This topic carries considerable interest when analyzed from different dimensions like Volume, Variety, Velocity (typical Big Data facets) or Analytics/Real-time, Structured/Unstructured, Local/Cloud, Hardware trends etc.
Share your perspectives in this space, the choices you have and will make.
In this talk Ashok will motivate the need for an experimentation platform. He shall discuss the various modules of an experimentation platform and share how to go about designing or building such a platform. He shall also work on the basic mathematics to develop the intuition (not just share the final formula) of when a result is statistically significant.
Flipkart.com, the website, has been around for a while. And this has given us ample opportunity to make mistakes. The good side of that is that we've learnt from those mistakes and moved forward. In this talk, Sidd will go over some of those mistakes, what we learnt and how we've implemented those learnings.
Creating a Catalog Management System is non-trivial when you talk of scale. Scale which operates at all levels: size of catalog, volume of data, and the multi-dimensional flux of changes associated with the data. Besides, an ideal model to represent the relationships and the elasticity of data is a non-trivial science.
In this talk Utkarsh will with you try and figure out what part of it is Science, and where we cross boundary and think Art.
One of the challenges in distributed systems is to ensure transactional integrity without compromising on performance and scale. We will talk about how we solved these problems when building Flipkart's Supply Chain systems.
We will cover the alternatives considered, including distributed transactions, compensating transactions and asynchronous messaging and go into some detail about the approach we chose and the technology we created to solve a large class of problems when dealing with such systems.
Motivation for supporting guest users and what is the scale difference guest users create for session management system (from 4 million to 60 million sessions). Security issues around session hijack, replay attacks, spamming, dos attacks and downstream impact in supporting guest. Session Management Scheme options Pros and Cons in terms of security and business needs. Design of session data store for services like cart and recommendations.
Messaging is not just queues. Messaging is not just notifications. Not a tussle between push and pull. We have not thought about messaging, really! Messaging can be a platform. We will discuss precisely about that.
We try to delve at the complexities in storage system requirements and design. When storage requirements range from storing millions of files a few KBs in size to each being multiple GBs, when I/O requirements range from hundreds of parallel writes to thousands of parallel reads; can there be a storage system that can handle all storage issues and live to tell the story?
You say, "How hard is it to generate traffic?", I'll say "Super easy!". Then you'll say "How hard is it to generate relevant traffic?" and I'll say "Super tricky". Understand the challenges that come with doing online marketing at Flipkart scale. What we have achieved so far and what lies ahead of us.
How can you divide hardware into multiple small usable fractions without much overhead? How do you create fully functional VMs with the click of a button? How to enable both ops and dev to add more capacity with minimal friction? This talk focuses on technological innovations which enable Flipkart's "Kloud" infrastructure to scale up to 1500 servers that work together to serve over 30,000 orders a day.
Keeping a large scale commercial website like Flipkart secure is non-trivial. With large scale, comes large number of eyes and some of those are prying eyes who want to break a few locks. With Flipkart.com as the case study let’s take a sneak peek into the way we handle web-security. This session will take you through the different kinds of security threats, vulnerabilities and how we solve them.
First approach to any fraud detection system is generally based on hand crafted rules. But as the time progresses fraud pattern can change and hence rules needs to be reviewed and updated. We wanted our system to automatically adapt to changes. And hence we built Frank (named after Frank Abangale Jr., greatest con turned fraud preventer), a system which not only learns rules but also keeps updating them continuously. The system gets the intended behaviour by leveraging the idea of individual records acting as social entity and collaborating to find fraud among themselves.
We are always short on time. There is way too much to deliver with too few hands on the table and very important things at stake. This talk goes into simple things engineers can do at team level and other tricks that are easy to do at individual level that help you make most of your time.
The fallacies of distributed computing are a convincing motivator to plan and test for fault tolerance and failures.
DiFIT is an http endpoint with provides interfaces with simple abstractions to manipulate system/application states in a distributed environment. One can use DiFIT to control operations on System Under Test and enable developers/testers to inject faults and write test oracles to validate recovery and remedial measures in an automated fashion.