Stop using hyperscalers. Then when an outage does occur, it doesn’t take down half the internet, and instead only affects a much smaller subset of services.
Problem is that as a provider, if you are sure you are confident you’ll get hit by an outage at some point anyway, it’s actually better for you if a bunch of other big names are brought down at the same time.
Instead of “that one service sucked”, the story is “aws sucked”. If it happens too much people will more widely say “ok they suck for using aws”, but for now the transparency gets them treated more like being affected by an unavoidable external condition.
I’m grateful a lot of sites I like didn’t use aws, but I’m not exactly a common demographic and even I won’t know if she is the services even move or not until another such outage.
Like sure, it wouldn’t be all together like this, but that’s also not a reasonable ask for a lot of big cloud customers without huge investments for not actually anything extra reliability.
I read your comment. You basically repeated back what I said.
As for “not actually anything extra reliability”, that’s not true. This is literally the definition of all your eggs in one basket. If all these services were instead spread out amongst smaller providers, there wouldn’t have even been any news about it because it would have affected just a few services. But instead half the internet went down.
Even one of the applications I manage was down because of a single RTE npm dependency used on the forms. This is when we discovered that the npm module wasn’t bundling the whole thing but in fact dynamically pulling the js from a CDN hosted on AWS, because our prod instances kept erroring out for everyone (No, I did not write this application and I’m already replacing the dependency).
The argument isn’t about spending thousands for a lateral shift in reliability, the argument is to decouple everything from a single failure point.
Can you name a more reliable alternative? With citations?
Because every major cloud provider has outages. On prem clouds also have outages. Everyone does.
Stop using hyperscalers. Then when an outage does occur, it doesn’t take down half the internet, and instead only affects a much smaller subset of services.
Problem is that as a provider, if you are sure you are confident you’ll get hit by an outage at some point anyway, it’s actually better for you if a bunch of other big names are brought down at the same time.
Instead of “that one service sucked”, the story is “aws sucked”. If it happens too much people will more widely say “ok they suck for using aws”, but for now the transparency gets them treated more like being affected by an unavoidable external condition.
I’m grateful a lot of sites I like didn’t use aws, but I’m not exactly a common demographic and even I won’t know if she is the services even move or not until another such outage.
Okay, you know those have outages too right?
Like sure, it wouldn’t be all together like this, but that’s also not a reasonable ask for a lot of big cloud customers without huge investments for not actually anything extra reliability.
Did you read my entire comment? I know it’s more than one sentence, but your entire comment would be irrelevant if you read the whole thing.
Did you miss my point?
Why would a company move away from AWS?
Because everyone has outages…
Why should companies invest tens of thousands of dollars to move?
I read your comment. You basically repeated back what I said.
As for “not actually anything extra reliability”, that’s not true. This is literally the definition of all your eggs in one basket. If all these services were instead spread out amongst smaller providers, there wouldn’t have even been any news about it because it would have affected just a few services. But instead half the internet went down.
Even one of the applications I manage was down because of a single RTE npm dependency used on the forms. This is when we discovered that the npm module wasn’t bundling the whole thing but in fact dynamically pulling the js from a CDN hosted on AWS, because our prod instances kept erroring out for everyone (No, I did not write this application and I’m already replacing the dependency).
The argument isn’t about spending thousands for a lateral shift in reliability, the argument is to decouple everything from a single failure point.
Yes, the argument is about spending money to migrate, because how else do we decouple the single points of failure, if not to migrate away?
Remember, this is a capatistic hellscape. Everything is about money.