 head. Who's there? Hey, close. Hey, how are you guys doing? And by the way, I spoke on this subject of FNW and a couple of months ago, and had a fabulous side conversation with one of your folks, Chris, on the same subject. I'm glad to see you raised the red flag because I've been throwing it out there for years. But the one thing that hasn't really cropped into the discussion, I think is an important one is the cost. And, you know, we all know the drives are getting denser and more capacious, but they're also getting cheaper and cheaper and cheaper. And there's a bit of a tradeoff between the cost of the media and also the cost of the environmentals. But as you start looking into the future, and we get to these 80, 20 and 40 and 52 terabyte drives, they're going to be pretty damn cheap when it comes to a cost per terabyte basis. So the concept of keeping, and by the way, when I talk about data dispersion, we do that to touch more for a performance aspect than anything else. But when as the cost gets cheaper, you can begin to replicate this more and more times. And that really seems like an opportunity, I think, from the technology. If, for example, you kept four copies of the data on four separate drives, not necessarily in a network environment, but within a single array. Yep. Yeah, so just to talk about cost and replication, certainly there will be some applications where making four copies make sense. But there's, we've seen some markets like social networking, those guys are so price sensitive. There's no way they're going to make four copies. So what they're looking for, and it's not just, you know, the cost of the drives, it's like four times as much electricity, four times as much Well, if somebody loses the tweet, that's not going to end the world. But if you start looking at the enterprise applications, the databases, the thing that really drives businesses alive. Yeah, I think you can afford to keep multiple copies and reduce the risk of data loss. Again, so let me just offer this as an alternative. So certainly four copies is one way to create reliability. And it has known pluses and minuses. A different way to create if you're if what four copies gives you is the ability to tolerate the simultaneous failure of three things, which is pretty good. Another way to do that, you know, is to and if let's say those were four raid copies, you'd also have the overhead of raid. So you would be physically storing probably five to six times as many bits as you started with. Another way to do that would be to do something like, you know, an extreme example. Well, let's just say, you know, like a 10 of 16 in with dispersal where you're you're spreading 16 slices, which are each one tenth the size of the original copy across 16 different servers in such a manner that you could tolerate the simultaneous failure of any six of them, that would give you comparable reliability to four copies. And the but then when you add up the physical number of bits with four copies, you're going to have if they're raid copies, you know, at least, let's say five to six X what you start with. So it's, you know, 400 to 500% overhead. With the 10 of 16 dispersal, you know, there the slices are each one tenth the size of the original, there's up to 16 of them. So at most you have 160%. So you'd have 60% overhead in your storage for comparable reliability, versus, you know, 400 or 500%. So so there will be many customers that will want to have, you know, that kind of increased efficiency in terms of electricity, floor space, power, you know, all that kind of good stuff. Yeah, I think I sort of categorize the world into two components. One is the the old structured data that we sometimes tend to ignore. But really, like I said, runs the applications worldwide and then the unstructured content depots. And the the impact of data loss between those two worlds is entirely different. And you tolerate loss in one part of that world without serious disaster. And the other part of the world you can't. Yeah, we may be talking about two different solutions. Right. And if you and if you're dealing with multiple sort of raid one copies, the rebuild time and the impact on performance during rebuild is substantially less. Isn't that correct, Klaus? Yeah, absolutely. So so so back to one of the earlier comments, if I will, if I look holistically at a database application transaction application, raid one with multiple copies may continue to be the most cost effective and reliable way to go forward for that type of data for the rest of the data, which is, which is where the bulk of the storage is some other algorithm may be more appropriate. So go ahead. That's my point. I think, I mean, even rate five or eight, six and database applications, the rate is done at the right level is acceptable in terms of performance. And, you know, the other part of this, and it gets into the cost is what I call capacity efficiency. So we're seeing a lot more duplication compression, you know, different ways to get more out of the capacity. And so that sort of mitigates what you just said, Chris, in terms of the cost of additional copies of storage. Yeah, but why? But so for example, if you look at deduplication or compression, if that gave you, let's say a forex gain and efficiency, that would be pretty great. You know, or you know, that's typical, you know, with dispersal, I think you, you know, in the example I mentioned earlier, you would also have about a forex gain inefficiency. So there's no reason, you know, so if you were to dedupe and then store the deduped data on a dispersed storage system, the realized efficiency would be the product. So it'd be four times four. So it'd be one, it'd be 16 times more efficient. So that's, you know, so forex efficient is great. But forex squared is even better. Yeah, that's that's pretty compelling. And I agree with you that in a database environment, absolutely, you know, most of the time, you know, replicated copies makes a lot of sense. But you know, if you look at the number of bits that are in the world, you know, either on a, you know, looking at like an average hard drive, how many bits, what the bits are, or kind of coursing through the internet, the majority of bits now since since about it was around 1999 are now unstructured content. And it's now about 65 to 70% of all bits are unstructured content. That that's the area where I think dispersal really plays. We're not suggesting for structure content that that's the right fit. But unstructured content is the largest amount of data and it's the fastest it's where all the growth of storage is. No, absolutely no question about it. But that's also I mean, you mentioned rate and buy in earlier. And that seems to be, in my mind, a reasonable approach for a lot of this unstructured data. It doesn't have the performance requirements necessarily that right, does. And, you know, if it takes a month or two to rebuild the drive, and you've got three or four or five extra equivalent spares, fitting out there, that's acceptable. Yeah, but your earlier, your earlier point, Chris, of if I've compressed, if I've de-duped, and if I've encrypted data, and I lose a bit if the impact is substantially greater. Yeah, absolutely. So that always has to weigh in on this. Yeah, with good encryption, if you lose a bit, the data should be gone. That's how encryption works. And that's also true of compression. The more efficient compression becomes, the more dangerous you are of bit loss. And what's happening is compression is also being driven by Moore's law. Because of the, you know, greater availability of massive amounts of processing, people get more and more sophisticated compression techniques, such as duplication. And as they do that, which they will inevitably do, because Moore's law is not going to stop, then the physical storage system, or the application that's producing this highly compressed storage will become less and less tolerant to bit failure. I mean, you know, if you don't use compression, bit failure is not a big deal, but you start really cranking up the compression and it starts getting more efficient, which of course, it's going to. And then, you know, certainly if you layer encryption on top of that, then individual bit failures become huge problems.