Gartner: Stop worrying and love the cloud, with all its outages and lock-in | Hacker News Reader

Gartner: Stop worrying and love the cloud, with all its outages and lock-in

theregister.com

12 pointsLinuxBender11 comments

I disagree. Cloud _anything_ is extraordinarily expensive.

We've started colocation in local datacenters for our non-critical systems and haven't looked back. Yes, these non-boujee datacenters do have occasional outages. In fact, just a single outage in two years, involving a power feed, which lasted an amazingly short 36 hours before it was fully restored.

Did we care? Not really. These are data warehouse, batch, large query, and ml workloads for analysis. The hardware costs the same as one month of cloud providers and has already paid for itself in savings multiple times, including salary cost.

A 36 hour outage is “amazingly short”?

Notice that the data center isn't actually facing the user:

> These are data warehouse, batch, large query, and ml workloads for analysis

Assuming the average load is 80%, it would take ~7.5 days to catch up with all the work accumulated during the outage. In a long run, that's pretty harmless.

if you can handle a 36 hour outage, probably not the target for the cloud.

36 hours without access to medical records would probably cause some downstream issues.

For a _power failure_, yes! It was actually a pretty wild failure.

They had a litreal Derecho blow through, and it knocked some neighboring power lines down (that weren't attached to the datacenter) against their outdoor switch gear. This caused power that wasn't supposed to be there to be fed into their feeds and blew out some transfer switch gear.

Their main feeds are actually underground, but never did they think other powerlines hitting their equipment would be an issue. They've since encaged all of their switchgear with an approved grounded metal fence.

Obviously YMMV depending on your use-case (not too many things can tolerate a 1.5 day downtime), but yes, in general people don't look at lower-level infrastructure enough.

Even at the lower end of the scale, I'd wager 90% of webapps can run fine on a Linux VPS (DigitalOcean, etc.) and the owner can save quite a bit of money compared to a full-stack deployment on a public cloud, with users never noticing a difference in uptime or performance. On the higher end, at places with $20k/mo or above in cloud spend should at least do a cost/benefit analysis on running their baseline compute on dedicated servers and whatnot, and branching out to the cloud for bursts.

On the other hand we have had a bunch of demo prototyping APIs built that ultimately weren't adopted as the client needs changed, and because they were built with Lambdas, it essentially cost nothing to run them. While having servers run for a couple of years before they admitted that priorities had changed and they weren't going to use it but could we built them something else would have had a cost attached to it.

Cloud was cheap in this scenario.

Planning, determining your usecase, workload, and you can usually get a good idea for cost comparison. Especially if you have PBs of data storage costs, data centres or even having local hosting can slash costs.

People are usually pretty bad with risk/cost management.

Weak. If you're B2B (and your business partners are in The Cloudz), or you're B2C and your customers have the opportunity to go to E2E managed telecomms (mobe, and they live in a coverage area) and have utilized it then they've got you covered.

Do people actually give a shit what Gartner says?

I think MBAs do

Way too much.

Reply on news.ycombinator.com