There’s been a lot of talk the last few days about Facebook’s Open Compute project, where they have published info about their servers and data centers. It’s interesting reading. But, arguably, not specifically relevant for many folks.
Say you’re a startup. You’ve built the next great thing, you’ve got a few beta customers, and it’s time to get ready for real use. You’ve dabbled with shared hosting, and have vowed to never do that again. You’ve thought about virtual private servers, dedicated/managed hosting, and cloud services like EC2 or Windows Azure. But in the end, you’ve decided you’d rather own and operate your own servers. So you need a home for them.
Unfortunately, there’s not a ton of guidance out there for you at this point. There is a lot of superficial advice that Google will point you to, but not a lot that’s very useful for the startup CTO. What should you look for in a data center?
This article will talk about some high-level things to think about. Parts 2 and 3 will dig in further into what’s really important, and how to plan for the future.
You get what you pay for
Way back in 2002-ish or so, before I started NewsGator, I had a couple of servers at a mom-and-pop colo shop. I think it was like $50/mo or something for a couple of servers – just insanely cheap. It was basically a small office suite, with maybe 5 racks of equipment. It seemed a little warm in there, and the only security was the lock on the suite door (meaning I could get to other folks’ servers). I wasn’t there long, but a friend was – he told me a lot of stories about his servers overheating, and even one time when his servers were down and unreachable for a whole weekend without any notice. He later found that the company had put the entire rack he was in (shared with others) in a truck and hauled it across town to a new facility. Yikes.
What do I need?
You really want to find a company you trust; someone you can call a partner. You’re putting a lot of faith in them, so choose wisely. If you can afford it, go with a top-tier hosting facility in your area. If you can’t afford it, think harder about whether you really want to colocate. A few high-level things to think about:
- physical security – who can get into the room your servers are in? How? Once they are in the room, could they get to your servers?
- network – make sure you’re comfortable with the internal and external connections to the net, and how traffic is managed in the event of something going down.
- bandwidth – think hard about what you need, both in terms of sustained and burst bandwidth.
- physical space – is there room for you to expand as you grow?
- cooling – your equipment doesn’t work if it gets too hot.
- power – can they bring in enough power for your equipment? What happens if there is a grid power outage? How often to they test their generators and other power equipment? Have they ever had an actual power outage?
- maintenance – sooner or later they’re going to have to do maintenance on, say, the router you’re connected to.
- SLA – they can’t guarantee your site will be up, but they can guarantee that you will have power and network access. What does the service level agreement say?
- company – is the company in good financial health? You don’t want to have to make an unscheduled move.
Now, when you look at a list like this for the first time, you’re probably thinking things like network/bandwidth are the most important to worry about. I did. I spent a lot of time worrying about that. But what I found was that ended up being the least of my worries.
This has all been pretty high-level so far…but in part 2 tomorrow, we will look at network and bandwidth; then in part 3, we’ll talk about the big daddy of them all, power and cooling.
Pingback: Choosing a colocation vendor – network and bandwidth (part 2) | Greg Reinacker's Weblog
Pingback: Choosing a colocation vendor – power and cooling (part 3) | Greg Reinacker's Weblog