Silicon Valley Hybrid Cloud Infrastructure Startup, Datrium, Drops 3rd Shoe

For most companies, just one major release in a year is notable. And though Datrium’s 2017 product evolution to date is the result of years of planning, our latest release is phenomenal by any standard. That’s why I am stopping for a moment to consider the impact of, not one, but three shoes dropping in a year.

If you don’t know, Datrium’s mission is to provide effortless hybrid clouds. In April, we unveiled a complete, policy-driven, VM-centric cloud data management dimension in DVX 2.0, consolidating primary and secondary storage. It eliminates the need for most users to have to buy, learn and operate separate data protection infrastructure. This week, with the announcement of DVX 3.0, we’re extending the throughput and capacity of DVX by an order of magnitude

>100% growth Q1 to Q2

Now with more than 150 deployments, we have grown our Fortune G500 customer list. For example, our much simpler operating model opened the door to a customer who picked us after a year of testing an HCI alternative. Only after a few weeks of testing, they decided to go with Datrium’s because of the 2x higher performance they got in their mixed-use, mixed-server environment. This is why Datrium more than doubled sales from Q1 to Q2. What can I say, nothing beats simple.

And then there’s Data Cloud

Even HCI vendors who pride themselves on integrated backup will mirror over RAID for hot data, or only have erasure coding on all-flash systems, sometimes with expensive post-processing. They’re high-touch and can’t compete for integrated secondary storage. Not to mention, they require additional products and complexity.

Why get a flash array plus a backup array plus backup/recovery software?

Why get additional WAN accelerator hardware? 

Why learn how to manage the combined hardware and software complexity?

Many of our customers asked these same questions and decided to go for massive cost reductions by leveraging DVX alone for all compute, all storage, all data protection and DR needs with unified management. We were there when one of our education customers needed to keep up to four years of daily snapshots in a single governance archive replica for multiple primary sites, at low cost.

Split provisioning is the future of private clouds

Split Provisioning is the future of private clouds because hands down it’s the best way to simplify and de-risk scalable infrastructure for VMs and containers.  It’s great small, but it works even better as you grow incrementally to rackscale. Compare the new DVX 3.0 to your favorite all-flash array and you’ll find a single DVX can now support up to:

  • 200 GB/sec. read throughput
  • 8GB/sec. write throughput
  • 18 Million IOPS (4KB, tested with Broadwell and SATA SSDs, so watch this space)
  • 1.7 PB effective capacity at about $0.50 / GB list price
  • 128 Compute Nodes sharing storage
  • 1.2 Million VM or stateful-container snapshots

Performance is on Compute Nodes with flash, and persistence is erasure coded across a cost-optimized object store pool (Data Nodes), as nature intended.  So it doesn’t suffer HCI’s erasure coding performance problems for hot data, Datrium’s patented erasure coding approach is log-structured.

As you add Data Nodes in DVX 3.0 pool, you add both capacity and write throughput linearly. All nodes contribute to drive rebuilds, while a 4TB drive rebuild in a single data node takes about five hours, at four Data Nodes it’s only one hour. When a Data Node is added, loads rebalance automatically.

Is the data node pool “scaleout”? 

The Data Node pool is a form of scaleout, but the word “scaleout” has latency baggage. In DVX, write IO is direct from a host to particular drives with one hop and zero host/host chatter; reads are in-host with no hops.  By contrast, scaleout arrays receive a request and route data internally among controllers, typically requiring a separate backend fabric, hurting performance by increasing latency.  Scaleout HCI has a ton of chatter across monolithic node types, which also affects latency predictability. 

Multi-Hypervisor, and Bare Metal Container Support

Per our announcement last month, DVX has extended our VMware support with bare-metal Linux KVM or container hosts, including exciting new partnerships with Red Hat and Docker for future optimizations.  Not only can there be a skyline of different host types, they can be running different hypervisors. With DVX, you can snap a single stateful container on one host and use it on another immediately—awesome for accelerating DevOps. This works seamlessly on the same system in combination with DVX’s award-winning support of vSphere.

DVX scales up to 6000 VMs in a Snapshot / Replication Protection Group  

Unlike HCI, which can sometimes offer up to 200 VMs in a snapshot policy (and no direct container support), DVX scales up to 6000 VMs and/or containers directly in a single protection group, all with dynamic binding to auto-include a VM or container to an existing group and associated set of policies.  Each DVX can now support over a million snaps, replicating to many other DVXs or to AWS.

We’ve also added our own VSS provider to enable pause-less and fully transparent application consistent snapshots for Windows VMs. Most of our customers use MS SQL Server, and large database configurations can take far too long to snap when stunning the VM with vSphere. Not so with Datrium VSS – Windows servers will not skip a beat and I/Os will continue to flow!

Pause for some personal gratitude

I couldn’t be prouder to work shoulder to shoulder with the amazing Datrium team. While DVX is getting asymptotically close to effortless for large private clouds, it has taken years of discipline among brilliant collaborators to build it and make it rock solid. That’s why there isn’t anything else in the market like it.

A special thanks to our customers, resellers and technology partners too. We’ll never forget that you all took a leap of faith with Datrium in our start-up days. We commit to continuing to do our best to earn your business every day. We look forward to the rest of 2017.