Actually, I agree with you, being the sysadmin you mention. Unfortunately, the up-front cost seems to be a barrier to many a startup founder. To me, this suggests a lack of good leasing options, though it could be that's because it's not worth making a business out of it.
I understand you don't have to worry about hardware issues
I've always found this assertion to be misleading at best. One still suffers the consequences of hardware failure (both inherent and human error), yet one has very little, if any, ability to prevent them. You're at the mercy of the providers.
It just doesn't make sense outside of well funded startups or established businesses.
I'm not sure it makes sense inside them, either. I've found that the break-even point for buying versus cloud is consistently around 10 months. The rule of thumb for when the transition from cloud to colo makes sense for a startup that doesn't yet have any of their own hardware is about $10k/mo, assuming they're growing at a rapid clip.
The only situation where it would make sense is compute-heavy (but light on memory and I/O) loads that peak much higher than the average. I have yet to encounter such a one, since it's tough to meet the parenthetical requirement, thereby bypassing the Deutsch distributed computing fallacies.
It's not actually that costly to own a colocated server. A decent server will cost you around $2000 which will include 8 cores & 24gigs of memory. You can colocate it for around $60 a month, which is about $227 per month for 12 months, but is approximately equivalent to 12 small amazon instances.
Colocation is still useful in startups if you want to create/delete instances all the time in a development type environment where you're testing out the infrastructure that integrates between VMs.
So if you don't ignore the upfront cost, you break even at the 3 month point using his numbers. His argument is every bit as strong regardless of the shape of your cows.
I do agree with the OC, but I also think his situation leans unusually in favor of the colo option.
That is, 3.7 month break-even is a much stronger argument than 10 months, so I'd be interested in hearing what others have found to be the break-even point with various cloud providers.
I use boxes in the cloud for various things, but not yet for hosting. We have a couple big fast machines in a cage, of the sort that would cost a grand a month if they were on EC2, that we run all our sites from.
Once you have a single project that justifies the colo box, the marginal cost of starting a new project with a new website is zero. No need to configure a $20/month slicehost box and cross your fingers that it can handle a Redditting. You just put it on the monster box that can handle anything you throw at it until it demonstrates that it needs a big machine of its own.
Assuming perfectly spherical cows..
Actually, I agree with you, being the sysadmin you mention. Unfortunately, the up-front cost seems to be a barrier to many a startup founder. To me, this suggests a lack of good leasing options, though it could be that's because it's not worth making a business out of it.
I understand you don't have to worry about hardware issues
I've always found this assertion to be misleading at best. One still suffers the consequences of hardware failure (both inherent and human error), yet one has very little, if any, ability to prevent them. You're at the mercy of the providers.
It just doesn't make sense outside of well funded startups or established businesses.
I'm not sure it makes sense inside them, either. I've found that the break-even point for buying versus cloud is consistently around 10 months. The rule of thumb for when the transition from cloud to colo makes sense for a startup that doesn't yet have any of their own hardware is about $10k/mo, assuming they're growing at a rapid clip.
The only situation where it would make sense is compute-heavy (but light on memory and I/O) loads that peak much higher than the average. I have yet to encounter such a one, since it's tough to meet the parenthetical requirement, thereby bypassing the Deutsch distributed computing fallacies.