 Okay, this is Apur. I work for Mintra. I have played a number of roles there. In interest of time and due to my laziness, I haven't prepared any presentation. So you have to trust my word. One of the questions which I see happening in most of the talks was that how do we monitor system parameters? There are systems like Nagios or Senso who are alerting them. But when you want to debug them, there is not much information available in those systems. So let's say your CPU shot up to 100%. You got a warning from Nagios or Senso. But when you log in, there is not much available there to debug for you. Also, there are times when your application started behaving erratically. It's the application, either a lot of errors or response time increased. At that time, you can't correlate whether your box was behaving weirdly, whether your database was behaving erratically, etc. So there is a very good tool called StatsD, which has been outsold, an open source by a company called Etsy. There is a very popular blog post from Etsy on StatsD, which you can check there. They have explained how this system works. If I have to give a 30-second introduction of that, it's a service running over UDP, where you can push in any packets. You can virtually store everything. The storage pattern will be a.b.c.d, so they create separate directories for that. The things you can store there are things like counters and times. So let's say you want to capture how many requests are coming to your server. That can be captured and what was the response time of the request that you can also capture. You can also go deep. Suppose you want the response time of one particular function. Your application is doing 10 things out of that. One happens to be executing a DB query and you suspect that that's the thing which is most unpredictable, so you can capture that as well. So basically it can capture anything and you have good graphing tools available. One is graphite, another one is upcoming, which is called Grafana, where you can plot those things because you are capturing n number of different metrics. Let's say you are capturing your database response database as well as your application. You can just suppose both of them simultaneously. One important thing which I have seen people using graphite, which I haven't seen people using graphite, but we are doing, is that we are monitoring system parameters into it. There are a lot of utilities available in Linux like iostat, VMstat, SAR, W, et cetera, which gives various things like load average, CPU usage, memory usage, et cetera. So we are storing that as well, simple crons running on all of the boxes which we have deployed using puppet, so it's simultaneously available to all the systems. We are in a format that environment and dot box, et cetera, we are capturing this. In case any of you are interested, then I can explain to you in detail outside.