This month the invitation from Koen asks what would happen if we came back from the cloud. Back to On-Prem.
Cloud solutions have sliders which magically reduce downtime, performance issues, and account credit. When you build it, infrastructure and platforms are your challenge, and you get one shot to right-size it. Capacity, resilience, capabilities. They’re your challenges to solve. There is no cake slider.
It’s the capabilities I want to consider here. The capabilities we have available to us day-to-day. Potentially two opposing sides:
From a positive perspective, on-prem is still on-prem. The foundations haven’t moved, and we’re not breaking new ground. The capabilities are stable and predictable. Quirks or challenges will have solutions scattered across StackOverflow, Google’s search results, or assimilated into LLMs. That’s reassuring.
Personally, I’ve felt more confident having everything under our roof and team’s control. I recently had a Managed Instance lose DNS during routine patching, but triaging was painful due to the abstraction. On-premises the full stack is yours.
Moving back to on-prem lets you define your capabilities through choices of platforms and tooling without the chance of them shifting or deprecating under your feet like in the cloud. Platforms will have a lifecycle defined on your terms. You can mindfully build towards the capabilities that matter, whether that’s through infrastructure or services.
Rebuilding on-prem allows you to define the blueprint across all layers of your solution stack. This can be a daunting task, but a valuable one where you can set an organisation up with the right capabilities for years to come.
On the other hand, on-prem is still on-prem. That’s likely not where you’ll find all the organisational data. In my experience, the business data will be supplemented across SaaS platforms and document libraries. The use of connectors in the cloud with no- or low-code solutions enable connectivity. The enabler is the technology, and those capabilities aren’t natively available on-prem.
As part of reverting to on-premises, conversations would need to consider integrations and 3rd party or licensed tooling to fill the gaps where local features have trailed as vendors push towards their cloud equivalents. For disparate cloud solutions, this can lead to complex on-premises setups to replicate the capabilities which were cloud-native.
On-premises, the capabilities are what you build. I don’t miss the tough days fiddling with clustering, poking availability groups, or wrestling with gMSAs. The reassurance of controlling the full stack comes with the responsibility of maintaining that. You’ve got to repair your own time machine, Doc.
Falling back to on-prem limits your capabilities and careful planning is needed to ensure the technology can effectively service the business needs. AI is a prime example where local adoption is possible, but prohibitively expensive at scale.
On-prem is still on-prem.
Koen asks about building everything on prem. That’s not something I’ve experienced as a conscious decision, so I can’t speak directly to that. For me, hybrid feels like the right choice, not just a default position.
On-premises provides stability and control for what you need to secure, and the cloud delivers the capabilities your business needs. New capabilities will come from the cloud first so it’s likely that hybrid isn’t a transition, it’s the end state. No need for Judgement Day.
Andy Brownsword 
2 replies on “T-SQL Tuesday #199 – Back to the Future IV: On-Premises”
[…] Andy Brownsword reminds us that it’s still the same animal: […]
I write this as I am on my way out the retirement door as a DBA at the end of the month and this will likely be my last post here. It is based on 4 decades of experience in this industry and a degree in Business (Production Management) from UC Berkeley.
Ultimately, the decision to go to the cloud with your compute is a multi-faceted decision with each business needing to understand the implications of the choice and not simply follow trends if they are truly interested in preserving the long-term interests of their business.
I make this comment as today, unlike when I entered this business, decision making management in a firm is done by a cadre of people who act as free agents with a build and move on (to the next business where they can pump up and cash in) and up mentality. When I entered this business in the mid 80’s, these decisions were made by executives whose careers were tied to the employer with your financial interests more tightly aligned with the long-term interests of your employer.
Back in the day, a decision to outsource IT processing, was done as a long-term financial decision. How much does it cost (in those days compute was expensive) do build all this stuff and where is the return on this capital? Such was the decision at General Motors to acquire Electronic Data Systems and move all its IT assets (and personnel) to EDS. I came into the picture there as an EDS hire in the restructuring of GM’s IT assets to what I like to call the Mainframe Cloud of the day (although they also took on the limited non-mainframe compute as well).
Through the 15 years I worked there, processing became orders of magnitude more cost efficient (through both equipment purchasing efficiencies, Moore’s law, and processing efficiencies – aligning computing needs around the clock to match the systems demands – very similar to today’s cloud management). In fact, the efficiencies were so great that at one point, EDS actually outearned its parent (bad year in auto business). Talk about the tail wagging the dog.
Along the way GM decided that spinning off EDS would allow it to grow its non-GM business (with competitors perhaps) as well as show the independence to be able to do that. However, what they didn’t account for was that their growing cost of IT (it never really went down despite all the efficiencies as there was more of it needed) was now being driven by the now more relatively expensive labor component thanks to Moore’s Law.
Not only that but now having EDS’ management not under the thumb of GM’s board of directors allowed GM only one lever to control the cost of IT – look for other vendors. And surprise, surprise! All of EDS’ competitors charged similar rates and there was always the friction cost of moving things. The return on that was marginal at best.
So, what was GM to do about this? In the original contract acquiring EDS, GM carved out its Research Computing division and its one datacenter on GM’s Warren Michigan campus where GM employees would work on GM mainframes (the big compute iron of the day). I worked with these people from time to time in my days at EDS (I was laid off as a young DBA shortly after Y2K in a cost-management move to please the new stockholders of the spun off company). They still knew how to engineer a datacenter.
So GM did something relatively intelligent. They brought back in house compute that was very specific to them – think custom engineering systems specific to GM. In that regard, they stopped some of the financial bleed of outsourcing getting what they wanted done where they could manage the costs of getting it done. They left much of the commodity compute outsourced (and to this day I believe many commodity systems remain outsourced likely in SAAS type situations).
So, in effect, GM runs a hybrid cloud approach to computing. While having been decades gone from there, from what I read, it seems to be working for them. And it makes reasonable business sense. Don’t outsource expertise that gives you strategic advantage in YOUR business, but don’t invest capital in a business where you are not the most efficient provider and allow someone else to do that with better returns.
I think there is a lesson in this for all of us (not just business execs). In your career, you invest YOUR time and possibly YOUR money in YOUR business (the value you add in exchange for wages). To be effective in this you must understand what the long-term business YOU want to be in and are YOU going to be competitive in this business or just another commodity supplier? Investing your time and efforts in things where you have a competitive advantage and allowing others to worry about the stuff you use to do your thing is a good use of your time and money. Trying to “master” stuff that everyone else masters will just keep you chasing the commodity job.
Just make sure that you understand the problem you are trying to solve, whether it is a profitable problem to solve, and keep moving your skill set to tomorrow’s profitable problems and your business will always be profitable and you won’t be beholden to those who provide you revenue.
Peace to all and thanks Andy for providing me a place to give my career thoughts…