Others have tackled this issue better than me, but I have a different angle on this.

Lots of you are replacing outdated systems, typically running on premise (i.e. within your network, whether in a broom cupboard in your HQ or in a colo facility where you pay for usage). Some of those run directly on bare metal, but a lot run on a hypervisor of some description (VMware, Hyper-V, KVM etc).

A lot of those are what I call resource intensive applications – they need a lot of feeding. Compute requirements typically 32 CPUs+ or 368GB RAM+. This was OK on premise, you already had or built the capacity to allow this. A one-off capital expenditure (cap-ex) whether you run the application for 1 year or 10 years!

The interesting part is when you decide to update/upgrade this on premise application and over the years while you’ve remained on an older version it has built cloud capabilities into the newer iterations. It’s typically now wholly cloud deployable/migrate-able or potentially retains some hybrid elements. Often it’s the compute bit that is easiest to cloud-ise.

Application Lifecycle Cost

How long does an enterprise application service typically last? This is a key question, because the operational expenditure (op-ex) that is the basis of cloud charging is fundamentally based on this key metric.

I’ve asked this question to application owners in the past and they might say 3 years or 5 years – but that’s actually not a realistic answer. Stay with me on this one, the more accurate answer is “Well, we expect it to be 3 years but then however long after until we upgrade/replace/migrate/decommission it”.

If I was you – I would immediately double the op-ex figure for my application service when presenting the business case for it. The cost figure you put in your presentation to business owners is completely wrong.

Actual Costs

Over on X I’ve seen some crude comparisons between a server you pay for versus the equivalent compute you rent from a cloud provider. The argument there by a quite famous X/tech personality was rather aggressively anti-cloud “Look it’s cheaper to buy outright and host it yourself”.

Here’s an example (NOT the guy mentioned above btw):

I’m not anti-cloud but I am against certain workloads being put into the wrong cloud!

What I mean is, your business might be all in on AWS, Azure or GCP. You assume everything has to be deployed into 1 cloud provider, usually because you’re convinced the network works faster or costs are easier to reconcile to cost centres.

There’s a reason why AWS makes the biggest profit margins across the whole of the Amazon group. A lot of organisations are spending a lot of money, and much of it is wasted. I’ll talk about cost optmisation strategies in a future post.

My Example

So we have a fairly heavy compute requirement, although a containerised application service it still has a huge CPU and RAM requirement. Let’s say 48 vCPUs and 192GB of RAM, throw in 1TB of SSD storage for fun stuff like your holiday pics

*Dell PowerEdge R750: 48 CPU cores, 256GB RAM (64 x 4), 1.9TB SAS SSD (https://www.dell.com)

Obviously some caveats, with VM pricing in Microsoft Azure you can of course reduce cost via your MCA agreement with Microsoft and of course with both Azure and AWS you can purchase 1 or 3 years Reserved Capacity (reducing by roughly 30% or 60% the total VM cost over the period). So that £100,260 could be approx £40,000 with a 3 year RI applied.

Additionally there are other alternatives to Hetzner used in the table above, for example (vCPU and RAM are equivalents as these providers have limited options compared to the bigger cloud players):

• Digital Ocean – ‘Droplet’ VM 24v vCPUs, 192GB RAM £48,300 for 5 years
• OVH – r3-256 VM 32 vCPU, 256GB RAM £37,527 for 5 years

Food for Thought

The prices above don’t include the operational overhead for support (largely these are similar across all these Server/VM types). There is a price to add to the On Premise server, that of installing it and the required power/network/security/colo components.

You can see the AWS & Azure costs are equivalent, and even with the 60% 3 year Reserved Iinstance (RI) discount, remains significantly larger than the other costs. The others are similar and the preferred option would be the hosted service with Hetzner. It will save you at least £25K over 5 years if not more.

Connectivity

You might think it’s a hassle to provide reliable connectivity between 1 provider and another. A secure IP Sec based VPN connection is sufficient, and you then have the future option of further savings by deploying more cost effective VMs (and maybe other services) with the provider who charges less.

Some organisation already do this with their multi-cloud setups, with Azure and AWS in the mix.

The Future

VMs are not the future of enterprise anyway, I always have and will continue to advocate for getting rid of IaaS based services (no-one needs the hassle of operation VMs). Where it suits move to microservices and loosely coupled frameworks, or serverless options and take advantage of who has the best pricing for specific services.

Or for longer term maintenance go for simplicity with low-code or no code and make that application service easy to move, update or modify when the need arises (it will, might not be a you problem right now but will be yours or your successors in the future). This will save a lot of migration/upgrade/transition headaches and is the way most enterprise applications should go but pretty much none of them do.