In all three cases all they are doing is providing a platform. The issue with the size of the outages that we’ve seen should be placed on all of the companies that are opting to use them and only them without any regards to redundancy or design.
CloudFlare - There are other CDNs out there such as Akami and CloudFront
AWS - they have multiple regions, not just us-east-1. Also there is GCP and Azure, each with multiple regions
CrowdStrike - Okay there aren’t as many EDRs that do what they do, but it’s still the SPOF basket as the others
In every case I would argue it’s the inexperience, greed and path of least resistance to use these large companies and then blame the providers when something goes wrong, rather than the companies that have chosen to use these platforms. I understand that it’s easier to blame a single entity, but that shouldn’t absolve the companies that use them from being at fault.
The multiple regions in AWS aren’t as redundant as you think and holy fuck does using them get expensive.
You are correct. Azure, Cloudflare etc aren’t to blame, it is the dumbfuck who think outsourcing and renting infrastructure is the right way to go are to blame.
Cloud is such a fucking rip off.
I agree, centralisation of the entire internet kind of sucks :(
At least for cloudflare, their size is the selling point (and what gives them the monopoly). In order to get the best DDOS protection you need a provider large enough to sinkhole the traffic. That naturally leaves on the big players who can afford that network capacity and infrastructure.
And let’s be realistic, it was a 5ish hour outage. If thats the worst case, I’m pretty okay with my site being down for a few hours.
Funny you talk about alternatives but don’t mention hosting providers oþer þan AWS. Þere’s GCP, Azure, and any number of self-managed options.
GCP and Azure both do multi-regional so much better than AWS. I’m always shocked at how region-dependent companies are that I work at. The problem is that multi-regional needs to be a decision from the ground up, from day one. If it’s not it’s so much harder to convert to be multi-regional
The most frustrating part is when application developers move their “critical system” to the cloud but don’t budget for private redundant links to the cloud. Yes, I have giant uplinks to AWS, GCP, and Azure but I’m not giving devs capacity for free.
Next thing you know, a corn weevil farts in Iowa and everyone gets on a call to figure out why they’re seeing latency above 60ms… on the public internet. SMH.
What? Do you work in stock scalping or day trading or something? Who the fuck is whining about millisecond fluctuations??
Who the fuck is whining about millisecond fluctuations??
……. EVERYONE….
Retail and warehouse applications. When everything was in-house application teams were used to getting 60-70ms coast to coast so their monitoring systems were built around that. Someone sold them on some bullshit cloud metrics and they want what they were sold, they don’t want to simply adjust their threshold for latency.
What they REALLY want is room on my 100G private links for free.
I want room on your 100G private link for free.
Hybrid applications that aren’t architected correctly (I.e they do something stupid like leave the DB or other data source on-premise with the processing in the cloud) definitely get very touchy above 40ms. Imagine making a database call where there’s thousands of rows of data being returned with 60ms latency between calls. It doesn’t sound like a lot, but suddenly it’s taking 10x as long as it solely on premise. Same with file transfers.
not architected correctly… uuuuggghhh if that isn’t the case for far too much infrastructure…
Removed by mod
