Why We Virtualised
Introduction
In 2025, IT infrastructure is increasingly virtual. This trend has been going on for a long time, and there’s a reason - cost.
Today we’re seeing virtualisation in networking, storage and other areas - but here I’ll be focusing on servers.
I’m writing this to correct the common view that virtualisation was pursued because of under-utilisation. We were only using 15% of most of the computer resources that we had, and virtualisation allowed us to be more efficient.
This is true; most machines are idle most of the time, waiting for work.
But it’s not the whole truth. It’s the headline that we used to give to Executives to get them to sign off on what were typically very expensive projects - the RAM and storage requirements for virtual hosts are high. Early virtualisation migrations were typically costing an order of magnitude more than the typical IT infrastructure project within an organisation. They therefore often had to be cleared at very high levels.
And the thing about Executives is that they’re not renowned for their humble modesty. They’re more likely to be in the role because they’re more than capable - and alarmingly willing - to tell everyone about their virtues and good works.
Which is how we get to “We virtualised because we were underutilising our hardware, it saved us loads of money”, with the silent part being “and I’m an awesome Executive, please give me lots of money.”
Savings, Savings, Savings
The headline was that we’d save money by solving underutilisation. This was the headline because it had very real fiscal effects. Once you’d sunk your {tens of thousands|hundreds of thousands|millions} into virtual hosts and migrated all your servers onto them, you were using a lot less electricity. This was a real, tangible, and almost immediate saving.
But there were other savings to be had. Ones that were more consequential for IT, albeit perhaps not as obvious on the budget spreadsheet.
The two points I’m going to offer are:
- Cabling infrastructure
- Patching
There are others, but these were the ones that were closer to my heart; I’m sure network or storage specialists will have their own points of view.
Cabling Infrastructure
The cabling infrastructure reductions from virtualisation are hard to understate. Every server required at least two cables - power and data. Many servers would have redundant power supplies so that they could keep running if one failed; that’s quite standard for all but the lowest specifications of server.
All this mounts up. The data cables also need to go somewhere, usually into a switch. Which also needs power.
Did I say that each server only required one data cable? I’m sorry, I misspoke, they need two. The other one will be for what’s called “out of band management” - basically the ability to completely control the machine remotely, even down to powering it off or on. The alternative to this would be to have to start adding in cables for the keyboard, mouse and monitors, with those cables going to some kind of KVM switch. Which will also require a power cable.
Prior to virtualisation, server rooms fell into two broad camps; the ones where cabling was neatly run and tied together, and ones that resembled some kind of evil creature’s nest.
Virtualisation required better networking, but in addition to reducing the number of servers you were also reducing the number of network switches and your power cabling, and at larger scales you’d also be reducing the number of UPSes that you needed.
Patching
Today patching is something most computer users are aware of. We’ve settled into a neat cadence of monthly patching on most machines.
Servers have slight complications for patching, but it’s mostly a scheduling job. Everything needs to be patched outside of normal usage hours, and you need to make sure that cluster members are appropriately sequenced.
Except it’s not that simple. Because it’s not just the application and operating system that needs patching. There’s also the drivers for the network card, for the video card, for any storage controllers or RAID arrays, and more.
You can make this simpler by standardising your server fleet - only buying from certain models, which you know have certain hardware. Of course, the higher end machines will have a different set of network controllers or storage adapters, but at least you’re minimising the variations.
Except that over time not every service upgrades at the same time, and before you know it you now have two or three generations of hardware in your server room. Which translates to three or four of each type of hardware at least.
In 2010 I was working for a large company that was aggressively virtualising its server fleet. When I spoke to the Wintel team about it, they said patching was the main internal driver. They had a huge spreadsheet which showed what type of hardware each server had, and if a new driver came out they had to filter it and see which machines would need patching. Two people were employed almost full time doing this - they actually tried to rotate the job around the team over time to protect people’s mental health. It was described as “soul crushing” and “feeling like a punishment”.
By contrast, virtualised machines have very standardised hardware - whatever virtual hardware the host presents. Better yet the drivers for them are often simply included in a package that’s simple to update such as the VMware Tools package, or are part of the Linux Kernel drivers set.
Suddenly the only pain points are patching your hosts, which you have far fewer of. And you can do that by live migrating guests across to another machine, patching that host, and then live migrating back. It’s much easier to schedule this so that the patching and testing of the hosts can now be done during the day, and migrations to/from done out of hours.
It’s hard to describe how much better patching is with virtualisation. It’s not a panacea, but it’s just so much better than where we were beforehand.
Conclusion
Virtualisation was attractive for many more reasons than just utilisation. I’ve not touched on things like how it made backups better through snapshots, or how it made storage allocation and management easier. That’s partly because I feel those are advantages that came after we virtualised, rather than reasons to virtualise.
But it’s definitely the case that it was way more than just utilisation. That was just the headline we gave for the executives…