsolescence and forklift upgrades.
As you watch the NVMe-oF game unfold, look beyond the Photoshop and InDesign glitz and peer “under the hood” for the real story. If you claim you can do DAS replacement, do it for more than a few servers per array without degrading application performance. If you have the customer’s best interest in mind, unbundle the drive purchase from the platform sale.
Pure Storage certainly deserves a Participation Trophy in the Enterprise NVMe-oF Array game for trying and we applaud their efforts around trying to address the requirements of ‘modern stack’ applications. But participation does not make a winner. Let’s leave that to competition in the field. Play ball!