Which means: don't do it yourself, unless you're an expert, which you are most likely not. Suck it up, it costs more money, but when you're just starting out and can't hire people dedicated to managing your servers, you should just pay another company to do it, otherwise you're just being penny wise. Downtime costs a lot more.
There's a balance. Also one of the traits startups need IMHO is the ability to 'become an expert quickly' if you need to, and if it makes sense time wise.
VPSs are great, but I've never got the point of things like engineyard where you're paying masses to get people to do sysadmin.
There's so many reasons a site can be down, you can't just move all responsibility to another company - at least I don't think that's a good idea at all. By all means move hardware to them, net connection to them. You don't want to worry about power outages, etc
And that's where you make a mistake: sure, the founder can be a competent sysadmin, but the founder simply cannot replace a team of sysadmins. You need to be available 24/7 -- there's always that pressure on your social live that you need to remain available, etc etc etc.
You think you can do it, as a founder, especially when you've got some sysadmin skills, but you simply cannot do as good a job as a team of operations people who are stand-by 24/7. And you simply cannot afford such a team for yourself when you're starting out.
And that's where you make the mistake. Look at twitter. Look at facebook. Look at anything else. They're all broken from time to time, all offline every so often. People are pretty forgiving especially at the start.
I'm reminded of a talk Chris Wanstrath gave where he said:
"Would you pay $100 an hour for an untrained accountant? Because if your consulting rate is $100 an hour and you do your own accounting, that's exactly what's happening."https://gist.github.com/0a2655aed6a26fa15a02
So I guess the relevant questions are:
1. How much is your time worth?
2. How much are you willing to pay for an untrained sysadmin?
Totally agree. We have local test setups, but even our dev site is on Rackspace (we also have remote virtualbox servers on our machines now which can mimic the main and dev sites).
I'm logged in as root now trying to make sense of what's choking it.
I just nuked about 100 http processes that were just sitting there. We'll have it up this afternoon I suspect (the guy that set this up is much more familiar with the framework than I).
I guess the question is why are people still building/managing their own servers for startups? EC2 and Rackspace cloud is cost wise equal, with the ability to have cold standbys and somebody else manage the infrastructure.
The only case that I see for having your own machines is for a development environment... Though if there was a reputable firm hosting SVN/Trac/Bugzilla etc. one might not even need to have anything more than boxes for compiles -- in this world it's your macbook..
I submit that this is an overly simplistic view of what "the infrastructure" is, a view common among those who write software: the infrastructre is anything that isn't the code they write.
There are very few scenarios where a non-trivial site can be kept up (that is, not "dysfunctional or down" as the OP mentions early on) with merely application code and generic, undifferentiated, off-the-rack infrastructure.
To scale even modestly, one must customize. The customization can start with the web framework and can eventually lead to hardware components (though the "lowest" I know of is Google's custom mobos). There are also operational considerations, such as
I think there are two separate questions for any enterprise. The first is, is further customization an optimization. The second is, is the optimization premature?
To suggest that, for all startups, the answer to the first is always "no" and/or the answer to the second is always "yes" would be overly broad.
For the folks that have created many web apps many of these questions have default answers based on experience but for a first timer like me I can't afford to learn by failing at every fork. I need to minimize my failed choices by searching for external input from people that have been through it before.
Great point. We chose Rackspace.
But it was more our software and database changes that we're getting a handle on. We haven't even looked at what many concurrent users do to our site yet. It needs to get shaken down, in a big way.
