RAID vs No RAID choice

That’s why you use a RAID configuration. If none was present and your drive died you can pretty much start over now.

But it’s a good time to learn about RAIDs

As an SNO you are responsible that the data you are storing is always available, if it is not you are failing audits which will result in suspension and eventually in disqualification. In your case if there is no RAID configuration and no backup your identity is lost. You will have to create a new identity and start over. And try to learn from your mistakes.

1 Like

Official recommendation is, not to use any RAID and rather setup one node per hard disk. Data is already redundant in the network, and the space you “waste” with redundancy in a RAID could be used to earn extra money. Yes, it’s unfortunate if you lose a node, but statistically you’re better off by setting up one node per hard disk.

3 Likes

That would mean you have to create a new identity every time you have a drive failure and then wait 10 months to get back to the same earning rate as before. Even though you have some overhead with a RAID i would still recommend it over no redundancy at all.

You know how statistics work? There is an average failure rate of 2% per hard disk per year, IIRC. So how often does one have a failed hard disk?

And if you have a second node running instead of wasting it for redundancy, this node has already made you money and will take the most traffic while the new one is vetting.

This discussion has been done to death here. Search the forum if you want to read more about it.

3 Likes

Raid is not part of any Storj recommendation.

2 Likes

This statistic means that 2 SNO of 100 those who follow the storjlabs recommendations lost their nodes in this year. Maybe it will be you, maybe someone else, but definitely not me, because I have raids.

3 Likes

That also means 98 of 100 would make more money.

The odds of getting hit by a bus tomorrow is >0, yet you get out of the house every day. Maybe not at the moment, but you get the idea… Or maybe not :wink:

2 Likes

Wrong. Money makes on used space and egress. This not about raids.

I don’t get, what you wanted to say. No RAID means more usable space and more used space means usually more egress since you have more data customers can download.

I say that you are mistaken in this.
If you used space depend of count of hdds you have, just by another one or two.
In a long term you profit depended of you reputation and age of yours nodes. You can stay on one hdd per node but you never earn more money than me with raids.

The opposite has been demonstrated multiple times on the forums. You will generate more revenue on average, in the long-term by running one node per HDD (assuming all of your nodes fill up), though your revenue will be more variable in the short-term.

The extra revenue that you earn by not dedicating some disks to redundancy will be larger than the revenue you lose when a disk fails.

1 Like

It is wrong assumption. This “calculations” make SNOs who play in storj. Im not play with it.
I have replaced two drives to this moment. If i had not setup raids i would loose some thousands dollars. But all that i doing is just replace cheaper hdds on the fly.

Please check your tone. You are not the only professional SNO, such as there currently are.

I just share the experience. English is not my native language and if you can distinguish tone in my texts, then it’s just translation difficulties

But since you have touched on this topic, what volumes do you personally keep?

The part that’s somewhat offensive is the implication that serious or professional SNOs would agree with you, and that those of us who disagree are “lesser” or “just amateur” SNOs. This is both a bit rude and is a combination of multiple logical fallacies (ad hominem, appeal to authority, no true scotsman).

Many SNOs have worked through this thought experiment and the general consensus (though it’s not unanimous – there are some who disagree) is that unless your drives are failing on average more than once per year, running a single node per HDD will result in more revenue in the long term because the extra income provided by the additional nodes will be greater than the income lost when an HDD fails and a node needs to restart from scratch with a new HDD. If your drives are failing more often than once per year then you’re probably just going to lose money buying new HDDs anyway (with the exception of warranties that you may be able to claim).

I am currently operating multiple nodes whose capacity sums to almost 11TB and will likely be bringing a few more nodes online in the next few months.

My strategy when I have two disks of the same size is to run a node on a RAID1 using both disks. When the node is nearly full, I remove one of the drives from the mirror, reduce the number of RAID devices to 1 (it’s no longer a mirror) and start a new node on the second drive. This gives me redundancy while the second drive would be unused anyway so if one drive fails early then I can absorb that without losing the node. However, once that node fills up then it becomes more profitable to start a new node on one of the drives and effectively double the income potential.

There is definitely an argument to be made for using RAID5/6 to operate very large nodes simply because it eases management tasks. You aren’t having to deal with setting up new identities and you only have one node to monitor. Note that this is a decision unrelated to node revenue, but perhaps you decide that it’s worth it to earn a little less revenue if it makes the node easier to operate. That is definitely a fair position and I wouldn’t argue with it.

Here’s how I put it: using RAID levels with redundancy is an insurance policy against a drive failure. When a drive fails, the policy “pays out” as you don’t lose any revenue. However, every insurance policy has a premium – a price to be paid for that policy. In the case of Storj, you pay the price of lower income potential for the benefit of a disk failure not lowering your node’s revenue. You give up some revenue to stabilize your revenue stream.

In the short term, an insurance policy can pay off. If I get a new homeowner’s insurance policy and my house accidentally burns down next month then I will have made substantially more money as the policy payout will be far greater than any premiums I’ve paid so far. However, in the long run, most people can expect to lose money paying insurance premiums. We’ve just decided that it’s worth the extra security to pay some money in exchange for eliminating the risk of catastrophic loss.

That’s exactly what happens when you run a Storj node on a storage layer with redundancy. The insurance policy can “pay off” if you have a bunch of drive failures shortly after launching the node – but the odds are that, in the long term, you’ll make more money simply by not having the policy and running a node per HDD.

3 Likes

Please do not ascribe to me your thoughts. All that I had in mind I expressed in my example and can repeat in other words.
Large farmers who are serious about business and earn their money operate with dozens of discs. Discs are consumable and are strewed regularly. If such a farmer keeps a node on the disk, he will go broke, as the old nodes all die, and the new ones do not give any profit. Only losses.
That’s all, there is no theory of excellence. Only calculation.

Personally i hold 75TB used, 30 free and over 40TB unconnected

Analogy with household are completely wrong. Wrong analogies lead to erroneous conclusions.
Now i have in holds more that i get paid. 90% of this holds from 5 old nodes. If i loose this escrow i loose much more, ten times more, than cost of few disk for redundancy.

And the statement that without redundancy you can earn more has the same meaning as the statement that Achilles will never catch up with a turtle.

1 Like

Why would you take a post of a new SNO asking for help to discuss raid vs no raid? It’s not the right place and doesn’t help that new sno… there are literally dozens of threads where raid vs no raid is constantly being discussed.
It is of absolutely no help to the person who started this thread. Even worse, most of you basically ignored his questions and only talked about this topic.
Please just stick with the official recommendation and point to a thread where raid vs no raid is being discussed but don’t use every thread about a dying hdd as an excuse to start this dicussion again.
We’re here to help people and not spam their thread with a different discussion.

3 Likes

In this case, I propose to wait for the real admin, who will transfer our discussion to the right place.
Answer was given to the OP a some time ago.

Indeed; my intent was to provide a counterargument to the advice given that RAID would be a solution, not start a discussion. Left unchallenged, OP might consider this to be the standard recommendation to users in this position.

In the future I will just link to a thread where this discussion has played out when rebutting this advice.

There is absolutely no possible way that you would have lost thousands of dollars if you would have only lost 2 nodes out of likely many many more you would have had had you run it with one node per HDD.

My node has been online since the beginning and has a total of $250 held amount. Worst case that is what you would have lost per node, but more likely it would have been a small fraction of that. Since you’re running RAID, lets say you have RAID6 with 8 HDD’s. In that setup the alternative would have been 8 nodes on 8 HDD’s and only one would have failed. You’d have lost at most $250/8 in that scenario, so what… $30 bucks? That loss would have been compensated by the money you could have made with the additional HDD space available or the money you could have saved by buying fewer HDD’s in the first place.

With annualized failure rates of HDD’s being less than 2% during the first 5 years of their lifetime, the math simply doesn’t add up. Even if you take into account the up to 10% annualized failure rates of older drives it doesn’t add up.

This is all independent of household vs datacenter. The math is the same. Unless you don’t count the cost of the additional drives needed for redundancy.
The only counter argument would be that managing multiple nodes at scale would add cost through management overhead. I can understand that. But if you’re doing this at that scale you should probably look into automating that process to begin with. I imagine a hot swappable server that simply detects a new disk and sets up a new node automatically. Some monitoring to see where nodes are starting to become full, so you can add new ones in time. If a node dies, other nodes will take over incoming traffic on that IP with little loss.

But in the end, the math doesn’t change with scale… the numbers just get bigger.

I think we’re just bumping up against something that’s really counter intuitive to people who are used to running large data operations. Doing things on individual disks is simply not done normally.

2 Likes