Companies run A/B tests to accelerate innovation and make informed data-driven decisions. At Microsoft alone, over twenty thousand A/B tests are ran each year helping decide which features maximize user value. Not all teams and companies succeed in establishing and growing their A/B testing programs. In this paper, we explore multiple-case studies at Microsoft, Outreach, Booking.com, and empirical data collected, and share our learnings for iteratively adopting and growing A/B testing. The main contribution of this paper is the A/B Testing Flywheel. This conceptual model illustrates iteratively navigating the value-investment cycle with the goal to scale A/B testing. In every turn of the flywheel, teams need to invest in order to increase the A/B testing momentum. We describe the investments in software development processes that have been advantageous in getting the flywheel to turn. We also share example metrics that track the progress towards sustainable A/B testing momentum.

Source and object selection and retrieval from large multi-source data sets are fundamental operations in many applications. In this paper, we initiate research on efficient source (e.g., database) and object selection algorithms on large multi-source data sets. Specifically, in order to acquire a specified number of satisfying objects with minimum cost over multiple databases, the query engine needs to determine the access overhead for individual data sources, the overhead of retrieving objects from each source, and possibly other statistics such as estimating the frequency of finding a satisfying object in order to determine how many objects to retrieve from each data source. We adopt a probabilistic approach to source selection utilizing a cost structure and a dynamic programming model for computing the optimal number of objects to retrieve from each data source. Such a structure can be a valuable asset where there is a monetary or time related cost associated with accessing large distributed databases. We present a thorough experimental evaluation to validate our techniques using real-world data sets.

As the number and size of large timestamped collections (e.g. sequences of digitized newspapers, periodicals, blogs) increase, the problem of efficiently indexing and searching such data becomes more important. Term burstiness has been extensively researched as a mechanism to address event detection in the context of such collections. In this paper, we explore how burstiness information can be further utilized to enhance the search process. We present a novel approach to model the burstiness of a term, using discrepancy theory concepts. This allows us to build a parameter-free, linear-time approach to identify the time intervals of maximum burstiness for a given term. Finally, we describe the first burstiness-driven search framework and thoroughly evaluate our approach in the context of different scenarios.

Satisfying energy constraints while meeting performance requirements is a primary concern when a sensor network is being deployed. Many recent proposed techniques offer error bounding solutions for aggregate approximation but cannot guarantee energy spending. Inversely, our goal is to bound the energy consumption while minimizing the approximation error. In this paper, we propose an online algorithm, region sampling, for computing approximate aggregates while satisfying a pre-defined energy budget. Our algorithm is distinguished by segmenting a sensor network into partitions of non-overlapping regions and performing sampling and local aggregation for each region. The sampling energy cost rate and sampling statistics are collected and analyzed to predict the optimal sampling plan. Comprehensive experiments on real-world data sets indicate that our approach is at a minimum of 10% more accurate compared with the previously proposed solutions.

Performing data-mining tasks such as clustering, classification, and prediction on large datasets is an arduous task and, many times, it is an infeasible task given current hardware limitations. The distributed nature of peer-to-peer databases further complicates this issue by introducing an access overhead cost in addition to the cost of sending individual tuples over the network. We propose a two-level sampling approach focusing on peer-to-peer databases for maximizing sample quality given a user-defined communication budget. Given that individual peers may have varying cardinality we propose an algorithm for determining the optimal sample rate (the percentage of tuples to sample from a peer) for each peer. We do this by analyzing the variance of individual peers, ultimately minimizing the total variance of the entire sample. By performing local optimization of individual peer sample rates we maximize approximation accuracy of the samples. We also offer several techniques for sampling in peer-to-peer databases given various amounts of known and unknown information about the network and its peers.

The ability to provide reliable in-network storage while balancing the energy consumption of individual sensors is a primary concern when deploying a sensor network. The main concern with data-centric storage in sensor networks is the ability to provide reliable and load balanced storage. Energy and wireless range constraints make centralized approaches for storage impractical, and in-network data-centric solutions can be used to reduce the number of messages sent over the network. However, these solutions quickly become expensive when combined with fault- tolerance, load balancing and routing. In this paper, we present a novel data-centric storage and query routing mechanism for sensor networks. The routing mechanism is constructed upon the neighborhood information of individual sensors and is completely independent of geographical information. Our data resilient algorithm is capable of recovering from multiple simultaneous failures in the network while adaptively adjusting the load distribution of the newly generated sensor data. Comprehensive experiments on both real-world and synthetic data sets indicate that our approach is more effective and efficient than the previously proposed solutions.

Peer-to-peer (P2P) databases are becoming prevalent on the Internet for distribution and sharing of documents, applications, and other digital media. The problem of answering large-scale ad hoc analysis queries, for example, aggregation queries, on these databases poses unique challenges. Exact solutions can be time consuming and difficult to implement, given the distributed and dynamic nature of P2P databases. In this paper, we present novel sampling-based techniques for approximate answering of ad hoc aggregation queries in such databases. Computing a high-quality random sample of the database efficiently in the P2P environment is complicated due to several factors: the data is distributed (usually in uneven quantities) across many peers, within each peer, the data is often highly correlated, and, moreover, even collecting a random sample of the peers is difficult to accomplish. To counter these problems, we have developed an adaptive two-phase sampling approach based on random walks of the P2P graph, as well as block-level sampling techniques. We present extensive experimental evaluations to demonstrate the feasibility of our proposed solution.

Peer-to-peer databases are becoming prevalent on the Internet for distribution and sharing of documents, applications, and other digital media. The problem of answering large scale, ad-hoc analysis queries ― e.g., aggregation queries ― on these databases poses unique challenges. Exact solutions can be time consuming and difficult to implement given the distributed and dynamic nature of peer-to-peer databases. In this paper we present novel sampling-based techniques for approximate answering of ad-hoc aggregation queries in such databases. Computing a high-quality random sample of the database efficiently in the P2P environment is complicated due to several factors ― the data is distributed (usually in uneven quantities) across many peers, within each peer the data is often highly correlated, and moreover, even collecting a random sample of the peers is difficult to accomplish. To counter these problems, we have developed an adaptive two-phase sampling approach, based on random walks of the P2P graph as well as block-level sampling techniques. We present extensive experimental evaluations to demonstrate the feasibility of our proposed solutio