Backblaze Scales Storage Cloud(backblaze.com)
backblaze.com
Backblaze Scales Storage Cloud
https://www.backblaze.com/blog/how-backblaze-scales-our-storage-cloud/
43 comments
Generally really appreciate how much information Backblaze publishes. I also liked when OKCupid did similar before they were bought out.
Yev from Backblaze here -> I LOVED the OKCupid data dumps, they were so fascinating! Pornhub used to do a great job of surfacing their data as well, but obviously the subject matter was much different.
It's amazing to see the magnitude of declining storage prices laid out so clearly.
Their prices do seem to continue to increase though
https://www.backblaze.com/blog/2023-product-announcement/
https://www.backblaze.com/blog/2023-product-announcement/
Physical drives is just one part of the service. Everything else is subject to inflation. Usual solution is to automate and cut staffing, but we know that often doesn't work out well.
I'm sure they like money as much as the rest of us do.
Profit is like oxygen, not the point of life but required for survival.
Too much oxygen is as bad as too little, if not worse.
Pure oxygen will kill you at >2 atmospheres of pressure. (That's roughly 30 feet underwater.)
But unlike oxygen, having more money allows you to do more stuff. Once your oxygen needs are satisfied, any more is basically wasted. You could make the argument that money has diminishing utility as well, but it's totally incomparable to the utility curve for oxygen.
Backblaze is not profitable
What type of "not profitable"? By itself that's not a useful statement.
Do they have a bunch of cash in the bank?
Do they have higher revenue than Cost of Business?
Do they have free cash flow? Is it growing?
A business that isn't profitable and also isn't growing is a much different beast than a company that is investing its cash reserves to grow the business.
Do they have a bunch of cash in the bank?
Do they have higher revenue than Cost of Business?
Do they have free cash flow? Is it growing?
A business that isn't profitable and also isn't growing is a much different beast than a company that is investing its cash reserves to grow the business.
They're definitely growing, and will likely be profitable at some point in the next few quarters: https://ir.backblaze.com/news-releases/news-release-details/...
But even on a non-GAAP basis (which excludes stock-based compensation), their net income for the most recent quarter was a loss of $5.5 million.
But even on a non-GAAP basis (which excludes stock-based compensation), their net income for the most recent quarter was a loss of $5.5 million.
That just looks like good accounting.
> What type of "not profitable"? By itself that's not a useful statement.
They're a public company so you can check for yourself: https://www.sec.gov/ix?doc=/Archives/edgar/data/1462056/0001...
They're a public company so you can check for yourself: https://www.sec.gov/ix?doc=/Archives/edgar/data/1462056/0001...
Nobody said labor or land prices have gone down.
They are able to fit 4x as many terabytes into a single server compared to when they launched B2, so it's as if their labor and land prices have dropped significantly.
It's fair to criticize the per-TB price increase.
Apparently they've been losing money, so from that perspective it makes sense to increase prices, but it doesn't seem like it's based on their costs going up.
(The unlimited home backup price going up is a separate matter and has much more cost-based justification.)
It's fair to criticize the per-TB price increase.
Apparently they've been losing money, so from that perspective it makes sense to increase prices, but it doesn't seem like it's based on their costs going up.
(The unlimited home backup price going up is a separate matter and has much more cost-based justification.)
Honestly, what I'd be interested in is buying basically permanent space.
And how much to just pay for 1 TB up front and I never have to pay for it again?
With steady decline in storage costs, that should be a sustainable business model. It gives you upfront capital.
And how much to just pay for 1 TB up front and I never have to pay for it again?
With steady decline in storage costs, that should be a sustainable business model. It gives you upfront capital.
pCloud will sell you "lifetime" cloud storage - current price for 2TB is 400eur, but they do black Friday/such sales too.
They operate for many years on 30% net loss. How company can last for so long without going bankrupt?..
Ask Uber. E.g. https://www.statista.com/chart/amp/31223/net-income-loss-of-...
Somehow people may finance an utterly losing operation for a decade or more.
I wish Backbkaze stays around though, even if they charge more. They'd still be charging less than AWS, and they seem to have a good culture / attitude.
Somehow people may finance an utterly losing operation for a decade or more.
I wish Backbkaze stays around though, even if they charge more. They'd still be charging less than AWS, and they seem to have a good culture / attitude.
I am reading Uber report: https://d18rn0p25nwr6d.cloudfront.net/CIK-0001543151/bbb206f... and my understanding is that they are profitable in timeframe since IPO(, 2020+, meaning revenue is about or larger than R&D, Sales and infra), but they have some large "other expenses" entry, I speculate it could be some acquisition.
They mention transferring data over “encrypted dark fiber connection”.
What does that actually mean? I thought fiber was dark until it was plugged in.
What does that actually mean? I thought fiber was dark until it was plugged in.
It means they are running a private network over a dedicated fiber line that carries only their traffic.
Okay, that’s a different meaning than this: https://enterprise.spectrum.com/support/faq/internet/what-is...
I can't access the link (403), but I've consistently seen it used to describe the state in which you rent the fiber. As in, instead of renting the service of shuttling data (where the provider puts the light on the fiber), you rent the entire fiber in its dark state and put your own light into it (which means you get to choose what optics you use).
Or, to use an analogy: You rent an unfurnished apartment. You're going to put furniture inside, of course, but you're still renting an unfurnished apartment.
Or, to use an analogy: You rent an unfurnished apartment. You're going to put furniture inside, of course, but you're still renting an unfurnished apartment.
Over here dark fiber has meant private, dedicated fiber since the '90s. As opposed to connections managed by your telco. Seems similar in the US per https://en.wikipedia.org/wiki/Dark_fibre
Yeah, they misused the term. They bought dark fiber, after you've bought dark fiber it's obviously no longer dark.
No, they didn't misuse it. You can buy lit fiber too. The difference with dark is that you get a lot more control in terms of what transmission tech you use, link-layer capabilities, etc
Your point does not address the thing the person you're replying to said. After you buy (really, lease) some dark fiber, and run traffic through it yourself, now that fiber can no longer be sold to anyone else as dark, because it's been lit up, by you. Dark fiber is only dark fiber for its first subscriber.
Whether that first subscriber themselves considers the thing they've bought "dark fiber" or "lit fiber" once they light it up, depends on if they're made an exclusivity agreement with the provider. If they have, then the fiber is still "dark fiber" from their own perspective. If they haven't, then the fiber should be considered "lit fiber", because now that it's been set up for their use, the provider has likely put it on the market for others to lease as well, so it's not going to stay "dark" for long.
But really, these are silly terms. Customers don't care about buying "dark fiber" vs "lit fiber"; they care about buying "dedicated fiber" vs "[dedicated throughput on] shared fiber." (A lease on dedicated fiber allows for nearly-arbitrary throughput scaling over time at no additional OpEx — just the CapEx of upgrading the transceivers on either end once-a-generation or so.)
The only two things that you know by buying "dark fiber" from a peering provider, are:
1. Demand for that path was lower than predicted at the time you requested it, so there existed at least one fiber-infrastructure developer who had run some fiber along the path but not managed to sell it off to any particular peering provider yet. (Peering providers don't tend to buy-and-hold fiber without either immediate or predicted base-load demand; rather, they index available-to-purchase fiber from infra developers, offering those to their customers at snapshotted rates; they then buy the fiber from the developer in response to a customer Purchase Order, eating any loss from purchase-price volatility.)
2. You might have to pay the peering provider you're leasing from, for the initial set-up work of lighting up the fiber: routing a secondary fiber link from the optical switch in the fiber-infra developer's overland or shoreside termination point, over to the peering provider's cable hotel; terminating that fiber into a [maybe fresh] optical switch or transceiver in the cable hotel; etc.
Whether that first subscriber themselves considers the thing they've bought "dark fiber" or "lit fiber" once they light it up, depends on if they're made an exclusivity agreement with the provider. If they have, then the fiber is still "dark fiber" from their own perspective. If they haven't, then the fiber should be considered "lit fiber", because now that it's been set up for their use, the provider has likely put it on the market for others to lease as well, so it's not going to stay "dark" for long.
But really, these are silly terms. Customers don't care about buying "dark fiber" vs "lit fiber"; they care about buying "dedicated fiber" vs "[dedicated throughput on] shared fiber." (A lease on dedicated fiber allows for nearly-arbitrary throughput scaling over time at no additional OpEx — just the CapEx of upgrading the transceivers on either end once-a-generation or so.)
The only two things that you know by buying "dark fiber" from a peering provider, are:
1. Demand for that path was lower than predicted at the time you requested it, so there existed at least one fiber-infrastructure developer who had run some fiber along the path but not managed to sell it off to any particular peering provider yet. (Peering providers don't tend to buy-and-hold fiber without either immediate or predicted base-load demand; rather, they index available-to-purchase fiber from infra developers, offering those to their customers at snapshotted rates; they then buy the fiber from the developer in response to a customer Purchase Order, eating any loss from purchase-price volatility.)
2. You might have to pay the peering provider you're leasing from, for the initial set-up work of lighting up the fiber: routing a secondary fiber link from the optical switch in the fiber-infra developer's overland or shoreside termination point, over to the peering provider's cable hotel; terminating that fiber into a [maybe fresh] optical switch or transceiver in the cable hotel; etc.
I'm pretty sure the point of saying it was dark fiber, meaning there's no one else on the link, was to show how secure and performant their interconnects are. The point of the article is mostly marketing, after all.
FWIW I've heard the term "dark fiber" used in both ways as well. Whenever there's ambiguity in jargon, I just avoid that jargon and use more words to describe the actual concept.
does that mean that they run hardware as long as it lasts?
and in this case made sense to proactively move to "denser" storage?
and in this case made sense to proactively move to "denser" storage?
the server to support the hard drives is quite a large part of the total cost, so replacing drives is probably an economical solution to increasing storage
Was going to say the same... A relatively modest server in the past 5 years can handle a few dozen SATA drives without issue. Going from 8tb drives to 16tb doubles capacity. IIRC they tend to cycle out drives around the 5 year mark.
I'm wondering if they've experimented with higher endurance lifecycles risking larger % of failures relying on redundancy and replication to migrate data, or encountered increased OpEx costs.
another solution is to distribute the task of drive maintenance. see storj for example.
Are the participants in this systems bare metal professionals? Prices seem suspiciously low.