4 ms·

Didn't people also say the same thing about AWS a while back when that had a downtime? I guess the internet has multiple "Single-Point-Of-Failure"s.

benbristow 6y ago
Ironic since the internet (ARPANET) was specifically designed to not have a single point of failure
- KirinDave 6y ago
  Doubly ironic since in doing this, they created a system where a protocol is the SPoF; namely nonsensical or false BGP advertisements can quickly kill the internet as a whole if done correctly.
KirinDave 6y ago
That's not actually a contradiction. When an individual website is considered as a system, it will have multiple points where the failure of a component inhibits the system. It's possible to have both cloudflare and your database as "SPoFs" and that "single" is not meant to imply everyone only gets one. It's absolutely true that if us-east-1 in AWS has a bad day, a significant fraction of the American digital economy will shut down. For some companies, the same is true of Azure and Google's various comparable offerings. I read your post as skeptical. Why would you be skeptical? If you care about keeping your product up, you absolutely should have a fallback for cloudflare if you're a customer of theirs. Now, you might not care (and actually, for most folks I submit you need not care), but the folks making sure Ambulances get timely push notifications and realtime driving instructions probably care quite a bit.
slazaro 6y ago
A "single point of failure" doesn't mean there's only one of them. It means the whole fails when the SPOF fails. But you can have many of them. A steel chain made of links has as many SPOF as links.
whinybastard 6y ago
it's a series of single points of failure on different levels, so, not multiple points of faiure, but much worse, single point after single point, which means the house of cards fails as often as any of them. The internet of 2020 is a monolith
riknos314 6y ago
This is why AWS is puts effort into blast radius reduction. The Physalia paper is a wonderful read that explains both the justification and high-level approach to implementing said isolation [1]. Another great example is shuffle sharding [2], notably used in Route53. The end goal of this is that aws outages would be isolated to subsets of aws customers where possible. So instead of half the internet failing it's only 10% (obviously the exact numbers depend heavily on implementation). [1] https://assets.amazon.science/c4/11/de2606884b63bf4d95190a3c2390/millions-of-tiny-databases.pdf https://assets.amazon.science/c4/11/de2606884b63bf4d95190a3c... [2] https://aws.amazon.com/builders-library/workload-isolation-using-shuffle-sharding/ https://aws.amazon.com/builders-library/workload-isolation-u...

benbristow 6y ago

Ironic since the internet (ARPANET) was specifically designed to not have a single point of failure

KirinDave 6y ago
Doubly ironic since in doing this, they created a system where a protocol is the SPoF; namely nonsensical or false BGP advertisements can quickly kill the internet as a whole if done correctly.

KirinDave 6y ago

That's not actually a contradiction. When an individual website is considered as a system, it will have multiple points where the failure of a component inhibits the system. It's possible to have both cloudflare and your database as "SPoFs" and that "single" is not meant to imply everyone only gets one. It's absolutely true that if us-east-1 in AWS has a bad day, a significant fraction of the American digital economy will shut down. For some companies, the same is true of Azure and Google's various comparable offerings. I read your post as skeptical. Why would you be skeptical? If you care about keeping your product up, you absolutely should have a fallback for cloudflare if you're a customer of theirs. Now, you might not care (and actually, for most folks I submit you need not care), but the folks making sure Ambulances get timely push notifications and realtime driving instructions probably care quite a bit.

slazaro 6y ago

A "single point of failure" doesn't mean there's only one of them. It means the whole fails when the SPOF fails. But you can have many of them. A steel chain made of links has as many SPOF as links.

whinybastard 6y ago

it's a series of single points of failure on different levels, so, not multiple points of faiure, but much worse, single point after single point, which means the house of cards fails as often as any of them. The internet of 2020 is a monolith

riknos314 6y ago

This is why AWS is puts effort into blast radius reduction. The Physalia paper is a wonderful read that explains both the justification and high-level approach to implementing said isolation [1]. Another great example is shuffle sharding [2], notably used in Route53. The end goal of this is that aws outages would be isolated to subsets of aws customers where possible. So instead of half the internet failing it's only 10% (obviously the exact numbers depend heavily on implementation). [1] https://assets.amazon.science/c4/11/de2606884b63bf4d95190a3c2390/millions-of-tiny-databases.pdf https://assets.amazon.science/c4/11/de2606884b63bf4d95190a3c... [2] https://aws.amazon.com/builders-library/workload-isolation-using-shuffle-sharding/ https://aws.amazon.com/builders-library/workload-isolation-u...