What is a HyperParallel Flash Array?

The Pavilion HyperParallel Flash Array delivers order of magnitude improvements in throughput, bandwidth, latency, density, and flexibility, while dramatically reducing costs when compared to other storage solutions. But what exactly is a HyperParallel array?

Your New Array is Using Old Technology

Storage array architecture has not really changed in the past 20 years. Interfaces have changed as we have gone from SCSI to Fibre Channel or iSCSI, and drives have evolved from SCSI HDDs to NVMe SSDs. Still, the underlying architecture is the same. Arrays today, including the latest All-Flash Arrays (AFAs) from the major vendors, are all still based on a dual-controller, where each controller is essentially a self-contained server. Those two controllers are then clustered together to provide services. Best practices for failover ensure that the system never delivers performance beyond what one controller can provide. This legacy architecture is the fundamental bottleneck in storage arrays today.

A Modern Design for Modern Workloads

Pavilion changed all that with the HyperParallel Flash Array. The Pavilion HyperParallel Flash Array is not designed as a server cluster. It is designed like the fastest thing in the data center, a switch. By using a switch-based design, the Pavilion solution can support more controllers, more drives, and more network connectivity in a smaller footprint.

True, Linear Scale Up and Scale Out

Since it is based on a switch-like design, and not a server cluster, the Pavilion HyperParallel Flash Array can support up to 20 independent controllers in a single 4RU system. Since each controller is independent, customers can begin with as few as four controllers, which is already double what is possible with other arrays, and scale up to 20 in a single system. Each controller can deliver up to 1M IOPS. So, the minimum configuration of four controllers can offer up to 4M IOPS, 10 controllers will provide up to 10M IOPS, and a full system of 20 controllers will deliver 20M IOPS.

When scaling out across systems, dual-controller arrays are limited. Because they are being clustered together, every time a system is added, the total performance of every system in the cluster is decreased by some amount. This is why traditional scale-out systems only scale to a few boxes; those diminishing returns tend to diminish pretty quickly.

With a HyperParallel Flash Array, since it is using a switch-based architecture and not a server cluster, performance also scales out across any number of systems in linear fashion. Whether using a parallel file system or global namespace to scale out, every incremental Pavilion system delivers the full performance of every array. So if one system can deliver up to 120GB/s of throughput, two systems will offer up to 240GB/s, six systems will provide up to 720GB/s, and so on.

Maximum Capacity and Performance in a Small Space

Legacy arrays are performance limited by the CPUs on the controllers. Numbers vary depending on the CPU used, but typically a CPU on a storage controller gets overwhelmed by IO traffic greater than 15GB/s or 2M IOPS. Some dual controller systems will offer more drives, which increases capacity, but does nothing for performance because the controller is maxed out. As a general rule, dual controller systems are performance constrained with anything more than 24 NVMe SSD drives.

Since the HyperParallel Flash Array has multiple controllers, it does not have this limitation and can deliver the maximum performance possible from up to 72 NVMe SSDs.

Order of Magnitude Improvements in Performance, a Fraction of the Footprint

It is easy to deliver massive performance with enough big iron. For example, one major storage vendor claims to be able to deliver 300GB/s of throughput. What they try not to say is that it takes them 24 4RU appliances to deliver that performance. That means in addition to the cost of acquiring 24 storage arrays, users will also need multiple switches to connect them, multiple racks to hold them, and multiple people to manage them.

In comparison, a single 4RU Pavilion HyperParallel Flash Array can deliver up to 120GB/s of performance and over 2PB of capacity with current drives.

Put another way, on a per RU basis, the leading storage vendor above only offers 3.125 GB/s of performance. The Pavilion system delivers 30GB/s. That is almost 10X the performance per RU.

What it Means to You

Ultimately, the value of the Pavilion HyperParallel Flash Array, or any piece of flash array technology, is not a function of throughput numbers or rackspace. It is about the value that can be delivered to the organization.

The Pavilion HyperParallel Flash Array, when combined with Pavilion HyperOS 3, make up the Pavilion HyperParallel Data Platform. The Pavilion HyperParallel Data Platform enables organizations to extract more value from their data, in less time, so that they can make better decisions, and do more.