In the beginning…
We have come a long way since I started in the IT world, if I think back to my beginning as an Enterprise Storage Architect (some 16 years ago) – we were just embarking into the realm of virtualization in the enterprise – logically segmenting physical compute and storage workloads to be able to distribute among virtual machines, and we still had dedicated storage drives for specific workloads. You would not typically share drives that were intended for SQL Logs with another workload in real to avoid performance degradation.
This concept also extended to compute and memory resource pools too, the idea of bare metal (which is direct access to the underlying hardware providing you with complete control and use of all of these server’s components) existed in a different form back then.
Oh yeah, and everything was on-premises – so you had bare metal in your data center – so no sharing server resources with other companies, and hardware was typically owned (and operated) by the companies that were intended on using it. In summary, there was no “cloud” or true concept of multi-tenancy.
Looking back on that now, when we had a whole lot of “tin” sitting in a private data center, it is amazing to see how much has changed and how much has stayed the same. The concept of bare metal was alive back then (but probably didn’t justify coining a specific term for it because inter-consumer sharing of computing resources wasn’t mainstream)
So while there was bare metal, there was fundamentally a different understanding and appreciation of what it represented to what it does today. Physical servers lacked scalability, agility, and flexibility and usually attracted maintenance.
The year 2022 – what do we have now?
So let’s fast forward 15 or so years, till today, and break down what the options are on offer across the multi-cloud universe. – there is the same concept of on-premises Dedicated Hosts (and deeper appreciation thanks to the hosting/Managed Service Provider era surging in the last 10 years), Cloud IaaS, and Bare Metal. On the face of it, these each deliver very similar resources by way of definition, but without divulging too much, the fundamental difference between the three is how these resources are consumed.
Let’s provide a quick description of each of these, the difference and pros/cons of each, and some of the leading vendors that offer the service starting with Bare Metal.
Bare Metal
What is Bare Metal?
Bare metal represents the legacy and evolution of what I covered above, dedicated hardware that can be purchased, configured and consumed privately by organizations (or anyone for that matter). It is a physical server that is only used by one customer and is offered as a whole unit to a user. It is typically intended for dedicated services to run on it which may require a certain SLA or performance level (which might not be able to be achieved with shared hardware). Bare metal typically comes without a base operating system, giving you the freedom to install what you please, onto it.
The key point here is that the hardware (and subsequent resource pool) is not shared outside the one user or tenant who is renting the hardware.
What Bare Metal is not
To be clear, a bare metal cloud is not the same as a dedicated host even though they are a lot of the same features – They both come as a physical servers and both types of servers are also single-tenant machines. Bare metal however is usually comprised of newer hardware and billed hourly or even per minute whereas Dedicated hosts are sold with a longer-term commitment (months or years) and probably represent the heritage of legacy hardware.
What are the benefits?
The obvious benefit of having a single tenant dedicated server is you don’t have the “noisy neighbour” problem which means no other workloads are going to contend for the server’s resources, and since the workloads you create on the dedicated server have full access to the complete pool of disk, CPU and memory they will receive maximum performance, reliability, and stability in this setup.
Also, as you have direct access to the underlying hardware, you can choose what you create on this, and is not pre-decided for you as it is in some of the other infrastructure-as-a-service offerings out there. While the hardware part of the stack (including power/cooling etc) is managed by the provider along with the network, you are responsible for everything above this.
And since you don’t ultimately own the hardware, it is the provider that will need to think about hardware maintenance, and refreshes (so you get access and use to the latest generation hardware they use)
What are the drawbacks?
Certainly, the most obvious drawback of having a dedicated piece of server hardware is going to be the cost, as the hardware is not shared then you can expect to be paying top dollar for renting the hardware from the provider.
Another consideration is that it will take you longer to get everything ready for production as you will need to install the operating system, device drivers, applications, etc., whereas other offerings can come with these pre-installed.
Vendors that offer bare metal to enterprises
While many different hosting companies offer bare metal servers to customers, if we just look at the major hyperscalers out there, four major providers come to mind – AWS, Microsoft Azure, Google Cloud, and Alibaba Cloud.
Amazon Web Services
Amazon offers “EC2” which are bare metal instances or virtual servers providing direct access to the compute/memory and storage of the underlying server. What is cool here is that AWS offers a range of benefits for EC2 instances such as Elastic Load Balancing, Auto Recovery, and Auto Scaling allowing dynamic capacity and performance management but keep in mind these have the potential to attract further costs should you consume more resources.
Microsoft Azure
Microsoft takes a similar approach to AWS, by offering a dedicated host (sitting in their data center) aimed at the highest level of component redundancy, performance and flexibility to manage your OS and application installation according to what your requirements are. Azure Bare Metal Instances have an isolated tenant meaning networking, storage, and compute resources are not visible or shared by other tenants.
Alibaba Cloud
Similarly, Alibaba Cloud offers ECS Bare Metal instances allowing an elastic scalable computing service as legacy physical servers, while there is a virtualization technology at play here (to leverage dynamic scaling/elastic features), Alibaba claim to be able to offer the same performance as a physical server.
Google Cloud
Lastly, Google pitch their Bare Metal solution as “OEM Hardware” that is certified to run multiple) enterprise applications specifically calling out Oracle apps as a target workload (i.e., Oracle Data Guard, Oracle Data Pump, or Oracle Recovery Manager).
With low latency, and highly available interconnects – their offering seems way more geared to specialized workloads, whereas the other three hyperscalers provide a more generic pitch for any workload.
All these vendors’ bare metal solutions are typically billed on a time-based subscription (usually hourly).
In the next post, we’ll look at the two other available services – taking a similar approach to detail what they are, the differences, and the pros/cons of each.
Stay tuned!
Follow our Twitter and Facebook feeds for new releases, updates, insightful posts and more.
