When I started in the hosting industry ten years ago, it was worth about $40 billion. That’s pretty impressive bearing in mind the internet was born just 20 years earlier. Today it’s valued around $550 billion. Its staggering growth shows just how much the hosting industry has expanded in the last decade.
In that time, we’ve seen the explosion of cloud hosting and the cementing of market share by the three big cloud hosting providers - AWS, Google Cloud and Microsoft Azure. We’ve also seen bare metal hosting providers modernize away from their traditional inflexible approach, long-term customer agreements and slow provisioning to far more scalable, flexible and cost-effective services.
I’ve experienced these transitions as both a customer and a hosting provider, and I’ve learnt some important lessons along the way.
So, to celebrate a decade in the industry, I thought I’d share some of the lessons I’ve learnt. Before I kick off, however, I think it’s important to caveat that these are very much my experiences and therefore come with a certain amount of bias. But I hope that as a hosting customer or buyer in 2023, they prove useful as food for thought, a discussion point or to provide a flavor of the hosting industry then and now.
Here are my six lessons:
The adage “you get what you pay for” could’ve been written about cheap infrastructure. What happens when infrastructure goes down? Your services or operations also go down, leading to loss of money and potentially reputation. Are you more likely to have regular downtime with cheaper infrastructure? Of course you are.
But when customers are buying infrastructure, the question of downtime and how many periods we as a business have experienced, almost never comes up. No one seems to consider the long-term cost of maintenance when purchasing infrastructure.
As a customer, I urge you to ask any infrastructure hosting provider that you are considering working with - “how many maintenance-driven downtime instances did you have this year?”. Also, look closely at what hardware that provider is recommending for you. Do they offer just one path from your server out to the internet? What happens if that path is disrupted at any point? Your server will go offline. So, specify multiple ISPs, core routers, switches and network ports to build in as much redundancy as possible.
It will cost a little more upfront but will cost your business far less in the long run.
Despite infrastructure being business-critical, it’s surprisingly undervalued and that’s shown in the way that many companies buy it. Specifically, who buys it. You may or may not be surprised to learn that the evaluation stage of buying infrastructure generally falls to junior employees who are not trained in buying and don’t have the wider business context. They are also often biased towards a certain supplier, which tends to be one of the hyperscale providers.
The well-known saying ‘nobody ever got fired for buying IBM’ could be updated today to ‘nobody ever got fired for buying AWS’.
Really, companies should have a dedicated procurement team or even better, a c-level team that works closely with the procurement team. And they need to invest in evaluation from the outset. Considering all options including non-hyperscale, the wider needs of the business and what it might require the infrastructure to enable 3, 6, 12 months down the line. As the industry and behaviors shift, so too should your infrastructure. But if you don’t evaluate regularly, how will you actually know where changes need to be made?
When purchasing hardware and looking at multiple different hosting providers, it’s tempting to want to make like-for-like comparisons based on a list of components.
Unfortunately, comparing providers in that way doesn’t work because the quality and performance of the components each one uses can differ vastly. Take CPUs. If you compare the same CPU from Intel or AMD to the same CPU elsewhere, there should be very little difference in performance. But when an infrastructure supplier offers desktop/consumer grade CPUs in a data center environment - which they are not designed for - the contrast is obvious. There are also cheap and expensive SSDs. As expected, the performance of cheap SSDs is significantly poorer than the more costly ones.
Likewise, budget suppliers will often cut corners when it comes to networking, especially in areas where customers are unlikely to look. Servers are offered with port speeds of 10Gbit/s or more, and customers assume that they will be able to use almost 100% of that available port speed at the same time. But – and this is where providers can be sneaky – the connection capacity between the top of rack switch and the distribution switch is only 20G, causing packet loss and disruption to services.
So, when you are next buying infrastructure, throw away that list of components and instead talk to the hosting providers about your business goals and how you need your infrastructure to support you achieving them. And expect that the packages they offer will vary between providers as different price points in the market.
If you are not doing a proof of concept (POC) with at least the final two providers in your decision making process you’re doing something wrong.
Yes, they cost money and time, but they are worth the investment. Particularly, if you go beyond just the technical aspect of the partnership and look at what it’s like to do business with each hosting provider.
So yes, measure the network performance, application performance on an exact CPU and the disk set up with specific performance targets in mind. But also test each providers’ invoice process (does my accounting team like the format of the invoice), communications (how long does it take for your account manager to respond to a query) and importantly, support. Break something on purpose, put in a ticket or message the support line and see if their response tallies with what the salesperson has promised.
The ‘business’ side of the customer-provider relationship is the most variable but also what you will be interacting with on a daily basis.
No matter what the technology is, it sometimes goes wrong. Even the mighty AWS has had its fair share of failures. I can safely say that I have never worked anywhere that hasn’t experienced an unexpected outage.
How hosting providers respond when these outages occur can make all the difference to its impact. During my decade in the industry, I’ve seen support from infrastructure providers transition from humans to bots. And don’t worry, this isn’t going to turn into a bot-bashing rant because they absolutely have their place, but when it comes to something as business-critical as infrastructure, its humans customers want to be able to talk to.
To know that there is someone on the other end of the phone, WhatsApp, Telegram - whichever your platform of choice - who can talk through the problem, reassure you that something is being done to rectify it and retain communication throughout is priceless.
Unfortunately, it’s often not the case because it's more resource intensive for the supplier. But I believe - based on my experience as a customer and what we hear from our customers - that good quality, human support is worth the money.
If infrastructure was music, the cloud has topped the charts in recent years. Its promises of scalability, efficiency and lower infrastructure costs have had businesses flooding to take advantage.
It’s only recently that the reality of those promises is starting to come out as companies working with hyperscale cloud providers have found themselves locked into proprietary technology that is costing far more than they expected.
They absolutely still have their place, but I believe smart infrastructure buying today should take a hybrid approach, for two reasons.
Cloud that follows the daily curve of usage is never going to be as cost effective as bare metal. In fact, it could be as much as 10x the cost. Putting in place a baseline of bare metal with cloud auto scaling allows scaling to take place, but if the peak stays consistent, it can be backfilled by bare-metal, reducing reliance on the cloud and therefore reducing spend.
Splitting your infrastructure across different vendors from both bare metal and cloud, increases the resilience of your infrastructure. If your bare metal provider has an outage, for example, you can move those workloads over to a cloud provider for the time that the outage is taking place and vice versa.
As some of my favorite lyrics say, “I’ve got a lot to teach but even more to learn”. I’m excited to look back on these lessons in 10 years’ time and see how they will have likely moved on. Because, if there is one thing we can guarantee in the world of hosting, it’s evolution and growth.
But in the meantime, I’d love to hear any key lessons that you’ve learnt from your time in the industry - whether that’s one month or 15 years - so please get in touch directly.
With a decade in hosting and video game hosting, Isaac knows the big mistakes and how to help customers avoid them. Dog dad to Lilly, he owns all the tools and lives his best lawn life.