Backblaze Hard Drive Reliability Stats for Q3 2015 | Hacker News Reader

Backblaze Hard Drive Reliability Stats for Q3 2015

backblaze.com

83 pointsdiscreditable19 comments

Meta: There are lots of very good VPS providers around like Vultr, Digital Ocean, Linode, etc., but none are able to compete with EC2 on the storage front. Backblaze should partner with or even create one that leverages its immense storage capability to provide S3 and EBS competition integrated with a cloud offering. That would be awesome, since EC2 actually kind of sucks on the compute and network front (slow, flaky, no IPv6).

Backblaze employee here. I am surprised Amazon S3 still doesn't support IPv6. In our very first invite only beta to Backblaze B2 we don't plan to support IPv6, but we plan to support it soon after. Maybe it is much harder for Amazon because they have so much legacy code and we could plan for it up front?

Yev from Marketing here -> What do you MEAN we're not going to support it?

It was cut for time to market reasons from the first release. Sorry about that.

I thought my department was in charge of release timing? Engineering be damned!

What's my department in charge of?

I don't know...what department are you in?

When you learn something from your company on HN there's something to be said about your internal communication or lack thereof ;-)

I'm not going to throw you the first stone, I have to read the local newspaper if I want to know what's going on at my (very small) company.

Amazon not supporting IPv6 is inexcusable. They're basically holding half the startup world hostage on deploying it.

https://www.digitalocean.com/

https://www.softlayer.com/

https://www.vultr.com/

https://www.linode.com/

https://www.arpnetworks.com/

https://www.ovh.com/

I don't understand why everyone has to use Amazon. Their price/performance for compute is not great and their per-node reliability isn't any better than any of the above providers either. (Google compute has auto-migration on hardware problems so they score higher here, but still no IPv6.)

Amazon does have a great service ecosystem but most people don't use most of it. I'd expect big-and-dumb "enterprise" IT people to pick a provider by picking the one with the largest features list and/or the highest "nobody ever got fired for" score, but I'm surprised that startups do the same. All of the above providers support IPv6 out of the box and have for a while. One of the ways startups can win is by making first-handed technically-informed vendor decisions instead of going with the herd.

Google Compute doesn't even support IPv6 and Google is a big pro-IPv6 company.

I suspect that somewhere under the hood these services are using some expensive enterprise SDN stuff that still doesn't support IPv6.

I'm not sure if you meant just that, but they certainly use in-house developed SDN stuff. That will at such scale still be "expensive" and "enterprise" tough.

Interesting but I don't understand what "222.77%" failure rate mean for ST1500DL003 in 2015 ? Each disk died twice !?

Yev from Backblaze here -> Not necessarily, it just means that if we had a set of those drives and replaced them multiple times over a given period, that number can be quite high. We're comparing the drives that failed against the number we have spinning, not the total we've had in service.

Thanks for explanation. (I would have preferred a definition "failures / number of disk bought", but, in practice, the result is probably very similar for all other models)

Well that confirms my anti-seagate bias is not unjustified.

Fairly obvious somewhere in Seagate's management chain, there are people making purposeful decisions to cut corners with quality and care in manufacturing.

By the way, didn't Google research show that smart data is not a good predictor of drive failure? Or was that just on SSD?

They did, in 2007. Things might have changed by now.

https://static.googleusercontent.com/media/research.google.c...

Their failure rate has dropped quite a bit so far in 2015, so it looks like they've made changes to correct their previous poor failure rate. Still a long way to go to become the front-runner, but they're going in the right direction at least.

I'd be surprised if they just not finally decided to improve since I have the same feeling as ck2 due to anecdotal experience dating back to the year 2000.

Reply on news.ycombinator.com