GHD and others also sell on ebay and may offer free shipping or you can use re-shipper, or find similar recycler company in EU – I’m sure there are plenty of datacenters. I doubt it’s US specific. But nobody argues that HDD prices are inflated.
That classification is meaningless. “WD”, “Seagate” and “Hitachi” are not drive designs. Each brand has produced many unrelated families, platforms, capacities and generations with very different failure rates. Backblaze publishes failure rates by model for exactly this reason.
Nothing you linked establishes that.
The quoted 65% is cumulative survival of a heterogeneous population from new through six years. It is not the probability that a six-year-old drive you buy today will survive another year, and it certainly does not make a used Ultrastar a coin toss.
This is basic reliability engineering. The usual model is the bathtub curve: elevated early failure rate from defective units, a long useful-life region with comparatively low failure rate, then increasing failure rate from wear-out. A drive that has already survived years of operation has already passed through the early-failure part of that distribution. Wikipedia also explicitly notes that you cannot even know where a product sits on that curve without sufficient population data for that product.
So the relevant quantity for a used drive is the conditional failure rate from its present age onward, for that particular model and population. Your 65% cumulative survival number does not provide that.
“Some drives” is doing heavy lifting here. Particular models can develop high AFR as they age. There is no universal progression where HDD failure probability simply climbs 10%, 12%, 14% with age. Backblaze’s own data shows substantial model dependence, and even their current commentary treats age together with model and population size.
The 65% figure is especially bad as a universal HDD-lifetime statistic. Backblaze published another article less than two months earlier saying that, using their then-current data, 88% of drives were expected to survive six years, attributing the improvement to newer drives, servers and operating practices.
And anecdotes about one WD Red dying after three years and another arriving DOA establish nothing about the reliability of WD as a brand, much less about the reliability of some unrelated Ultrastar model.
So yes, disks fail. Old disks eventually enter wear-out. None of that supports “used is a coin toss”, “WD is meh”, or any universal age-based failure sequence. You need model-specific data.
Irrelevant anecdote
If we are talking anecdotes – in my personal experience over last decades, when I stopped overpaying for a privilege to experience early failures and switched to used drives exclusively – amount of postage I buy to RMA drives reduced drastically. IN fact, none of the used drives I bought in the last 6 years (out of about 20 or so) failed so far. And yet, I did RMA plenty of new drives in the past. But this too, is an anecdote.
No, the reasons for the that rule are still present.
Are you in US? Maybe that matters? US1 satellite and what not? 90% of all data is from Us1 on oll my nodes.
No, I moved it to the other topic the message I responded to belonged, 1 minute after I posted it here :). Nothing was deleted.
You would need to provide some quotes here. “Use what you have” is what storj repeatedly asserted
Yeah, that is immediately a no-go, let alone if you are considering buying new drives is a very bad idea – see bathtub reference above. GoHardDrive and ServerPartsDeal both have ebay stores and yes, you’ll probably pay shipping to outside of US, but various reshipping services exists, and I suspect EU might have similar datacenter recycler companies. I’m not in EU so I can’t be arsed to search.
Ehhh… It’s not an investment. It’s an expense. Buying drives periodically is part of running a server. With or without storj. Storj is not a factor here.
If you want storj to pay for your drive – this is not happening, by design. At some point it was possible accidentally, when drives were cheap, but that was not the intention.
RIght. You buy 24TB drive for yourself. and unless you need all 20TB of space at once, you give leftover to storj.
Storj cannot shrink nodes, so advice was to run many, they will all share ingress, and then you rm-rf one by one when you need to get space back.
I like to keep 40% free on my arrays, not to scramble at the last moment and to never run out of space. So when drives are cheap— I buy them. when they are expensive – I have a huge cushion of free available space.