Comment by DaiPlusPlus | Hacker News Reader

Comment by DaiPlusPlus | Hacker News Reader

Parent Full thread

Physical co-lo is only cheaper if your time is worthless (ditto for Linux-on-the-corporate-desktop, or anything else that’s nominatively gratis).

My current SaaS gig runs in Azure (App Services and containers - no VMs nor “serverless”) and being completely free of the need to do sysadmin work is just so nice. When I was colocating my own rack about a decade ago there was always something that needed my attention, be it security updates, some new unexpected error in the syslogs, RAID drive failing, etc - that I’d need to commute to the DC an hour away or pay the DC’s helping-hands fees. Back when I was younger it was fun and valuable to gain those experiences, but right now I’m a west-coast SWE and no employer wants me to spend my time doing sysadmin/SRE work when my time is literally worth ten times more doing product work and letting MS’ Azure staff keeping things humming along.

For high-availability, AWS/Azure/GCP all the way! There’s no way a micro-ISV SaaS vendor can compete if they have to burn cycles doing unprofitable infrastructure work.

> Physical co-lo is only cheaper if your time is worthless

Physical colocation is nearly always cheaper than AWS for most use cases.

Exceptions are when your usage is so tiny that it doesn't need even a single full-time server, and, when your usage is so erratic that provisioning for the spikes doesn't make sense.

For everyone else, you're paying a very large premium for AWS over what it'd cost to buy and operate some servers.

Or when you need anything more than to run a few very basic services ( like a PHP app with apache and mysql).

AWS have managed services for a lot of different things, which save a lot of time on maintenance. Object storage? Message queue? Any type of database? ML? etc. etc. etc.

And of course, there's the scalability. With on-prem it's a careful balancing act of buying enough capacity to have space to expand but not too much to waste it, and you have to regularly keep an eye out lest you hit a limit with multi-week lead times ( e.g. you're out of storage, and buying a new shelf can take weeks to be usable).

Look, i never said aws good/bad, i just said do the math and don't trow your money blindly at some IaaS/PaaS/SaaS etc.

That's really the only answer. Do the research and the math for your own business/use and see if AWS is a better value than colo or your own hardware.

Oh certainly if you want some beefy server with graphics cards, tons of memory etc, running 24/7, it's probably cheaper to co-locate. A g3s.xlarge costs $13k for 3 years with amazon, I suspect kitting your own server out and hiring 1U of space and a 1G internet access port would be much cheaper over the 3 year contract and not run into nasty bandwidth surprises.

The benefit of AWS-style comes in at

1) Flexibile workflows (I need the grunt on mondays, but not the rest of the week)

2) Tiny workflows (I just want a couple of $5/month VMs in different cities and some DNS to host a couple of toys)

Now if you have the $8/hour self drive, that's great if you use it once a year on a 10 hour trip to a holiday destination. It's awful if you use it for 2 hours a day.

Again, you're completely discounting the ops time involved in "operate some servers" - how many devops/SRE folks do you need to employ to do that? And 24/7 on-call?

I need the exact same devops team I have today. No big difference to them where the boxes sit and who owns them. It's the CFO who sees the bigger difference.

I do save the few staff hours needed on physical maintenance of racking, cabling and replacing hardware. But at typical startup scale that's not much work. In the past I've done it myself on the side. As the company gets larger it takes dedicated staff, but at that point the comparable AWS bill is so large I could pay their full year salary on a single month of AWS bills.

That why i wrote "medium business (traditional ones)", the HW stuff can be done by the DC peoples, one HD as spare (online) and one offline, if HD fault, send email "please exchange HD 5". If you can safe 20k a year it's even probably worth your time...well 60 minutes of it...but maybe not if your one of those rock-star programmers ;)

Reply on news.ycombinator.com