Building Private Cloud Storage - HCI or Dedicated Array?

Last week Martin and I had a great conversation with Kieran Harty, CTO of Tintri Inc. Tintri sells a range of storage appliances that deliver VM-focused shared storage. Rather than talk exclusively about products, we discussed how storage arrays are evolving to meet the needs of private cloud.

Requirements

The NIST definition of public cloud gives the basis of a set of requirements that should be expected in private cloud storage. As Kieran explains, we can take the features of elasticity, automation, billing and effectively a service-based approach and deliver these with today’s storage platforms. However, these capabilities need to be more dynamic than we’ve seen from storage over the years. Traditional arrays from 15-20 years ago (and their modern evolutions) had issues that made it hard for them to deliver in the cloud. Specifically, the issues relate to the manual processes that storage administrators used to have to do. Data layout (because of relatively poor HDD performance) was a key skill. Mapping and masking were tedious and error-prone.

New Technology

Many of these flaws have been addressed. Modern arrays can be driven by APIs and don’t need human interaction. Even the process of hardware discovery is much simpler. I recently authored a paper (sponsored by HPE) that talks about the new features of Smart SANs and Peer Zoning. Together these technologies take the manual effort out of discovering and zoning storage in a Fibre Channel network. With fewer steps and less work for storage administrators, mistakes can be minimised and technology can be deployed quickly in an automated fashion. The paper also covers the features of 3PAR that have cloud advantages. It also covers Synergy, HPE’s composable infrastructure offering.

Policy at the Object Level

One important feature of any storage being used in private clouds is the ability to apply policies at the right object level. Tintri applies features such as snapshots, clones, quality of service and replication at the VM level. 3PAR applies them at the LUN/volume level, although a single volume could be relatively (logically large) even if it held a small number of virtual machines. The use of policies starts the abstraction away from the physical constructs of the hardware. Being able to say a virtual machine or container gets a specific service level, regardless of the platform on which it sits, is a powerful feature.

Federation

One of the biggest features of private cloud storage is Federation. This means the ability to treat multiple arrays as a single unit and move workloads between them on demand. With Federation, hardware can be moved in and out of the infrastructure for maintenance and upgrade and to rebalance for performance and capacity.

HCI or Dedicated Array

This leads us on to whether dedicated storage or HCI is the right solution for private cloud. Of course, the standard answer is “it depends”. Good friend Howard Marks recently authored a paper in conjunction with Pure Storage that looks at the cost of delivering HCI with FlashStack compared to Nutanix HCI. The original paper generated some debate, resulting in an update to the figures. The 2nd edition shows that converged infrastructure with dedicated storage holds its own against HCI and could be cheaper depending on storage efficiencies. Obviously “mileage may vary” and other factors such as operational costs will also have an impact.

The Architect’s View

Dedicated storage is far from dead and evolving for private cloud. The nice thing about the issues we’ve covered here is that they address customer needs instead of focusing on speeds and feeds. That’s increasingly where the storage story should focus. After all, we’re all here to add value to the bottom line. You can find the recent podcast here (embedded) or over at the Storage Unpacked website.