How much on-premises is out there?
Hot new epigraph for daily life: There’s probably a lot of things I should be doing, but I’ve gotten good at not doing them.
Boy, in this issue you have a ramble - and then I lost energy and needed to get on with my day. I’ve written this issue up many times over the past ten years and didn’t do much work to find my previous typing. Also, with a little more searching and finding/paying for some analyst reports, you could probably get something useful.
May I suggest putting this on repeat as the background music for reading the blow?
How much on-premises is out there?
Update, after having emailed: the AWS CEO estimates that there’s only about 10% of IT running in public cloud: ““Essentially, IT is going to move to the cloud. And it’s going to take a while. You’ve seen maybe only, call it 10% of IT today move. So it’s still day 1. It’s still early. … Most of it’s still yet to come.”
How much “private cloud” is there? What rates are workloads shifting to public cloud? For over 10 years, this has been an obsession of mine, and every year it’s difficult to get beyond anecdotes. We all know, feel that public cloud is increasing in use and on-premises IT is lessoning. But, it’s near impossible to prove this in aggregate. I should start with some estimate right at the top. I don’t know, let’s say it’s 60% on on-premises and 40% in the public cloud right now, as of September 2022.
Now, let’s confuse that simplicity by thinking about it.
First, you have to ask “what workloads,” that is, what apps. Paying attention to the types of workloads we’re talking is important because so much of public cloud is SaaS…I think…versus software that organizations program and run on their own. And, also, there’s a tremendous amount of spend on supporting services like networking, managing employee devices, storage, and security.
Types of Workloads
There are applications that organizations buy and run, and there are applications they build and run themselves.
The buy and run category used to be called Commercial off-the-shelf software (COTS) because you would get, like, a DVD off a vendors shelf and install it and run it on your own. You didn’t programming this application, though in most large organizations you would customize it. For businesses, a lot of software used for billing, keeping track of inventory, HR stuff (you’ve probably used Workday in your career), finance (filing expenses with Concur), and collaboration stuff like email and calendaring is this type of software. And then there’s other categories that are industry specific like all the IT used to run factory floors, cash registers in retail, even controls for frying chicken.
There’s also services and management tools: storage to keep all those documents and spreadsheets, management tools to do everything from updating end user devices (laptops, computers, phones, etc.), and then a few broad categories like “networking” (you need the Internet up to do work!) and “security” (both in making something secure and auditing that it is secure. MULTI-CLOUD!
Another category is, broadly, “data.” This is distinct from storage (though, confusingly, storage vendors will place themselves in this category - but that’s sort of like shovel manufactures saying they’re in the home sky-scraper building category) and is focused on doing something with all the data, analyzing it to find out basic things about your business (the UK market has been buying a lot of motorcycle tires this year) and more complex analysis (out of 300 sales people globally, 10 in the UK, there are only 4 who have consistently sold motorcycle tires in the UK at or above our desired quantity for the past five years, furthermore, here are three characteristics these people have [most of their meetings are outside of corporate headquarters at factories, they have industry experience, and they do more frequent deals at smaller sizes than others] that we could try to apply to the other 296 sales people - or just simple analysis like, “our pumpkin return rates have grown 2% per month after switching to lower cost pumpkin farmers last fiscal year). You can also put predictive maintenance here which basically is “your train is about to break, according to past patterns, so you should fix it before it breaks during peak time.” (That kind of analysis would have been awesome to run on the security and baggage handler cut-backs at many airports.) At the moment, the huge growth and valuation of Snowflake makes this an attention grabbing category - rightly so.
Finally, there’s software that organizations build (design and program) and run on their own. There’s online banking, squad management, managing shipping stuff around (logistics), frying chicken(!), and, really, anything that the organization thinks it can do better (cheaper, easier to use [better design], or with features that no one else has) than the COTS alternative…or there is no alternative. You can imagine that the people selling COTS software are eager to change this thinking to “you don’t need to build it on your own.”
Allow me to make a broad sweep of the table: the only thing that matters for this question of how much on-premises IT is left is number four, the software organizations write and run themselves.
Here’s why the other categories don’t matter for the question…rather, here’s why I don’t care about them:
The nature of SaaS (how it is architected, what it takes to manage and run it, and the [often undelivered on] ability to add new features frequently), makes SaaS the best, only conceivable option for COTS software.
Management tools, by their nature, follow the workloads: they will run where the workloads are. More confusingly, they’re run in both places: you can run the “brains” of a management tool in the public cloud, but if you’re managing on-premises stuff, you’ll need to run at least a little piece (an agent, a “proxy,” whatever) on-premises…and a lot on-premises if you’re configuring and managing on-premises software. Pretty much all management startups-cum-mainstream companies try valiantly (foolishly?) to be SaaS only, but about three to four years add on-premises support. Furthermore, while management tools are important and a fine amount of revenue (IDC says something like “$31.2 billion in 2021”), there are many, many, many, many…you get the idea…more applications and workloads out there. By number, management tools are tiny in the overall “all the IT running in the world” (we’ll exclude off-planet IT for now).
Data. What to do with data? To me, this is its own thing, its own category. And, sure, if you were interested in all the money spent on “computer stuff,” you would pay attention here. But…I don’t know enough about data. My sense is that, like COTs-to-SaaS, the fundamentals of how you run your data workloads fit to public cloud so well that the public cloud will become the only option that makes sense (once people get over regulatory and security concerns - remember when putting your credit card into the Internet in the late 90s seems dangerous?). The real reason: I don’t know, I care about the next one.
Custom written software. When you write and run your own software, you have to choose where to run it. Choosing the infrastructure it runs on is incredibly important…rather…thinking through if it matters is important. There are likely thousands (hundreds of thousands) of little applications in the world that are perfectly fine to run on one tiny server somewhere - probably someone’s laptop. These are sometimes called “departmental” applications and are usually used by employees. These little apps have a huge tolerance for being flakey - while annoying, they don’t need to work all the time. There’s are usually really cheap to run: just get some existing hardware that you’ve long ago paid for and run it on that - maybe schedule a reboot of the server every night. And then there’s applications that need good performance (can be used by lots people people at once, are fast [usually this means it has low-latency, you don’t get a lot of spinning wheels while you wait], stay up [less need to reboot the server and wait]), and don’t loose data or the state of workflow use [“I was right in the middle of signing you up for this $50m loan to build that new dam, and then the system went down…so let’s start over again!”]). In those cases, it’s important to think through where the app runs, how much it will cost you, and so on.
Furthermore, most people believe that organizations will use custom written software for Business Success(TM): growing revenue by retaining existing customers and getting more money from them or gaining new customers; reducing costs to increase profit; or coming up with new things that only that company has so they can both grow revenue and charge a lot for (only one company sells 24 hour insurance for when you take your iPhone to the beach…for now).
I don’t know: there’s some McKinsey reports, and some implied connection between doing well at custom software and Business Success in the DevOps reports from a few years ago. Plus, “software is eating the world,” or whatever. Ask your regional IT thought-leader for more.
The workload matters when you’re asking if on-premises is dead, if it all goes to public cloud. It matters for curiosity, but also for how you think through what to do, both as users (buyers) of IT and sellers of IT (vendors, cloud providers, consultancies).
I’ve typed myself into a cul-de-sac. Let’s look at some charts.
What people are saying
When I try to figure out where workloads are running, on-premises or in public cloud, here’s what I can find without paying for anything and after a little less than an hour of looking around.
By kubernetes. To me, the only reasons you’d run kubernetes is to support custom written applications1. Here’s something from our (VMware’s) recent kubernetes survey:
There’s a lot of nuance here. Only on-premises is low and shrinking. But, “both” is harder to figure out. Does “mute-cloud” mean only multiple public cloud, or include on-premises - same with hybrid. (And “edge” usually implies private cloud, at least, not public cloud). Here’s the answer from the report: “More than 40% of respondents utilize hybrid cloud (combining on-prem and public cloud), while almost 50% are now cloud-only. “
Does public cloud usage include using a CDN? So, like, you could run 95% of your application on-premises and just use CDN for speed. I dunno.
Let’s take “[public] cloud-only” as the important figure: 50% of respondents said they run their kubernetes clusters in the public cloud only. But that doesn’t tell us how many workloads are running there. Still, it’s a good indication that public cloud only is…growing?
(I’ll be going over this kubernetes report more in a talk next week, recorded if you can’t make it.)
Next, let’s look at some Gartner stuff:
Gartner predicts that by 2025, 80% of enterprises will shut down their traditional data centers. In fact, [as of Feb 2019,] 10% of organizations already have.
This is from 4 to 5 years ago (depending on when the survey was conducted, analyzed, and then published). How has “shut down their data centers [already] progressed since? You could do some implied point based on the end point of 80% in 2025. I don’t know, sounds like too much work in Excel - I could probably buy a report to find out.
But, because I didn’t buy the extended report and inquiry with analysts, we have no idea what workloads these are. If it’s mostly COTS-to-SaaS, then sure! For the other workload categories, who knows.
Also from that write-up:
Gartner’s 2019 IT Key Metrics Data shows that the percentage of the IT budget spent on data centers has decreased over the past several years, and now accounts for just 17% of the total.
This is interesting, but you have to careful about just paying attention to revenue. That largely represents new workloads, I think. That said, 17% is really low. Because of the principle, “IT never goes away, it just gets fully amortized,” you’d expect it to flat-line at some point…at 5%?
However, if 83% of IT budget is spent on public cloud - that seems like a lot. If there are less workloads in public cloud, those are expensive workloads (subscriptions/opex add up versus one time buys/capex).
Tracking by budget, spend is lower (or existing data centers are highly optimized, fill amortized, and this cheap as a percentage of IT budget): “Gartner’s 2019 IT Key Metrics Data shows that the percentage of the IT budget spent on data centers has decreased over the past several years, and now accounts for just 17% of the total.”
And a few more:
From HPE, May 2022: “Research from IDC shows that on-premises, non-cloud workloads are actually on the rise, growing from 29 percent of total workloads in February 2021 to 35 percent of that number only six months later. The takeaway: Enterprises are now searching for an “optimal balance of workload distribution,” with migration off the public cloud the primary shift.”
And estimate from a financial analysts, from the summer of 2021: “McNealy’s base case for HPE is that long-term on-premises workloads are 30%. The upside case for HPE is that on-premises infrastructure is used for 40% of workloads.”
In the same, about 25 to 30% of workloads are on-premises in 2021: “Research from IDC shows that on-premises, non-cloud workloads are actually on the rise, growing from 29 percent of total workloads in February 2021 to 35 percent of that number only six months later.”
And so forth.
I’m just saying: It’d be cool to have some more surveys and models.
I get obsessed with this question because, increasingly, my feeling is that “everyone” wants the benefits of public cloud but has reasons for staying on-premises. I’m not sure most of them are good reasons - just “because it’s what we’ve always done, know, and trust” is likely the number one motivation.
What falls from this is that a lot of the toil people struggle with comes from having to run on-premises. You have to take on managing everything below the application, which is a lot of time, money, and attention spent on something other than the application. Or not! Maybe it’s worth it - even required! - to do many things on-premises. It’d be good to get some charts around how that’s going.
$200 of a conference that cares where the app are running
If all of that is of interest to you, you’ll be interested in our annual conference, SpringOne. It’s December 6th to 8th in San Francisco. We’ve posted a lot of the talks, including one from me if you can find it.
When you register, you can use the code COTE200 to get $200 off!
The Mysteries of Mondrian
In Apple News, in The New Yorker.
“By not wishing to say anything human, by completely ignoring oneself, the artwork becomes a monument to Beauty: transcending the human; and yet human in its depth and generality!” That’s as mordant an aesthetic verity as I know, but Mondrian’s guileless confidence in being understood is touching.’ I don’t what to make of that chunk of text, but it’s fun.
“Composition with grid 8: checkerboard composition with dark colors,” Piet Mondriaan.
Meanwhile, from Charmion von Wiegand, “Sanctuary of the Four Directions”:
And then this context to understand everyone’s mood mid-century, especially Europeans: “Mondrian arrived in New York, by way of England, in 1940, on a convoy that had lost five ships to German submarines en route.”
“The one handed swipe and wipe while holding phone.”
As Jorge Luis Borges said, “If a book is tedious to you, don’t read it; that book was not written for you.
“[The song ‘Chaise Longue’] was written in 2019 in a single day by vocalist Rhian Teasdale while she was sitting on an actual chaise longue…”
Meanwhile: “Douchebag Chet (Bill Paxton) checks out crazy hot Lisa (Kelly LeBrock) and says he’d like to ‘Butter her muffin.’ This refers to finding her sexually attractive and the desire to have sexual intercourse with her. Lisa then transforms Chet into a pus spewing massive turd-like creature. One of the greatest comeuppances that ever comeuppanced.”
We should start versioning The DevOps, so that every year when is thought lords & ladies write the “DevOps doesn’t work” piece, we’re forced to remember that version 1.0 to 4.0 actually did make things better, and version 5.0 is still keynote vision.
“The trouble is writers write from pain and Harold has closed up all his wounds, so all he can write now are these small things.”
“There is such political naivety in the US that it only takes one image of five Palestinians dancing in the street to obliterate the bigger picture.” This and previous from Alan Rickman’s secret showbiz diaries: the late actor on Harry Potter, politics and what he really thought of his co-stars
“Night Drive Beat”, “Commercial Boom Bap Ener…”, and “LOUNGE MELLOW CHILL POP.”
Stuff from me
October 5th - Escaping the Legacy Trap. I just did a (hopefully final!) review of the booklet we’re basis this talk on it - it was good! This is a webinar, so you can register for free, watch it live, or watch the recording. Whatever the case, sign up for it!
October 6th - State of Kubernetes read-out - I’ll go over our recent survey about kubernetes usage.
A long discussion with my co-worker Nate about how I got into computers and my ongoing quest to find out how kubernetes “won.”
National Coding Week 2022: addressing the digital skills gap - Three things I like(d) about coding.
Relevant to your interests
Announcing the all-new Substack Reader for web - Also includes subscribing to RSS. One day, perhaps I’ll check it out. But Feedly plus Newsify is deep in my bones at the moment.
Anything You Want, Derek Sivers - the follow your dreams approach.
Terry Pratchett: A Life With Footnotes by Rob Wilkins review – anecdotes, elephants and ‘an embuggerance’ - ‘Writing is an isolating job – “a life of homework”, according to my next-door neighbour.’ And: “keeper of the anecdotes.”
Cleaners at Amsterdam gallery ordered to let insects run wild in name of art - “Three months before the exhibition opening, cleaners were asked not to remove spiders and their webs. I’ve been going around weekly ever since to spot where webs started to appear – a very different way of looking around the building that I know so well.”
PaaS Is Not Dead - good stuff if you’re constantly asking why PaaS isn’t more successful. As always, the general answer is “are you sure it’s not?”
Speaking of: Building for the 99% Developers - I wish there was some framing that make the “1%” the weird ones instead of the normal people the ones who are thought less of.
This isn’t entirely true - I think there are COTS things being written to run on kubernetes…sort of weird: why wouldn’t you just make that a SaaS. Well, cause you need off-the-internet (air gapped) apps or apps that are more realtime (factory floors), or need to work off-line (in-store registers and inventory management). ↩