It’s The Web Project Guide Podcast, and we made you a new episode.
I’m lucky. I never have to think about web hosting.
I’m lucky because I’ve really only built small sites that can rely on one-touch hosting — through a domain provider, or as a part of the overall service (like Squarespace). I’m also lucky because, in my real day job, we have people who figure that stuff out. People who understand on-prem vs. cloud. Who know the actual politics of “five nines.”
Thankfully, we also know people who can talk about those things — namely, my co-author Deane Barker and his colleague Elias Lundmark, product manager for cloud hosting over at Optimizely. Hosting is still a weird topic with its own language, but it’s also made a lot easier to understand thanks to this conversation.
Hey, let’s talk about site uptime because I think it’s very sexy and glamorous to ask for an inordinate number of nines.
I wrote about this in the book: when I was at Blend, we would get RFPs all the time, and clearly the RFP had been passed around to all of the stakeholders in the organization and the CIS admin infrastructure person added, “Well, we’d like five nines uptime.” And so we would dutifully scope that out, what it would take to keep a website up for five nines.
And five nines would be 99.999% uptime, right? So this is something like 30 seconds of downtime a year, and whenever we’d price this out, I mean we would do it. But the numbers would just be stunning. I mean, to keep five nines of uptime, you have to have multiple redundancies across multiple different regions on different sites of the world.
When you talk to people about site uptime, do you think that there is too much lip service paid to things without being cognizant of how much this is all going to cost? It’s very sexy and glamorous to ask for crazy hardware, it’s also very expensive.
Very much so.
I start to think back around the time I was studying computer engineering and just solving for something like storage and that sort of availability — because spinning hard disks tend to fail quite a bit. Just handling that, both redundancy and backup strategy, and the disaster recovery if something really goes wrong, is that you come into setups like active-active, which are really expensive to build on-premise.
Yeah, that’s definitely where the cost of something starts to escalate. And when you have that requirement of, “All right, I need five nines,” then it’s certainly going to get expensive, it really is.
But then it’s also why public cloud providers are so good at that — they have so much redundancy built in, just because of the sheer amount of hardware that they have. It’s really easy for public cloud providers to provide “five nines,” whereas if you’re building something that is tailor-made for a specific application and doing that on-premise, it’s going to be much more expensive.
You can listen below, but we’d love it even more if you would subscribe wherever you can find podcasts, or at the following easy links: