 down the road, okay? Next part, laptops to cloud. So that's why I asked earlier, you know, you have this great bit of kit like on the left circa probably 98. Is that Windows 98? Yeah. Oh, Windows 95. That's good. And you say to yourself, well, I've got this code, it's taken forever to run on my laptop, which has, you know, a gig of RAM and a slow hard drive. I'm going to move it to the cloud. And then you find out, okay, you start asking questions, why is my Python job only running 20% faster? The problem is, and you'll see here on the next slide, I'll flip back, is that this incredible genius John von Neumann designed our computer architecture over 70 years ago. And in this classical world of computing where storage lives separately from computing, you end up in today's modern era being IO bound, okay? And what does that mean? Again, this is pro tips for the analysts in the room. You are being constrained by the speed at which data can get to the brain of your device. The actual processor in that laptop may only be 30% slower than the fastest Amazon EC2 instance, the X1, which is my favorite. But in truth, you're pulling data off a hard drive. The hard drive is going to be slow, it doesn't matter what you do. So for us, as we do these kinds of things, we're thinking about how can we improve speed? And in this case, it's about IO. So parallel Python won't get you there, splitting files won't get you there, Hadoop could be, but ultimately you end up being Nick bound. The reason why I bring this up, this is more of the meat in the sandwich part, you need to think more broadly about the tools you use and oftentimes help to define the architecture so that the tools you have at the airport to run your code are not just designed for all of the pilots, but they're designed for you to get a good experience out of the execution, okay? Good compute experience. So now here are a couple pro tips, RAM disks, IO is the enemy, use Redis, split files and parallelize, okay? I promise life will be better. Now the second part