By marrying hardware and software in container appliances, stealth-mode startup Diamanti has a chance to make containers mainstream.
Containers are a big deal, threatening to upend the comfortable world of virtualization. But as impressive as containers, the technology, has been, the business of making containers pay is still in its toddler phase.
Which is one reason that Diamanti, formerly Datawise.io, is potentially promising.
Diamanti has been a consistent contributor to Kubernetes, the container orchestration software that Google released as open source last year. But as I learned in a conversation with Diamanti CEO and cofounder Jeff Chou, the process for getting containers into VM-worthy enterprise adoption could be accelerated by a special, appliance-based approach.
For the love of developers
Containers are taking off as developers embrace this new, agile, lightweight approach to packaging applications. But adoption lags what we saw with virtualization, the last major technology transformation of the data center. VMs could just drop into a company's data center infrastructure. This is not true of containers.
Backing up a step, it's clear that developers are the key to container adoption. Developers love the ease of using Docker containers. More than ever before in our industry, developers are in the driver's seat for companies -- even outside of traditional tech. Containers speed everything up, allowing developers to take advantage of their benefits of higher productivity, application portability, and ease of scale out.
That's the good news.
SEE: Docker's no longer all about test-and-dev, says Docker CEO (TechRepublic)
The bad news, according to Chou, is that "Docker containers hit a wall once companies try to force containers into legacy IT infrastructure." The problem, he tells me, is that operational costs and complexity jump with all the manual tuning and new processes required to keep containers running in production at acceptable service levels. Because the infrastructure is designed for VMs and not containers, it typically takes a person months to plan and deploy.
Because of these constraints, "Companies brute force performance problems by throwing hardware at it -- they over-provision networking and storage." It's bad enough that doing so increases capital costs, but it's made worse by the fact that hardware utilization at most enterprises typically runs down at the 15% range. As Chou concludes, "With the capex today on hardware, no organization can afford to throw away 85 cents on the dollar."
Turning containers into an appliance
To solve this problem, Diamanti just announced a converged appliance as "the best way to overcome the challenges for containers," Chou avers.
It's the fast-track way to migrate to containers with no vendor lock in. Run any app. Choose your orchestration from Kubernetes to Docker Swarm to Mesosphere DCOS. Choose your preferred Linux OS. We're simple. We install in 15 minutes in your current infrastructure with no tuning or other changes required. Deploy containerized applications in under 60 seconds with no storage or networking configuration. Performance? We offer 1 million IOPs per appliance. We guarantee 100us from development to production versus 1ms for DIY and traditional solutions. And instead of 15% utilization of hardware infrastructure, we'll deliver 90%.
If all that sounds too good to be true, it's worth backing up to get some context.
In the Diamanti worldview, the modern data center looks more like Google than VMware. Containers in this world will be deployed mostly on bare metal, not on VMs, similar to how Google runs its data centers and cloud. "Deploying containers on a hypervisor defeats agility," Chou warns, "and then your application is lugging around an entire VM with its operating system, network stack, storage stack, and you're consuming a lot of memory. VMs are heavyweight."
"When you talk about nimble deployment of applications, you need to run containers on bare metal."
If this sounds a bit too radical for most staid and stodgy enterprises, it is. But Chou was quick to caution that this brave new world of bare metal containers won't happen overnight, or universally. "People say the previous tech goes away. No, it's hybrid. We'll always have VMs." This is so because there are application environments that require VMs, like client-facing virtual desktops.
Even so, he insists, "You'll also have a lot of what I call server-side applications that don't need a virtual machine per instance. We'll see a whole class of new applications that don't need to pay the VM tax in licensing fees or infrastructure resources."
For these, a Diamanti-style appliance may be perfect.
Making money while making friends
It could also be a way to actually make containers pay, which is something that often eludes successful open source technologies. Wide adoption doesn't necessarily mean profits at most open source companies.
Diamanti's Chou believes it's critical to start with a strong allegiance to open source -- it can't just be lip service. But it also must account for the real world in which the software will be run:
As a company, we are committed to open source. We see that devs and dev-ops need off-the-shelf tools with no vendor lock in. We also see that the world is going hybrid cloud. We designed our solution for that hybrid world so we're not strictly on-premises. Any modern infrastructure needs to account for both private and public cloud deployments with containers as the common currency. We also don't rely on hypervisors. That is new. We are an infrastructure play that supports multi-tenant cloud.
It's not just about software, however. "We are also evangelists for open compute, the hardware side of the house," Chou stresses. "We're hugely committed to the Open Compute Project (OCP). We believe in commodity hardware." The problem, Chou goes on, is that "OCP needs more innovation around I/O. To date their projects have been all about servers, disks, compute."
SEE: Hybrid cloud: The smart person's guide (TechRepublic)
As luck would have it, the Diamanti team delivered a lot of new technology at Cisco around UCS. And Cisco was happy to give away the converged network adaptor technology because Cisco didn't want to be in the NIC business. Because of this background, Diamanti has both the incentive and the expertise to innovate around storage and networking I/O, while giving much of it away.
This means that, according to Chou, "In the future, our customers will have the option of picking an off-the-shelf OCP controller in their server of choice that our software will run in." How does he think Diamanti will win despite giving so much away? In Chou's words, "We'll win in the marketplace by out innovating our competition."
In other words, they'll win precisely because they give so much away.
- A big step forward in container standardization (ZDNet)
- Docker and Mesos: Like peanut butter and jelly (TechRepublic)
- Docker announces Container-as-a-Service to bridge the DevOps gap (TechRepublic)
- Consider this operational challenge before implementing containers (TechRepublic)
- 10 things you should know about Docker (TechRepublic)