YC W12 startups' hosting decisions

GoDaddy still has a lot of mindshare, while far fewer people have actually heard the horror stories. And at least some of the people who hear the horror stories will figure it won't happen to them, or that it only happens to people doing something wrong, the same way they figure a PayPal horror story won't happen to them.

If you're referring to their domains, GoDaddy's pricing is competitive with the industry.

More than pricing, I thought using GoDaddy was frowned upon after the whole SOPA fiasco. I'm surprised so many have decided to use them.

There are a few big downsides to using Gandi as a registrar:

https://news.ycombinator.com/item?id=3405177

In addition to the problems outlined in that post, you're additionally subject to the laws of France and must pursue any relief in Paris [0]:

  You agree that You have been informed and expressly accept
  that if no amicable solution is found, in accordance with current
  French law, and unless there exists a clause to the contrary, any
  dispute concerning the validity, interpretation or execution of this
  Contract shall be referred exclusively to the competent jurisdiction
  in the region where Gandi is registered for the providing of the
  service  concerned,  and  therefore  at  location  of  Gandi  SAS
  (Paris).

I'm surprised more Y Combinator companies don't use EasyDNS: http://blog2.easydns.org/2008/07/08/5-questions-for-ycombina...

[0] http://www.gandi.net/static/contracts/en/g1/pdf/domain_name_...

If you had $164k (plus $1-2mm more usually), why would you use a lousy (but free) CA or a mediocre (paid product) CA? I mean, I use them for some specific non-web URLs, but the $99-499 to get a real (shorter/smaller file size, EV, wildcard, multi-cn, more widely present in browsers, or less broken intermediate certs, or at least business-verified cert), is totally worth it.

For DNS, I'd just use OpenSRS through a friendly registry reseller; I'm sure at least one YC company is an OpenSRS reseller.

I did a quick bit of checking and couldn't find anyone (like you).

My gut is telling me that few will open data centers in Canada when NJ/NY and Washington State are right there across the border. Latency won't be a big driver for that.

Would a EU/Australian data center do?

Ninefold has one for Aussies, and I bet if you looked there's competition available in the EU.

Also, see if you can't get Amazon S3 to store only in their Ireland data center.

If you use a US company then they are subject to the Patriot Act no matter where the servers physically are.

Canada Cloud Computing would be able to tell you.

> Is this is a list of what is most likely the cheapest options? The options with the fastest setup? Lowest commitment?

Comparing server-hosting options (dedicated, own hw+colo, vps, cloud) merely by paid $$/month, dedicateds and owned hardware + colo-housing always wins (in my experience). You can get a HP DL120G7 w/ good cpu and 16gb ram, incl housing and a TB of traffic for ~ $300/month[0]. Last time i checked the prices for an aequivalent offering on amazon (and others) was way beyond that (excl traffic). A startup could easily boostrap with mentioned hardware. And you're the only one on that hardware, you're pretty much guaranteed the power that you're promised.

Now if you want more than one machine, you'd probably put some virtualization on your server. For that, you need someone on your team who is able to and can spend the time to setup some virtualization on your HN. And care about it. (Can be lot of work or very little, depending on how much time/effort was spent in the setup process). If you don't have that person on your team, you'd have to pay someone to do it. This can push the costs significantly, maybe beyond those of the virtual competition.

Another benefit of VPS/cloud is that you never have to care about the hardware running your boxes. A harddrive dies? Not your problem, and you probably wouldn't even notice. If you have your own hardware, you could protect against downtimes due to hardware failure by having critical server-parts redundant, e.g. hw/sw-raid, redundant power supply. If you put some service-contracts with the hardware-vendor or the datacenter, you can have your failed hardware hot-swapped by someone in a decent amount of time (depending on what sla you're willing to pay for).

Fast setup is usually better on vps/cloud providers. Having fast setup times with own hardware depends on you choice of datacenter/provider i'd guess. They might have pre-setup/cabled machines ready to be activated within short call. If you need some more power RIGHT NOW and you're running on dedicateds and for some reason it takes 2 days to get the new hw, that i'd probably spin up a cloud machine temporarily. Preferably a prepared image that's usually shutdown and taken up for such a scenario.

please mind, that comparing those soultions to heroku is a different story. Heroku offer way more service and take away most of the sysops headaches from you. Of course that's a little more pricey, but again: for any server-instance based solution (vps, cloud, dedicated) you need _someone_ setup your httpd and a deployment of you app.

So back to you're question "What does this data tell us?". The hosting diagram tells me personally, that choices were made based on not wanting to spending time/money on sysops and/or more trust in cloud/vps solutions.

[0] http://www.leaseweb.com/en/dedicated-servers/configurator-cu...?

I think that's the biggest problem with data these days. Data is easy. There is tons of it, it's cheap. Analyzing, understanding and conveying that information is very hard though.

how would/could you derive that?

I'm not. My personal experiences with Rackspace (servers and hosting) have been pretty terrible. When things are working, it's OK, nothing monumental. But when something goes wrong you get poor communications, excuses, and low priority.