The Cloud! The Cloud! Take us to the Cloud it’s cheaper than on-premises, why? Because someone in marketing told me so! No, but seriously. Cloud is a great fit for a lot of organizations, a lot of applications, a lot of a lot of things! But just spitting ‘Cloud’ into the wind doesn’t make it happen, nor does it always make it a good idea. But hey, I’m not here to put Cloud down (I believe that’s called Fog) nor am I going to tout it unless it’s a good fit. However, I will share some experiences, and hopefully you’ll share your own because this has been a particular area of interest lately, at least with me but I’m weird about things like deep tech and cost benefit models.
The example I’ll share is one which is particularly dear to my heart. It’s dear because It’s about a Domain Controller! Domain Controllers are for all intents and purposes, machines which typically MUST remain on at all times, yet don’t necessarily require a large amount of resources. So when you compare a domain controller running On-Premises let’s say as a Virtual Machine in your infrastructure it carries with it an arbitrary cost aggregated and then taken as a percentage of the cost of your Infrastructure, Licensing, allocated resources, and O&M Maintenance cost for Power/HVAC and other. So how much does a Domain Controller running as a Virtual Machine run inside your data center? If you were not to say, “It Depends” I might be inclined not to believe you, unless you do detailed charge back for your customers.
Yet, we’ve stood up that very same virtual machine inside of Azure, let’s say a standard Single Core, Minimal memory A1-Standard instance to act as our Domain Controller. Microsoft Azure pricing for our purposes was pretty much on the button, coming in at around ~$65 per month. Which isn’t too bad, I always like to look at 3 years at a minimum for the sustainable life of a VM just to contrast it to the cost of on-premises assets and depreciation. So while $65 a month sounds pretty sweet, or ~$2340 over three years I have to also consider other costs which I might not normally be looking at. Egress network bandwidth, Cost of backup (Let’s say I use Azure backup, that adds another $10 a month, so what’s another $360 for this one VM)
The cost benefits can absolutely be there if I am under or over a particular threshold, or if my workloads are historically more sporadic and less ‘always-on, always-running’ kind of services.
An example of this, is we have a workload which normally takes LOTS of resources and LOTS of cores and runs until it finishes. We don’t have to run it too often (Quarterly) and allocating those resources, obtaining the assets while great, they’re not used every single day. So we spin up a bunch of Compute or GPU Optimized jobs and when it might have taken days or weeks in the past we can get it done in hours or days, which means we get results and we release the resources once we get our data dumped out.
Certain workloads will tend to be more advantageous to others to be kept on-premises or hosted exclusively in the cloud, whether sporadically or all the time. That really comes down to what matters to you, your IT and your support organization.
This is where I’m hoping you my fellow IT Pros can share your experiences (Good, Bad, Ugly) about workloads you have moved to the Clouds, I’m preferable to an Azure, Google or Amazon as they’ve really driven things down to a commoditized goods and battle amongst themselves, whereas an ATT, RackSpace, and other ‘hosted’ facility type cloud can skew the costs or benefits when contrasted to the “Big Three”
So what has worked well for you, what have you loved and hated about it. How much has it cost you? Have you done a full shift taking ALL your workload to a particular cloud or Clouds. Have you said ‘no more!’ and taken workloads OFF the Cloud back On-Premises? Share your experiences so that we may all learn!
P.S., We had a set of Workloads hosted Off-Premises in Azure which were brought wholly back in house as the high performance yet persistent always-on nature of the workloads was costing 3x-4x more than if we had simply bought the Infrastructure and hosted it internally. (Not every workload will be a winner )
Thanks guys and look forward to hearing your stories!