Hacker News Reader

Comment by soccerdave | Hacker News Reader

I've had ~25-30 instances running for the past 3 years and only had 1 or 2 instances have hardware issues, never had issues with EBS. Running on us-west-2 but it seems like more issues happen in us-east-1.

View on HN

I'm on Virginia zone D... the other day a 15TB EBS went down, even with status as good. Their explanation, which took a lot of time/energy to get, was that the 2nd replicated copy had a failure, and when rebuilding from the 1st good replicated copy (primary) it suffered an unknown error taking down that copy as well... I was upset to say the least.

1 or 2 failures out of 30 is a really high failure rate for HW.

I challenge you to build enterprise grade hardware, run it hard and have a hardware failure rate of ~1% a year.

Challenge accepted. Been building carrier grade equipment with significantly lower failure rates than that for >15 years. Gear I designed in my first year out of grad school is still in field use today.

New challenge: build your machines for low cost from commodity hardware, rent your machines out to millions of customers and never have a single customer have > 1% of the hardware they land on fail.

That's moving the goalpost too far for my taste. But assuming this is true and AWS/GCE are doing this then why are their prices so high?

We run bare metal dedicated servers at SoftLayer with up times in the 3-4 year range. Our failure rate is less than 1%.

Not that hard...

Agreed. Similar results here over the years. I shouldn't bash EC2 so hard, but they also shouldn't keep the same uptime estimates when they have degraded over time. My developer made it all pretty clear to me with a statement "if you tell me the strengths and weaknesses of the system, I'll code accordingly" Great developer, can overcome EBS failing with 3-9's, but only if they state that and not 5-9's!

I just find it weird that every time "the cloud" comes up on HN, people defend it as hard as they can, like running servers yourself is some voodoo magic to be shunned. Usually with examples of "well, X is only saving $56,000/month with this switch away from the cloud! surely they're making a terrible tradeoff in an increase in employees!".

The answer is no. People do these calculations before moving stacks. The cloud is where VC money goes to pad Amazon's bottom line. AWS is insanely overpriced if you actually sit down and do the numbers. I'm our company's part time sysadmin on a bunch of bare metal servers, I spend maybe 1-2 hours total per month kicking things/filing hardware replacement tickets/etc.

I don't understand this mindset against learning the entire stack. You should understand hardware, network and OS. Maybe I'm too old.

Reply on news.ycombinator.com