Bare Metal Is Reliable, But Doesn’t Have to Be Boring
Over the last ten years or so, I have deployed applications to all the flavors of software and infrastruture services, from virtual machines to containers to functions. Having done that, I naively thought of bare metal servers as just a substitute for really big VMs. (Spoiler alert: I was wrong!) Having just wrapped up my third month at Equinix, a leader in the bare-metal-as-a-service space, I’d like to share with other folks who come from more of an app developer background some of the “aha” moments I’ve had learning more about this type of infrastructure.
The Right Tool for the Right Job
Bare metal isn’t for everyone. I hope that’s not a contentious statement.
It depends on what you’re trying to accomplish, sort of like selecting a transportation method (walking, biking, driving, flying). I wouldn’t fly my kids to school, nor would I deploy a “Hello, World!” app on a bare metal server. The nature of the task should dictate which tool to use.
Since joining the team, I’ve been hearing that Equinix Metal, our bare metal cloud service, is a “platform for building platforms,” and now I get it. If you want to deploy a platform — say, one like Zoom, with millions of daily users — bare metal is a great choice for scaling.
Here are a few more example scenarios where bare metal really excels:
- High performance, low latency: Bare metal is ideal for applications that require powerful hardware and low latency, such as video streaming or online gaming. It provides direct access to hardware components (NVMe (nonvolatile memory express) drives, NICs (network interface controllers) and GPUs (graphics processing units), for example) without the performance overhead of conventional virtualization.
- Regulated workloads: GDPR and PCI-DSS may be the most common data protection and privacy regulations, but there are many more, and they differ from one part of the world to the next. A global bare metal cloud platform helps you keep sensitive data within the geographic confines defined by the relevant regulations to remain compliant.
- Predictable workloads: The cloud was made for elasticity, but if your workload’s demand is predictable, moving it to a dedicated bare metal server might save you in the long run. (Think of use cases around web hosting or email.) It’s worth noting that Equinix Metal makes the choice between cost and performance on one hand and elasticity on the other a lot easier with its automated, rapid server provisioning (more on this later).
Whoa, That’s a Lot of Networking!
When I first logged in to the Equinix Metal portal, I saw a lot more various networking resources (IP addresses, subnets, VLANs (virtual local area networks), BGP (Border Gateway Protocol ), gateways, Layer 2 and so on) than I was used to. This is not a cloud console that gives you a single URL that is set up for you automatically when your application is deployed.
There’s a reason for this: control. The further down the stack you go, the more options are available to you, the person responsible for architecting your application. As with most things in life, there’s a tradeoff involved. In this case, the trade is between control and convenience. Which of the two you need more of depends on what you’re trying to do. If you’re trying to set up a globally available platform, for example, I’d argue that you need more control, and that you need all those mechanisms under the hood to be surfaced the way a bare metal cloud surfaces them.
An API for Colo
The cloud ecosystem we enjoy today owes much of its success and vibrancy to the user experience of interacting with cloud platforms via APIs. Equinix Metal has created that experience for bare metal. Its users can provision, configure and manage dedicated servers anywhere in the world remotely via APIs. This helps infrastructure engineers get started deploying on bare metal in minutes, not months.
Two popular methods of interacting with Equinix Metal APIs are Equinix Terraform modules and the Equinix Metal CLI. (Many more methods are available, documented over at Equinix Labs.)
Example use of the Equinix Metal CLI:
|
1 2 |
# Provisions a c3.small.x86 in the Dallas metro running Ubuntu 20.04: metal device create -p $METAL_PROJECT_ID -P c3.small.x86 -m da -H test-staging-2 -O ubuntu_20_04 |
Example use of the Equinix Terraform module:
|
1 2 3 4 5 6 7 |
resource "equinix_metal_device" "device" { hostname = "tf-device" plan = "c3.small.x86" metro = "sv" operating_system = "ubuntu_20_04" billing_cycle = "hourly" project_id = "f2a2d7ad-123-456-789-10ebdf49cf84" |
Connecting to Other Clouds
While the term “hybrid cloud” may be cringy to some cloud purists, the reality is that many organizations use a combination of services by various cloud platforms. Many of these organizations, be they government, finance, telco or healthcare, have workloads running on premises that aren’t a priority for being moved to the cloud because of privacy/data regulations in those industries. Those organizations would still like to use the advantages of modern cloud services for those workloads if they can.
Bare metal cloud providers make this possible by offering private network connections to other cloud platforms. Equinix Fabric, for example, offers dedicated network connections from customers’ own colocated IT infrastructure — or their bare metal servers — to their cloud providers through services like AWS Direct Connect, Azure ExpressRoute, Google Cloud Interconnect, IBM Cloud Direct Link or any other supported cloud.
This is “cloud adjacency,” and it is the next evolution in the hybrid multicloud world. One of my colleagues described the concept perfectly in a recent blog post:
“Cloud adjacency is the key: companies get the in-country physical infrastructure they need to satisfy regulators. They can also take advantage of cloud services without sending data out of country or exposing it to risk. Only data that needs to be moved to support a particular cloud workload will be moved, and any data that does pass to the cloud will do so over private, secure interconnection.”
One Last Thing …
There’s a lot more to bare metal computing than I previously thought. There’s a rich and vibrant ecosystem at play here that runs some of the world’s most mission-critical applications. The ecosystem is, perhaps, somewhat underappreciated, mostly due to a lot of newer and shinier technology. While at a recent KubeCon, I heard time and again that folks want their infrastructure to be reliable and boring. I definitely agree with the first part, but I hope I’ve shown that infrastructure doesn’t have to be boring, at least not bare metal.
