Hacker News Reader

Comment by jchw | Hacker News Reader

> Today it is possible to purchase a refurbished 1 TB hard drive for about $20 — so buyers are essentially paying Amazon each month the entire cost of the hard drives used to store their data!

What a refurbished hard disk does not give you:

- Redundant storage

- Redundant power/network/etc.

- Geographic distribution

- A network API

- Any networking, or uplinks at all

- Access control

- A team ensuring the reliability and security of your data

- Global network of datacenters and pops

- Integration with other cloud services

All that and Amazon et al. do not store your 1TB of data in a single refurbished hard disk. This is beyond non-sense.

Sure, some of these costs are amortized, but I still think this is a lame comparison. If all you wanted was at-cost block storage with crappy reliability and speed, you could probably accomplish that in easier ways.

View on HN

> What a refurbished hard disk does not give you: [...]

The whole point is that a decentralized storage network can add such features on top of the 1TB refurbished hard drive.

The refurbished drive may not be reliable, but if you replicate your data amongst many separate drives with independent failure (geographically distributed; different providing entities) then you can create a system with greater reliability. By using erasure coding schemes (similar to RAID5/6), it's possible to reduce the storage overhead which supports reliability to a fraction of the principle amount. Thus, providing 1 TB of storage to the network could earn slightly less than 1 TB of reliable storage in return.

Comparatively, a single cloud provider has the disadvantage that, because they are a single legal entity, they can decide for whatever reason to not continue your service.

What I am not saying is that Sia is not viable or that cloud storage providers have no downsides, which they clearly do. I did not mean to imply that there is anything wrong with this service, just to express my annoyance with the comparison. I am not happy that my comment got so much attention, because I had no intention to overshadow the announcement with my single gripe.

The line that is annoying me and probably others is this:

> Buyers are essentially paying Amazon each month the entire cost of the hard drives used to store their data

...They're not. Because redundancy. Because electricity, CPUs, network, software development, etc.

The comparison is meaningful setup for discussing decentralized cloud storage.

The comparison establishes the fact hat Amazon's price is not coming from the physical storage HW itself, its the everything around it. Then you can move to discussing how to get other parts cheaper.

I think the use-case of cloud storage is mostly marketing by cloud provides to show relevance to the consumer. They are are loss leaders at best; used to lock you in to their services / track your every move; at worst.

I think the comparison with a $20 1TB hdd is perfect. Cloud providers, and phone makers, have convinced you to have your photos, music, data, online, always. Like a bank. Among other things, its much easier to mine your usage patterns that way and enforce DRM of other cloud services.

I have my IDE PC drive from 20 years ago. It still works. I can burn a CD if im that worried about failure. I dont need the redundancy of a billion user website, a networking api, access control, all that other crap you use in enterprises that comply with ISO standards.

Before cloud providers, people didnt take their photos to Apple and tell them to store them. The only reason u do it now, is because of marketing and service integration.

Wrong. The reason we do now is so that we can have our photo library accessible on our phones and synced between multiple devices.

Appreciate your response. My comparison to the cost of a hard drive today is not meant to be a direct comparison, but instead to demonstrate that traditional cloud storage is actually quite expensive. Sia is able to bring down storage costs by an order of magnitude and bandwidth costs by two orders of magnitude. While most HN readers understand the nuances of cloud storage costs, most non-technical readers are unfamiliar with the high costs of S3 storage and egress.

The issue I take is that the comparison is truly not meaningful. Yes, storage is cheap, but you are comparing the cost of a refurbished disk to a managed service.

This may lead people to think cloud storage is a waste of money, which I personally think is untrue (though Sia certainly may be able to do it for cheaper than cloud storage, too. Both things can be true.) While the cost to us is certainly more than the cost to the provider, the cost for us to replicate what a cloud provider offers is certainly greater at small scales where we benefit less from amortization and volume pricing.

A good point may be that we don’t actually need to replicate what a cloud provider does to offer a compelling alternative for various use cases. This I would probably agree with. It is certainly cheaper for me to have my own NAS and offsite backups than it would be to pay for cloud storage, though it is still surprisingly expensive (at least to the uninitialized. Check out what a fully loaded Synology 8 bay would cost! Multiply by number of backup sites. A lot of storage, but definitely not cheap.)

The comparison wasn't complete, but I think it was still meaningful. It's like comparing the cost of sand to glass; it should be obvious that sand is not glass, but it begs the question: how much proprietary value is there in glassmaking?

> This may lead people to think cloud storage is a waste of money, [...]

But that is exactly the argument being made -- that a trustless distributed system can substitute and obsolete cloud storage.

> But that is exactly the argument being made

Then I respectfully disagree. I don't think cloud storage is always the best use case, but I do believe it is a good value for what it actually offers you. Backblaze B2 is my ideal marker for what a 'good value' in cloud storage should look like. At $5/mo per TB[1] it easily crushes the metric of monthly cost being less than equivalent non-redundant refurbished storage media.

> that a trustless distributed system can substitute and obsolete cloud storage.

OK, I have literally no qualifications to say that this isn't true. After all, what I am NOT claiming is that it can't be done for cheaper. What I AM claiming, is that without a service like Sia, it is certainly not going to be cheaper for common use cases of end users. I'm talking about comparing harddrives to services.

> Buyers are essentially paying Amazon each month the entire cost of the hard drives used to store their data

I still think though, that a compelling case could be made without bad comparisons or hyperbole. Which makes this more frustrating than it needs to be imo.

[1]: https://www.backblaze.com/b2/cloud-storage-pricing.html - Assuming I am not deeply mistaken.

Well, the actual comparison is between the cost (for ownership) of a 1 TB (refurbished) drive vs. 1 month of 1 TB of cloud storage.

The Author could have done it as the cost (for ownership) of a brand new NAS with 2x2 TB disk (which can be - just checked on tiger direct) around 350+65+65=480 US$ vs 6 months of 4 TB cloud storage 6x4x20=480, but the overall message would have been the same.

Imagine that you rent a car for 6 months and during the 6 months you pay rates equating the price of the car brand new.

Obviously you're paying for more than the drive. That doesn't make it a bad comparison. There's no great reason for the drive to be more than half the cost, so even with 3 copies amazon still comes out as making a whole lot of profit. There's a reason that multiple competitors can provide all those things at significantly lower prices.

And more importantly, if you're focused on storage then don't pay for features you don't need.

One of the main innovations of Sia is providing a low cost means to turn that 1 TB refurbished hard drive into a useful cloud product.

Sia is able to provide, at 1/10th the cost of Amazon, high throughput, high reliability, high redundancy, geographically distributed cloud storage, and it can do that without high reliability drives or datacenters.

Off topic, but: I purchased three refurb 6TB HGST He HDDs in Jan. from a reputable shop. All three failed last month. I was so happy that the shop completely refunded all of them because the warranty is generally very limited on refurb HDDs. Normal RMA is 5yr on this specific model.

There's no way refurb drives are cost effective for these environments.

On balance, I purchased 45 refurb 4TB SAS drives around the same time. I had two DOA with 0% failure rate since then.

Also, very anecdotal, I know.

I'm wondering if the fact mine were He filled refurbs could've been a factor.

Beyond a functional test, checking the SMART data for an indication of overall drive state and He containment, and then giving the drive a quick clean, it would be interesting to know the supplier's scope for 'refurbishing' an He drive.

That all three drives lasted this long then failed within a month is very suspect. This sounds less like random failure indicative of the quality of refurbished drives in general, and more like a defect common to all three drives.

Because of such plagues, data-centre operators often mix drives of different manufacturers, models, batches and ages. Running heterogeneous devices in a redundant array helps makes the failure distribution more random so that the safety margin is less likely to be suddenly overwhelmed by correlated failures.

Makes sense to mix drives. I do this with my personal NAS (~24 drives) and has worked out nicely even at such a small scale.

Backblaze's failure/longevity stats are indispensable when considering which drive to purchase: https://www.backblaze.com/blog/hard-drive-stats-for-2018/

If you went by these stats, you'd buy HGST...

Not saying you shouldn't, I'd probably do so as well. Curiously, there's no 6GB HGSTs over at Backblaze.

Reply on news.ycombinator.com