A bare-metal cloud allows you to rent hardware resources from a public cloud service provider, or sometimes a managed service provider. With a bare-metal cloud, you get direct access to the hardware platform without having to go through tenant management systems. Therefore, one of the benefits of bare-metal cloud, as it is sold to the public, is the ability to better support high-transaction workloads that do not tolerate latency.
I’ve found that bare-metal is often used by tier 2 cloud providers, and managed services providers, as a selling point of their “cloud.” Indeed, enterprises that are still attempting to maintain control over their hardware and software often pick bare-metal to maintain that control, typically while not considering costs and workloads requirements.
If you are thinking of leveraging a bare-metal cloud, keep these points in mind.
First, make sure to compare costs to actual bare-metal, meaning hardware and software you can buy and install in a datacenter, or under your desk. In doing many of these cost models for clients, I’ve found that it is usually much cheaper to continue to buy your own hardware and software, including operations and maintenance.
Second, the performance does not seem to be much better than traditional, multi-tenant cloud services. While you would think that bare-metal will “kill it” in terms of I/O performance and lower latency, public cloud providers have done such a good job of managing access to underlying physical resources that the difference is not that dramatic. However, do your own benchmarking.
Finally—and this is the deal breaker for me—it takes much longer to spin up servers on bare-metal clouds than on traditional clouds. This means that you are trading your ability to expand on demand and change on demand for the marginal benefit of running on bare metal. Considering that agility accounts for most of the value that cloud computing provides, moving to cloud without gaining it seems downright dumb.
Now, there certainly are some applications for bare-metal—I get that. My point here is that the majority of workloads I see ending up on bare-metal cloud instances get no benefit from being there, other than the ability for IT to claim proudly that they are on bare-metal clouds. Let’s work in reality, shall we?