About the talk
Chaos engineering tests your application resiliency by thoughtfully injecting failure and starving resources. Complete failure is obvious, but how do you detect the warning signs of pre-failure stress? This session takes the capabilities of chaos engineering beyond resiliency to support capacity optimization. You already need to monitor performance to see when your code is bending before it breaks. Why not glean more insight from the data so you can prioritize efforts and respond rapidly?
So it's a technology from every side is a bender and operator, and developer and publish. This is a story in three chapters. Start with how this crazy, I did go to my head and then it's time to talk about what can be done by combining two products in the right way, and then finally, let me offer a few hints on getting stop. In a previous life, I drop capacity plans for the national computer systems and it was time to gather the numbers but the prediction was easy. We just tried out where the trend
was going and then flag when I plan to operate a little bit, I guess work in a complicated World. We're guessing our way into the future. What if we could find some 70s? This is where the Story begins and if you want to short version of this section, here it is. What does this number would have been for applications that I earned in previous jobs? And I realize it's probably from a few years back. That's why you said Simplicity, imagine your application this
weekend. So 100 make no second away after detected. I'm running for a full year is equivalent to a 97.6 Savage. So what happens if we make it a hundred milliseconds faster? With the second question did not count by capacity in this? Why? Because I didn't have the pool. So what do we need to do that? But we need two things and structure resources, so that you can experiment. And yes, for culture pants. This image released to buy that the impact of adjusting the dog. I mean, you can try them all down and wait for the complaining phone calls, but it's not a move would be to have a maid
of it. Sees, what is happening. Now. If only there was a way right now, that brings me. Now. I guess that kind of computing is really about experimentation and becomes the ability within a few of my grandma and that allows for CPU and storage and response time and so on. And the Mater, the Mater is application performance, monitoring area comes with Grandma. traditionally, we think about, The sort of testing entry wise. I blowed and volume about this in the book, The
Art of software testing. What one of these mean. Well, I think about that comes in like beating Vegemite sandwiches. You want to see how many Vegemite sandwiches you can deal with the first is, how many Earths can you fit in your mouth at one time? That's the lawyer test, right? And then what about the? Well, that would be if you imagine a white guy is delinquent, while you're eating them, how long before the title close up. But the thing is that I skipped breakfast this morning so that I wouldn't
want sandwiches. So, what you want to do, I what's my stock with the code that your customers use, my cell phone, so I can talk with your application to tell you where, that is able to observe what the car does, he uses what they actually use and where they been to see that in the back from here to where you need to experiment that you want to. And that's the first place to begin. Now this sort of waiting at where to begin. It's a little against the concept of computing but it's because it's
not type week. I'll take a random. But remember I'm just suggesting. This is a hint for finding a starting point. What would you like from disk for this conversation to die? When not say she was doing, was thinking capacity and then we experiment to see if that means Porsche license. So let's pull them. Out,, You can create outcome statements, and then experiment to load in the Badgers, just write out testing. It might be, you know, we can handle up the ex transactions before the system. We can
handle up to a transaction system. Now, you know, when the stock instead with increase, you know, where the restraints in order to do anything about it, without knowing the constraint. You don't know where to go to, send. This is fine, but we needed something about capacity. We need to ask different questions. Remember, what we're trying to do again, is iscar like capacity to the Foreman's to revenue, and customer experience. Here, might help with that. Either you or what kind of capacity
percentage? See what happens. Get measures of what happens when you get a list. I try to set up toggleable explicit, transactions. Car lightly, the customer experience, 20% abandoned Zora a 15% increase in tickets. So be thinking about that as you go through this process. Now, of course, you need solid set of applications. Is that any way for Blue Train testing in Canary, testing and version comparison transaction, pricing them and bringing all the milk together. We need to use tile to find Wildfire slow
down and that means monitoring around the title. So I your first hit. Black Hole diner in safety applications to buy. Remember to show you where the uses of your nail getting certainty Around Your Capacity to do, you know where to go in the application to address it. So you might find out of so 270. So true things. Clearly you need something that I did not have. When I stand capacity planning. I need you to adjusting the dog. Extra house today Grandma, you need some serious application for
customer experience and Revenue code to Pasadena performance, customer experience with connecting brand one and nine. Prateek. Jain is my current hero and he said how to see the grand experiment status inside after an Amex in case you got distracted. Now is the time to come back and take a screenshot of this, right? This is If you already have Grandma mathematics and vice, let me know how it goes. How is that said? Alright, so that's right. So what are the next steps?
How should you find the amount of tiles in your Enterprise application? Key business transactions, that you are building a shed back with the community after just trying to satisfy our curiosity and clients and also the user experience in customer Behavior. That with that, you're going to know what will happen if you lose capacity and when something's going wrong, the transactions from your processing and I will research and monitoring is a must-have for 39% of employees. So I can put this up
as if you'd like some here again upon to swap back and take a quick screenshot and I thought was that I want to thank you for your time this morning, and we'll move ahead, two questions. And of course. During the next few hours. I'll be in the chat room to take questions there as well. I love that part where you were just talking about like you seating capacity, not resiliency. And then how you talked about that blast radius where your testing for resiliency? And as you expand off, last 3 years and you just going for, what exactly is my customer singing and asking the right questions
and of course, when you talk about the hero that we have, that is able to actually bring together, our Dynamics in Gremlin, like that is like such a beautiful flow that I love when we see those integration top in the community, when it's not driven by us and it's like so you figured it out, you understand the value, and Now, you're sharing with the community, going back to the principles of engineering. We're like, we followed them for scientific method, but the last one is all about sharing the results, your failures, your progress. So I'm going to move on to Q&A.
We have the first question. This is a most serious question. I'm going to ask you, how many vegemite's I just. Can you fit in your mouth? Three triangles Australians that are always talking about how great it is unlikely that the berry farms near the system that users do not interact directly. Interact with so, Yeah, so they want to know kind of like when you have automated systems that thus, the users are not like they're not customer-facing. What measures
would you actually say that folks should be tracking in order for them to understand what's going on. Are you just written inside and then you get the mumps and ribbon, you get it between. Yeah, I think you could you touch upon an interesting point, many folks, like don't realize that their internal tools and though being customer-facing and sometimes the questions are like, how do we put my trays on internal to order? Like how do I make sure those are all up and it's like your
engineers. Do your customers like, you got to make sure there if they're always up. So that is very true. Where do you get the information from revenue? What is the source of Truth? Usually, so this is the source of Truth is a great while putting a drive. Right, but we needed to talk to the business. Am I say business in the game with it? That's an internal application on its own actions. Can we extract. We're out of that transaction goes by? And then yes. so, it's coming from the dollar as I love that answer. It's like part from a of just being like one point
of of truth. But also just looking at that data like we are talking about the events like this. I know we have a few more questions coming in but we're all for a time. So we're going to post the rest of Marcus questions that are coming in over at the Q. A shuttle for Marco, but we will love to thank you for being one of our speakers out of 4. Starting our morning following of the conversation from tacos. Bedroom, lights. Thank you, Marco.
Buy this talk
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.