Hacker News Reader

Comment by belter | Hacker News Reader

> So we need to hold regulatory bodies accountable as well - when they frame regulation such that organisations are cornered into this they get to be part of the culpability here too.

No, we need to hold Architects accountable, and this is the core of the issue. Creating systems with single, outsourced responsibility, in the critical path.

View on HN

As a CTO, when your company goes down you get fired.

When every company goes down you get let off.

The sensible thing is follow the herd and centralise. You're outsourcing the risk to your own job.

This is the point of much of the security efforts we see now.

Outsourcing of security functions, and things like login push a lot of liability and legal issues off into someone else's house.

It's hard to be the source of a password leak, or be compromised when you don't control the passwords. But like any chain your only as secure as your weakest link... Snowflake is a great current example of this. Mean while the USPS just told us "oops" we had tracking pixels for a bunch of vendors all over our delivery preview tool.

Candidly, most people stacks look a lot less like software and more like a toolbar riddled IE5 install circa 2000. I don't think our industry is in a good place.

> When every company goes down you get let off.

And when only companies who use a certain OS and/or Cloud vendor go down? ;-) Do you also get let off?

If it's large enough. Nobody loses their job when office 365 goes down, even if that happens once a year.

However if you decided to choose a small company which goes down once every 5 years, you're screwed.

This is one of the interesting aspects in Ethereum.

If your validator is down, you lose a small amount of stake, but if a large percentage of the total set of validators are down, you all start being heavily penalized.

This incentives people running validators to not use the most popular Ethereum client, to avoid using a single compute provider, and to overall, avoid relying on the popular choice since doing so can cause them to lose the majority of their stake.

There hasn't been a major Ethereum consensus outage, but when that happens, the impact of being lazy and following the heard will be huge.

How is it lazy and herd-like to _not_ run the latest and greatest? Sounds like Etherium's design is promoting a robustly diverse ecosystem rather than a monoculture.

> How is it lazy and herd-like to _not_ run the latest and greatest?

I'm not sure what you're asking here. Ethereum incentives don't make you run the latest version of your client's software (unless there's a hardfork you need to support). You can run any version that follows the network consensus rules.

The incentives are there to punish people who use the most common software. For example, let's say there are around 5 consensus clients which are each developed by independent teams. If everyone ran the same client, a bug could take down the entire network. If each of those 5 clients were used to run 20% of the network, then a bug in any one of them wouldn't be a problem for Ethereum users and the network would keep running.

If the network is evenly split across those 5 clients but all of them are running in AWS, then that still leaves AWS as a sigle point of failure.

The incentives baked into the consensus protocol exist to push people towards using a validator client that isn't used by the majority of other validators. That same logic applies to other things like physical host locations, 3rd party hosting providers, network providers, operating systems, etc... You never want to use the same dependencies as the majority of other validators. If you do and a wide-spread issue happens, you're setting yourself up to lose a lot of money.

If 90% of the world runs in AWS, and I'm the only one running on my own hardware, do I get a benefit when AWS goes down?

It sounds like you're describing the advantages of diversity, with a little game theory thrown in to sweeten the deal. Still not sure how that can be described as lazy, or did I completely mis-read the original phrasing?

I find that in today's world it is no longer about one person being "accountable". There is always an interplay of factors, like others have pointed out cyber security has a compliance angle. Other times it is a cost factor, redundancy costs money. Then there is the whole revolving door of employees coming and going, so institutional knowledge about why a decision was made lost with them.

That is hard to do for even a small company. How do you balance all that out for critical infrastructure at a much larger scale?

The problem is that even knowing that this likely to happen many companies would still put CrowdStrike into a critical system for the sake of security compliance / audit. And it's not even prioritization of security over reliability because incentives are to care more about check-boxes in the audit report than about the actual security. Looks like almost no party in this tragic incident had a strong incentive to prevent it so it's likely to happen again.

Reply on news.ycombinator.com