 Is everybody ready for our last set of sessions? That's so good when that works, okay, cool. All right, we have one more set of breakout sessions, and there's just two in this session. So the first person pitching their session will be Greg. If you want to come up and use the podium. Okay, and I'm glad that the last session was a brainstorming session because my session is also a brainstorming session, which is to say I made some effort to prepare for this, but really just realized I don't know what I'm talking about. So I need you guys to come and bring your opinions and views and help us figure, start figuring this out. Basically, which button is which here? This one? There we go, okay. Basically, MLab has been built, I think kind of built in the assumption that people were sitting at home on a hard-wired ISP with a machine and running speed tests, and we never really anticipated, I guess, how fast the world was going to move to mobile, and we haven't done anything on the platform so far to support the fact that the world is moving to mobile, and that mobile measurement is somewhat different than static measurement. So we just basically set things up in such a way that we can aggregate by IP address. IP address is kind of the only thing we... It's kind of the best thing to aggregate by because we have a few other bits of signature information we can use to distinguish whether there might be multiple clients at the same IP address, but really it's... Even when we do de-duping or try to reduce the impact of very high traffic clients, it's at the IP level. And so even in the home, one of the home or business environment, one of the things that we're realizing is that there are other factors beyond what we are measuring that may influence the connection behavior, and so we're becoming aware that we're not even measuring those things. We don't have very much visibility into those things, whether somebody's on a Wi-Fi, whether somebody's hardwired, whether they're at the far end of the house and the Wi-Fi connection is bad, whether they're an apartment complex and there's a lot of interference. What other traffic there is from the teenager that's upstairs screaming and downloading stuff, or what kind of connection it is. And then when we go to mobile, we kind of know even less of the important things about the environment. We don't even necessarily know that it's a mobile connection. We usually can tell that it's a mobile client device, but we don't know whether they're on a Wi-Fi or whether they're on a cellular network, what kind of network they're on and so forth. There are...a lot of this information is available on the mobile device, but we are not collecting it and we're not storing it and we don't have it available. So, big question is what should we do about this and how do we do it in a way that doesn't cause us privacy problems? Since we are radically open in publishing all the data, we need to be very careful that we don't publish anything that leads to privacy problems that would come back to haunt us and haunt our users. At the same time, though, the things that are much more relevant from an environmental point of view for mobile measurement are really the lat long of the client device, what tower they're connected to, what the signal-to-noise ratio is, and as people move around, if we're not careful about how we identify the tests, we could make it fairly easy to track people and we can't let that happen. So, we just want to get together, do some brainstorming, see what people think are the needs, why do we need to do things differently, what do we need to do differently, and what possible outcomes could we get if we bake some of this into the system. And Greg's session is going to be in the smaller conference room that we used on the first day for the net neutrality measurement. So, the one in that corner over by the entrance. Do you want to go with Greg? Okay, I'm going to let you talk. You know what? Okay, good. So, I wanted to make sure. You mentioned something about making slides. I was like, what did happen? Cool. So, a thing that we care about with the data that we have is this idea of we have all these measurements and people say, okay, what's going wrong on the Internet? And we go, it was really hard to collect the data. What do you want from us? Maybe you could figure that out. It's hard. And we'd like a better answer than that. So, I built a small system that starts talking, that allows us to start talking about that and hopefully can kickstart the conversation about how we should assess what is an anomaly? What isn't an anomaly? What are the key metrics or the key aspects of something that should be considered anomalous so that we can figure that out? And just basically this problem of anomaly detection is something everybody wants us to do, but everybody also agrees that we'll point to a random person and say they should definitely put the bell on the cat, right? And so, we don't know who's going to put the bell on the cat, but we should at least talk about what it looks like. And that's our goal for the session, to what it's an anomaly and help us help you to find them and agree on terminology.