Okay. Bottlenecks using traditional file system. So, imagine this isn't reusing NFS. Went out, we bought a net app FAS3200. It stores 2880 terabytes or 2.88 petabytes of data, one big rack of these storage devices in there. I don't know if it can communicate at 100 gig ethernet. It might be limited to 10 Gig for the book, just imagine okay, for a moment that it is capable of communicating at a 100 gig. You've get your Watson and you want to do some serious data analytics, crunching on all of that data over there because there's, we're willing to mine that data and extract some value out of it, see what's going on, see if we can determine some patterns. We'll get into that in machine learning and data analytics segments. But we just need to train. We need to get it all over here, we got to read it all. So you do the math on it. 28 terabytes divided by 100 gig turned into this many bytes per second. I don't know what the ECC checking an overhead as I think I downgraded by it might be 85% efficiency of that or something. Anyway that's a long time, 256000 seconds, just 71 hours. It's like three days. Yeah three days to read all your data, you want to wait around three days to read all your data? Those is nothing, okay, in terms of data volume. What happens if you've got 128 those net app file servers and you got 100 gig ethernet connection? Your pipe is too narrow. When you're dealing with data on this size, it's massive scale, it's no longer efficient to utilize this traditional file system. This is what has given rise to these new file systems we're going to take a look at. They're fast, but thy're not fast enough because the size of the data is just enormous. So better scheme is to push compute out toward the storage is. Have a very fast switch in the middle in case the server needs to access some data over here. Okay, fine, but we want to push compute out to where the data is. So this server is accessing this data locally and this servers is accessing the storage locally etc, etc, etc. We get much better overall utilization of the system and data only has to flow across tier when necessary. But the primary data crunching would take place this server accessing the data on the storage device.