Hacker News Reader

Comment by nsfyn55 | Hacker News Reader

>Don’t blow anything into the Cloud that you don’t have a personal copy of.

I don't understand this logic. Amazon's S3 offers service level agreements with failure rates that at one point implied the statistical likelihood of losing an object to be once in "thousands of years". When dealing with any sort of stable storage this is simply something I cannot offer. I couldn't produce a set up locally with the resources I have for making guarantees on the decade level let alone millennia.

With that said I keep personal copies, but the authoritative copy is what's in the cloud, because its a hell of a lot more stable.

TL;DR I hear this argument all the time. The cloud isn't perfect but its a hell of a lot closer to anything I could achieve. "not invented here" syndrome won't save your data.

View on HN

Until your control panel gets hacked and you lose all your data. See: codespaces.com (assuming it wasn't an inside job or a dumb mistake, but even then the same rules apply). So no, DON'T BLOW ANYTHING INTO THE CLOUD THAT YOU DON'T HAVE A COPY OF.

It has nothing to do with 'not invented here', it has everything to do with your inability to outsource your responsibilities.

The degree to which people rely on others to take care of their stuff is a huge blind-spot. Jason's advice is spot on in this respect, no matter what the up-time guarantees of the cloud solution you are using (and no matter what the redundancies), if you store all your data in the cloud without an off-line copy your company is 3 mouse clicks away from being history.

The 3-2-1 rule is a rule for a reason.

3 copies

2 formats

1 copy off-site

The cloud is a great place for that 1 off-site copy.

Assuming the rest of your data isn't in the cloud, yes. Otherwise reverse the situation. But you got it perfectly.

No argument at all about the last sentence, but you should have skipped the first four lines as wildly premature optimization.

The vast majority of the world would be much better served by (to appropriate your jargon) a 2-1-1 rule, because it's a straightforward treatment for a straightforward problem that is easily implemented. Yes, "format skew" and double-failure of backup solutions does indeed happen, but at a much lower incidence than "oh crap I deleted it!".

I think the 'three copies' rule refers to cyclic backups, not necessarily three copies of the one piece of data.

This protects against data corruption that is not detected immediately.

Another common way of doing this (and one that I prefer) is one where you rotate out a backup medium with ever larger intervals. So one gets set aside per week, then one gets set aside per month and so on. That gives you a series of snapshots in time that will allow you to pinpoint with some accuracy when an event happened. Longer ago you'll have less accuracy but this can help a lot in trying to triangulate who or what messed up. Just being able to answer the question of whether or not 'x' happened before 'y' was hired or after can help in narrowing down the number of suspects in case of a breach or other nastiness. It also prevents against back-ups for whatever reason not wanting to be reloaded (and you should guard against that by loading your back-up immediately after you make it, even so, the medium might fail the next time you try a read).

Better still if there is a streaming log of everything but only very few companies can afford that sort of solution for all their data. Those can be hard to restore from (by replaying) so there too a snapshot system can help.

Oh sure, there's lots to say about the design and effective use of a backup regime. I agree with all that stuff.

But if you're going to condense it to a "rule" that will help people not well-versed in the field, that rule can only be "MAKE BACKUPS!", because at least 90% of the data loss scenarios in the real world happen because simple backups weren't made.

Don't make it more complicated than it is, because someone will stop to do it "right" and then lose data because they didn't just make a copy on a USB stick.

Modern 3-2-1:

3 backup services 2 formats 1 local copy

I guess that could get expensive depending on the services.

Yes, it could be. The trade-off is usually measured by the cost to the company of re-creating or losing the data entirely. So for some stuff such a system is feasible, for others it isn't and then you'll have to compromise. Log data for instance does not need to be stored like that unless there are legal rules saying that you have to keep it no matter what.

For each business this is a delicate affair and you should spend some time on figuring out what you really absolutely have to have and what you could lose without losing the company or getting in trouble with the law.

What's the relative risk that my specific control panel will get hacked and I, specifically, will lose all my data, vs. the risk that data on my local store will be compromised by a virus on my computer encrypting my whole hard drive and refusing to let go until I send XYZ bitcoins to a specific address?

There's risks and then there's risks. As always, you can spend some time and money to mitigate some risk, and decide some risk is low enough that you don't care to mitigate it. I wouldn't just assume one risk vector is higher than another without some concrete numbers.

There's risks and there's bankruptcy. If you feel like gambling then be my guest. What's the risk of your house catching fire? Are you insured? What are the risks of getting burgled? Your car stolen, a car accident? Are you insured?

What are the risks of having a bad health issue happen to you? Are you insured?

I'll bet that you don't know the answers to any of those questions, and yet, you are probably insured against all of them.

As for your question in more detail: it happens often enough that for me if you operate your business 'in the cloud' and you don't have a back-up of your data outside of the cloud to guard against catastrophic data loss due to either malice or accident that I'll happily fail you. Think 'sysadmin was depressive, wiped out company' (or maybe you let him/her go and they took revenge or any one of a number of other scenarios that would instantly terminate your existence online if you had not guarded against it).

Risk management is consequences * incidence versus cost to mitigate. If the cost to mitigate is negligible, the risk is measureable and the consequences are terminal then it is a no-brainer to protect yourself against this. At least one of my customers wished they had set up a backup facility for their critical data (too bad, end of story for them) and there is one published case that we all know about. That's two that you can count on and most likely other trouble shooters have similar stories.

I see catastrophic cloud data loss as having a much higher chance of happening than many other items on the list of stuff that I verify before greenlighting an investment.

I see what you mean; I read "DON'T BLOW ANYTHING INTO THE CLOUD THAT YOU DON'T HAVE A COPY OF" and thought of only my personal data (because if it's user data, it's originating from the cloud and I'm not blowing it into the cloud; if anything, I'm pulling it out of the cloud).

Though if we unbox that a bit... How do we balance the risk of loss due to your cloud being owned vs. loss due to the risk of your users' PII being violated if someone attacks your in-house (ostensibly in-house reliability and security-managed) copies? Better make sure you're spending the money on security best-practices on all your copies, not just the "live" ones.

Side-note: What are your thoughts on two separate cloud providers as providing sufficient insurance? Say, having your data live in Google Cloud and at rest in Amazon S3?

That was Jason's use case that he had in mind, but for plenty of people the situation is the exact opposite but the principle remains.

Having two cloud providers would work assuming they don't share any critical infra-structure and assuming that it is not the same set of employees that have access to both systems. Also helps if you have a totally separate set of credentials for the back-up system and if that back-up system can only be unlocked by very senior people (preferably execs) after a catastrophe of suitable magnitude hits.

Whether you store your primary in the cloud or your back-up in the cloud the story is the same: have a copy somewhere else.

Finally: the only real back-up is one that is off-line. So from that (ok, ultra-paranoid) viewpoint it would be best if you actually went to an off-line medium to store your data in such a way that nobody can wipe it all out without physically destroying all copies.

All this of course after suitably weighing the importance of the data.

If your data is truly valuable to you, then yes, you host it (encrypted, I hope) in a cloud service to protect you from onsite issues; you host it live locally, to work with it — and you have a hard drive in a safe deposit box somewhere, as well, in case something goes wrong with the cloud copy.

Read this and then consider if you ever want to write something like it. https://www.facebook.com/orainwikihosting/

That's very painful to read. Codespaces went pretty much the same way, here is what they wrote:

"Code Spaces will not be able to operate beyond this point, the cost of resolving this issue to date and the expected cost of refunding customers who have been left without the service they paid for will put Code Spaces in an irreversible position both financially and in terms of ongoing credibility. As such at this point in time we have no alternative but to cease trading and concentrate on supporting our affected customers in exporting any remaining data they have left with us."

They promised a follow up once they got to the bottom of what happened but they never did, I'm still curious what the whole story was.

Once again. I only entertain statistical arguments. This anecdotal nonsense is for the birds. Any given service can fail at any time. This is known kaleesi. You can poop out case after case of some shit wiki platform losing your bird watching posts and it won't make a lick of difference. You need to show me that its better with the magic of statistics. Should be easy peasy show me that you have lost less data(proportionally) than amazon over your entire career.

I can make this even easier. Its almost certainly going to boil down to this question. Have you ever lost even a single byte of data? Cause if you have, you aren't even in the same ballpark.

"Have you ever..." is the wrong question when it comes to risk management; the right questions are "Is is possible...", "How likely...", "How can you avoid..." and "What's the cost to avoid...".

Have you ever been killed in a car accident? No? Then I guess you don't need a seatbelt, right? Has your house been burnt down by a lightning strike? No? Then no lightning rods and proper grounding for you, right?

You don't wait until a disaster occurs before you put a disaster recovery plan in place.

This is a misrepresentation. The "Have you ever" was to demonstrate a point. There are armies of IT housemen that believe they are better than securing data than the major storage providers. This attitude is pretty pervasive and patently nonsense. My point was to get them thinking. If you in your entire career have lost even a single byte of data. You are statistically missing the mark(in terms of data loss) by orders of magnitude. S3 is stupidly resilient. So much so that no amount of raid configs, backups, and redundancy on the part of the joker in the corner cubicle will ever even be even close.

So getting back to risk management. So we about to make a QB selection for the big game. Now that we understand that the implementation of folks like jaque here are like the 12 year old pee wee standout vs. the cloud provider's seasoned NFL quarterback. Which one do you pick to safe guard your business.

> Until your control panel gets hacked and you lose all your data

This is just alarmism. If you really wanted to demonstrate your point you would show me some data. The data would demonstrate that over the millions and millions of users on a number of cloud platforms that their rates of data loss are significantly higher than your "home spun" storage. Then you would take out the outliers and show an honest distribution on the most stable vendors(since I'd expect the majority of data loss over all vendors to have some amount of locality within a specific vendor). The result of this will be the rate of data loss on the most reliable cloud platforms. You can then compare this against your own success rates.

If you have ever(even once) lost even a single byte of data(for any reason), I do believe you will find yourself hopelessly outmatched.

> demonstrate that ... on a number of cloud platforms ... that their rates of data loss are significantly higher than your "home spun" storage

You seem to be missing the point.

People are not saying "don't use cloud storage at all"

They are saying "don't use cloud storage exclusively"

You might trust that your cloud providers will never be hacked, that your individual account on them will never be hacked, that they will never suffer a catastrophic failure, that they will never go out of business, that they will never be fully or intermittently down at the moment you need to retrieve some data, and so on and so forth, based on their claims and documentation, or that you won't cock up at some point and cause data loss in your account, but I prefer to have my own copy (or copies) as well as the ones that are "in the cloud". For truly essential data at least one of those copies is both offline and offsite.

Unless you've personally memorized all the data in your head, you're trusting someone or something. Even if you've printed it all out on acid-free paper and stored it in a bank vault, you're trusting the bank, you're trusting your paper supplier, you're trusting the courier who transports it back and forth. If you've stored it on "your" servers you're still trusting your hard drive manufacturer, your RAID firmware, your OS code, your hosting company, your sysadmins.

For most values of "I have x hours and y dollars to safely store z gigabytes", a pure-cloud solution (possibly involving multiple independent cloud providers) has a lower chance of failure than one involving local storage.

>For truly essential data at least one of those copies is both offline and offsite.

But the fact remains that where ever you choose to put it "offline and offsite" the odds of it being lost are orders of magnitude greater then its persistent and redundant storage on a reputable cloud provider. Even if you put it on the most stable storage you can find and lock it in a underground safe. You can't guarantee its integrity a handful of years from now let alone centuries.

To put it in other words you are advocating storing your money in a mattress because you don't "trust the banks"

You're entirely free to play fast and loose with your customers data.

It is interesting that even today there are still people arguing against having back-ups because 'the cloud', you'd think that we ITers learn from our mistakes but for some this has to be personal experience before the lessons are learned. Good luck if and when it happens to you, please re-read your comments here at that time. There isn't an IT person that I know that has not been saved by a back-up at some point in their career, the fact that you are lucky so far is not a reason to think you are exempt.

And no, nobody is advocating storing your money in a mattress because you 'don't trust the banks', the point is that you can keep a copy of your critical data for very little money to safeguard against an eventuality that seems to hit the IT world with alarming regularity, even if they use 'the cloud'. Fuck-ups, hacks, disgruntled employees and failures all happen, every day.

Maybe ask the people of Greece if they "trust the banks". Shit happens, a local copy of your stuff does no harm at all in the same way as some cash on the hip is cool in case your card stops working.

Do you keep your money in a shoe box under your bed or stuffed in your mattress? I'd hope no, but then that begs the question "where do you keep it?" I'd assume not in a bank because any bank can fail at any moment and they are only insured by an instutition as flaky at the United States Treasury department. Governments collapse all the time just ask the Soviet Union. Right?

The failure probabilities are relative. And everyone here is equivocating. As if the likelihood of failure is evenly dispersed. You indicate that storing customer data in the cloud is "playing fast and loose" but I'd argue the opposite. Not having a cloud backup is what is "fast and loose"

Imagine you are an independent bank. You need to move your customers deposits. Do you load sacks of cash into your corporate minivan or hire an armored car service? Well lets look closer. With the latter you are giving up control, right? You have no control over the quality measures an armored car service might take. Yet somehow not contracting them seems like foolishness. The reason is obvious. Its because you know in this case that transporting money isn't your expertise. Its not something you can focus adequate time and resources on perfecting. You also can't spread the risk of failure across a lot of customers, absorb that failure, and make your service better for the next go. The exact same logic applies to long term storage of data. If that isn't your only function its extremely hard to get it right.

I'd assume not in a bank because any bank can fail at any moment and they are only insured by an instutition as flaky at the United States Treasury department. Governments collapse all the time just ask the Soviet Union. Right?

If you bank fails, you will be made whole with money from the FDIC. Money is fungible. Any money will do.

Data is not similarly fungible. There's no IT FDIC to replace your lost data with other data that's identical.

Oh you can't trust the FDIC they are government insured. And governments collapse all the time. I gave an example. Like a I said there is only one safe place for your money. Thats in your own private bank that you build and maintain yourself.

When is the last time the FDIC failed versus the last time an online storage vendor folded?

Its true that FDIC has lost fewer deposits than S3 has lost objects, but as usual there is a caveat. There are 2,000,000,000,000 objects in S3. In order to make a comparison the FDIC would have to insure over 300 deposits for every man, woman, and child on the planet and maintain a failure rate of '0.000000001%' or The FDIC would have to operate in its current state continuously for many millennia without interruption. I don't get a sense that the contrarians here appreciate the numbers involved.

You're evading.

When was the last time the FDIC failed versus the last time an online storage vendor folded?

Anybody who remembers nirvanix is going to have the willies just reading this thread, those who were caught by it will see red.

Personally I think the risk that AWS fails is remote, but if I were to store a bunch of data with them I'd most definitely make sure that I would not be dependent on them. It would be a convenience at best, but never a dependency.

I keep my money spread out across multiple accounts with multiple banks up to the insured limit. Because yes, banks can fail.

> You indicate that storing customer data in the cloud is "playing fast and loose" but I'd argue the opposite. Not having a cloud backup is what is "fast and loose"

Basic reading comprehension failure, that is not what GP is arguing.

He's arguing that if your live data lives in the cloud your backup copy should not be in the cloud and vice versa.

>I keep my money spread out across multiple accounts with multiple banks up to the insured limit.

Insured by who? Not governments. Governments collapse all the time. The only safe place to keep it is in your own custom bank at home that you built yourself.

And that money is worthless if the issuing government collapses. There is no ultimate security.

> To put it in other words you are advocating storing your money in a mattress because you don't "trust the banks"

Money and data are not directly compatible here, as one "bit" of money is the same as any other in most respects, but I do always carry a minimum amount of cash in case my cards stop working or are lost/stolen, have a pot of "emergency cash" at home for similar reasons, and recommend others do the same. Most of my money is in the bank so I'm not protecting against massive institutional failure here, but I am protecting myself against a variety of possible temporary inconveniences and possible mistakes on my part.

I'd rather have two backups of which one is more likely to fail then the other than just one, or three rather than two, and mixing types of backups means not all of them are subject to exactly the same failure modes (my local backups, online or off, are unaffected by connectivity issues for instance). For data that is important, a little paranoia is healthy IMO.

> To put it in other words you are advocating storing your money in a mattress because you don't "trust the banks"

In this case, money can't be compared to digital data. You can make multiple copies of digital data on our own but not of your money tucked away into a mattress, that would be illegal.

It doesn't share the exact same properties, but you can make the comparison with any given copy. So yes the analogy holds.

It is not alarmism at all. All it takes is a single pissed off employee or a single fat fingered command and an entire business can be wiped out. Coming from the database world it the saying was there are people who are crazy about backups and there are people who have not needed their backups...yet. For some reason everyone ends up having to learn the hard way.

I also not advocating against the cloud, just that a single copy in S3 is not a backup solution.

In all fairness, you need more than just off-cloud backups to guard against a single pissed off employee (make sure you're locking S3 and your local backups with different credentials, and the credentials to read the backups can't be gleaned from the credentials that write them) or a single fat-fingered command (kingdomofloathing.com once blew away N days of user progress by accidentally running a backup script in the "restore" direction against prod data).

Of course! Offline backups are only one of the many things that need to be done. S3 versioning can help with fat finger errors, credential management helps with rogue employees, bucket policies can copy files to off to private buckets/glacier, etc...

There is no single perfect solution, so like security you layer your disaster recovery.

> make sure you're locking S3 and your local backups with different credentials, and the credentials to read the backups can't be gleaned from the credentials that write them

This is more or less an exact copy of how I would advise companies to set this up.

S3 seems great today, and is undoubtedly more technically reliable than any home solution, but history is littered with companies that closed down on very short notice leaving their users' data deleted or inaccessible. Not to mention that billing disputes or legal action could also affect your ability to get data from them.

(Jason Scott is speaking from experience here, as one of the people called in to do emergency archive response when these cloud businesses shut down.)

Edit: jacquesm points out the sudden death of cloudspaces.com, which I'd forgotten about: http://www.infoworld.com/article/2608076/data-center/murder-...

A single computer security problem (which happen on a daily basis, although not usually at this severity) could enable your cloud to be deleted.

The "counterparty risk" is small but ineradicable. The finance industry re-learnt this with Bear Stearns recently.

> history is littered with companies that closed down on very short notice leaving their users' data deleted or inaccessible

That's absurdly unlikely to happen with S3.

How likely was it that one plane would fly into the World Trade Center, let alone two?

I'm not going to go too deeply into speculative scenarios, but all kinds of business-interrupting calamities are possible with very low probability. We just had the story on the frontpage about the Juniper backdoor; do you think S3 isn't being targeted by multiple state intelligence agencies?

Basically I agree with https://news.ycombinator.com/item?id=10772321 - by all means keep the working copy online, but have a local offline backup.

Pull out the more general example. Unexpected shit happens. A bad code push deletes data, a disgruntled employee decides to leave a "present" after they leave, etc. All things that have happened before. Having a backup away from production makes it much more easy to recover from unplanned outages.

"That's absurdly unlikely to happen with S3."

How absurd was it that money market funds would fall below $1 during the recent unpleasantness ? It was so unthinkable that major parts of our entire global economy were predicated upon it never happening.

It happened.

Money markets are even more "serious business" than Amazon, or even all of tech is, and had even more big brains attesting to their inability to fail. They failed.

Right. If that happens, I likely have MUCH bigger problems to worry about, e.g., finding the nearest VAULT-TEC vault.

Yeah but once again this problem is statistical in nature. Admittedly the cloud isn't perfect but the question remains. Does the "Counter Party" risk outweigh the chances that my home spun solution would fail? Since the likelihood of failure on my part is orders of magnitude higher then the "counter party" risk would have to be huge, like really enormous. e.g Amazon would constantly have to be on the brink of imminent collapse. This is not the case.

If Amazon went out of business there is no chance at this point that the loss would be total and catastrophic. Its essentially a bank at this point. Its shutdown would be orderly.

The idea here, and you seem to miss this point entirely is that for your company to fail then both systems would have to fail simultaneously. That is why you have back-ups. Any one system can fail, but for two systems (both the original and the back-up) to fail catastrophically at the same time is statistically very unlikely, and with a back-up here I am just talking about a secondary system that just pulls the data in on a regular basis. Tarsnap or a simple script that stores the data in a set of directories, anything is better than nothing if you lose all your data. Rebuilding will still take time, but at least there is something to rebuild with.

For any one system to fail is perfectly possible and in fact should be expected to happen at some point and that's why you design against that.

You also make sure that it is not the same people that have access to both systems so that if your sysadmin walks onto the floor with a bad hairday your back-ups will still be there. And this is also why you test your back-ups to make sure they actually work.

And what you keep missing the point on is that Im not saying don't have a personal copy. I am saying if you have multiple copies the safest one is undoubtedly in the cloud and by no small margin.

Bank shutdowns are not orderly because banks are important; they're orderly because banks are heavily regulated. If Amazon has an orderly shutdown, it would be because of the selflessness of its owners and managers, which is a pretty fickle thing to bet on.

I certainly think this level of regulation is on the horizon given the number of businesses who use S3 as an authoritative repository. If Amazon needed to shutdown S3 even today I can see the government asserting a heavy hand in it.

You need to think strategically about what you put on the cloud, because it's easy to overdo it. The decision inputs for an individual, startup and enterprise are all different.

The cost of getting stuff out of Amazon is very dear. While the statistical likelihood that Amazon will lose your stuff is low, the likelihood of Amazon or a competitor doing something to cause you to rethink your use of AWS (or force you to move) is much higher. Wanting to reduce your AWS costs in the long term is a near certainty.

As vendors move away from perpetually licensed software this gets more important. What happens when a vendor (say Microsoft for strawman purposes) decides that some function that is important to you must interface with Azure, and Azure only? What if Amazon decides upstream network transfers are no longer free? Those sorts of changes can break your business, and vendors like Amazon/Microsoft/Google are fully enabled to change those terms.

"TL;DR I hear this argument all the time. The cloud isn't perfect but its a hell of a lot closer to anything I could achieve. "not invented here" syndrome won't save your data."

I think there is an easy rebuttal that should be considered...

First, the "nines" rating of any service or resiliency is just gibberish. Go find the statistical likelihood of money market funds "breaking the buck" or of CDS blowing up - both in 2007/2008. Those had a lot of nines too and a lot of very smart , well qualified people attesting to those nines (in venues even more serious than IT).

A highly complex system becomes incomprehensible, even to the people that built it. Those nines mean nothing.

Second, you absolutely can build something more stable and predictable than Amazon precisely because you're the one that built it - which means that it is more comprehensible and fails more predictably and gracefully.

I don't care who does the calculation and how many nines they come up with - if you load FreeBSD on two bare metal servers and put them in two different datacenters and run them with any kind of conservative and cautious sysadminning you'll have a better solution. Yes, it will be more expensive.[1][2]

The standard closure to a comment like this is to refer to Talebs Black Swan and Antifragile books ... which you certainly should read ... but even more important is "Normal Accidents" by Charles Perrow[3] which I hope will convince you to stop looking for complex things that never fail, and instead look for simple things that fail gracefully.

[1] ... but we have a HN-Readers discount - just ask!

[2] You know who we are.

[3] https://en.wikipedia.org/wiki/Normal_Accidents

> I don't care who does the calculation and how many nines they come up with - if you load FreeBSD on two bare metal servers and put them in two different datacenters and run them with any kind of conservative and cautious sysadminning you'll have a better solution. Yes, it will be more expensive.[1][2]

Nonsense. It's very easy to have that kind of setup fail - remove one from the load balancer, take it down for maintenance, whoops we removed the wrong one from the load balancer. I've been in similar-sized companies using dedicated servers or AWS, so I've seen both sides. It's like how people feel safer when driving themselves than when being driven by a professional, even though a professional is overwhelmingly likely to be a better driver - everyone thinks "oh, there's no way I'd make that kind of mistake".

It is very easy to overengineer "high availability" systems - I certainly think things like STONITH and dedicated control planes are more trouble than they're worth, and running a system that you don't understand is a recipe for failure. But I'll take FreeBSD on a basic AWS setup - four EC2 nodes (multiple AZs), and an ELB - over bare metal any day. Physical server maintenance is not where I have a competitive advantage, and would mean more complexity to understand, not less.

"Nonsense. It's very easy to have that kind of setup fail - remove one from the load balancer"

I'm sorry - you're already missing the point.

There is no load balancer. There's no firewall. There's no services running except for sshd.

I guess I should qualify what I mean by "better", though. What I mean is, "I know exactly how this will fail and it won't be interesting or surprising. Or take any thought or time to fix."

It will be comprehensible. "FreeBSD on a basic AWS setup - four EC2 nodes (multiple AZs), and an ELB" is not. You have no idea (nor do I, nor do probably most folks at Amazon) the different and fascinating ways that will fail.

(disclosure: we use EC2 instances for backup DNS. We are not anti cloud or anti amazon)

If there's no load balancer then is the second machine just a warm spare? Do you get paged to get up and reconfigure things when the first one fails in the middle of the night?

>Second, you absolutely can build something more stable and predictable than Amazon precisely because you're the one that built it - which means that it is more comprehensible and fails more predictably and gracefully.

This is from you and this is from the wikipedia article on NIH(not invented here) syndrome ...

>Not invented here (NIH) is the philosophical principle of not using third party solutions to a problem because of their external origins. False pride often drives an enterprise to use less-than-perfect invention in order to save face by ignoring, boycotting, or otherwise refusing to use or incorporate obviously superior solutions by others.

I am not saying don't have a copy, I am saying the if you have several copies the safe one is in the cloud.

What is your recourse if service level is not met? A monthly credit? Your data is probably worth more to you than the value of its hosting. Too many people rest on the laurels of SLA.

I argue frequently that the cloud causes cognitive dissonance. People want the ease of use and reliability of centralized outsourced services, but other people want data privacy guarantee and control of those services. If anything is a threat to one or the other, it's the fact we fail to understand the various requirements from distinctly different use cases.

These polarized "requirements" are usually rationalized away by each group. As a data privacy advocate, I believe that I can provide a reliable storage solution to my company without using a centralized cloud service, which then guarantees my privacy because "I'm in control". As an advocate of centralized cloud services, such as Amazon, I present that their team is better at security and reliability than any other team on the planet and that I can encrypt something and trust that my key management is secure. Both of these arguments have fallacies and assumptions.

The solution is to challenge ourselves to build better solutions. At what point in time did technology advancement ever slow down? At what point will we ever stop and say "Y'all, this here compute system is good enough and should be centralized/decentralized!"? Never, I say.

I think you missed the point. S3 isn't free. You pay for that service:

    and if you’re a person they are giving it to you without
    you signing anything accompanied by cash or payment that
    says “and I mean it“.

Are you talking to me? I never said it was free. No storage solution is free.

He's saying that the article does not apply for S3 and other paid services.

Yeah I guess I can see that. But then again I would never give important data to a 3rd party without some sort of contractual arrangement.

How will a contract save your ass if your data is lost? You can't outsource your responsibility.

Unless you're personally memorizing all the data then you're outsourcing the responsibility somewhere - either you're paying for a service that you hope will be compliant with its specification, or you're paying for hardware that you hope will be compliant with its specification and employees that you hope will be as skilled as they claim to be. There's no way to eliminate the risk entirely, all you can do is take the safest option available. These days that's often the cloud.

> These days that's often the cloud.

For plenty of use cases I actually agree with that conclusion. With the caveat that I would add: accompanied by a suitable non-cloud back-up of critical data, code and configuration data held on a medium that is not accessible by the same people that administer the cloud stuff.

I think for most levels of safety you would target, a pure cloud solution is going to be cheaper. E.g. if you decide you need a backup that's independent of your primary cloud provider and any staff with access to that, you can probably accomplish that more cheaply with a second cloud provider than by self-hosting.

Yes, a second cloud provider is a valid option, if you take a number of precautions (see elsewhere in this thread).

What is it with you?Will all your home brew nonsense save your ass when your data is lost. I am basing all of my arguments on "what is the statistically most likely to fail" and "given limited resources where would you store your authoritative copy". The answers to these are "whatever you come up with will have a higher likelihood of failure" and "the authoritative copy goes into the cloud" You must be a terrible gambler. "Will playing the odds save your ass if you lose your money on good bet" no. But making good bets will increase your return regardless of the outcome of any given bet.

> "whatever you come up with will have a higher likelihood of failure"

This is where you are simply wrong.

> "the authoritative copy goes into the cloud"

This we agree on.

So, where you are wrong is this: you are only considering technical failure modes but there are many others besides and so any comparison between 'the cloud' and 'your get-out-jail-free-card-backup' would have to be in terms taking into account all modes of failure, not just the technical ones and then on top of that you'd have to consider the likelihood of the backup failing to restore at the same time that the primary system goes down for whatever reason (including technical ones).

That's why I call it a blind spot, even after being told in 20 different ways that the uptime of Amazon does not come into play here you are still clinging to that. It is simply not relevant because that's not the scenario a back-up will most likely prevent against. It will also prevent against that scenario, on top of the ones where someone incapacitated by drugs, anger, depression or any one of another set of circumstances decides to take it out on your company. Or maybe someone makes an honest mistake (I've been called out twice in my career to restore data for companies that had been wiped out because of simple mistakes that should have never ever happened).

So back-ups are not a luxury, they are a necessity and the degree to which Amazon can outperform your back-up in terms of availability is not a factor in the whole discussion.

Well, unless Amazon blocks your access to your account. Have you accounted for that in your failure rate calculations?

Ask Bernie Sanders about the cloud and the implications of losing access to it at the whim of the cloud provider.

We've been doing it for a long time. It's called some reliable servers in multiple locations with copies of the same data plus some offline backups (esp tape). There's many non-technical, Fortune 500 companies whose data has lasted longer than Amazon's existence. Cheapest version of this involves two high-capacity boxes at two locations each with maybe a colo agreement. Let's you keep replacing stuff as servers and drives fail. I've even done it with embedded boxes (VIA Artigo's) in the past where data volume wasn't high.

Do you control Amazon? No? Then you're not the captain - you're just a sailor and someone else is in control of your data.

Now go put your stuff up on S3 but please keep a backup of the things you upload there.

I do keep a back up of things, but the back up is on S3 because that is the most stable storage option available to me. Everything else is the risky copy.

>Do you control Amazon? No? Then you're not the captain

This is textbook NIH syndrome. Rather than looking at the relative probabilities you look at something from an ideological perspective. Your argument isn't "Its safer with me" its "I want to be in control of it" This attitude is generally harmful. Think about your money. Are you in control of the bank? no right, but you still keep your money there instead of in a shoe box under your bed. Why? Because the bank is better at protecting your money than you are. They have big heavy metal doors and men with guns to move it from place to place. Amazon is like a bank for your data.

> This is textbook NIH syndrome.

I'm not saying you shouldn't use S3. You should just make sure that it's not the only service you have the only copy of your data hosted.

>Rather than looking at the relative probabilities you look at something from an ideological perspective.

No, I'm looking here solely at the risk of putting all your eggs into a basket you don't control.

This is the perfect point of the article - use the cloud (hey, it's "free") but KEEP A COPY. Otherwise you will get boned one day!

Might as well just change "use the cloud" to "use a computer".

What if you make a mistake a manually delete something in the cloud? What if a program has a bug in it and something gets deleted? If your data is precious, you need multiple copies of it, no matter where they live.

The problem with the "once in "thousands of years"." argument is that you never know when that once happens and it may well be... tomorrow and never more in the next couple of thousands of years.

That is what happened (once again) to John Meriwether and Long-Term Capital Management... A "once every 10.000 years event" which took place... today.

> Amazon's S3 offers service level agreements with failure rates that at one point implied the statistical likelihood of losing an object to be once in "thousands of years".

Where things like that in 2009 when this article was written?

> I don't understand this logic.

Curious if Upton Sinclair would have anything to do with it.

Lets put it this way, I know enough not to store data on the cloud(s) exclusively.

Reply on news.ycombinator.com