Siva Sudhir, Pooja Nilangekar, Bohan Zhang, and Aaron Tian will present their capstone projects. Bohan: OtterTune is really coming:  how to use OtterTune to tune your DBMS automatically Aaron: Fast Durability and Recovery in In-memory Databases Siva: Compilation of User-Defined Functions in Peloton Read More

For partitioned tables, maintaining good clustering properties for frequently accessed dimensions is critical for partition pruning performance. Naive methods of clustering maintenance could be expensive, especially when the clustering dimensions are different from the dimensions with which the data is loaded. On the other hand, approximate clustering is cheaper to maintain while still resulting in good pruning performance. In this... Read More

Current architectures for main-memory online transaction processing (OLTP) database management systems (DBMS) typically use random scheduling to assign transactions to threads. This approach achieves uniform load across threads but it ignores the likelihood of conflicts between transactions. If the DBMS could estimate the potential for transaction conflict and then intelligently schedule transactions to avoid conflicts, then the system could improve... Read More

PipeDream is a new distributed training system for deep neural networks (DNNs) that partitions ranges of DNN layers among machines, and aggressively pipelines computation and communication. Today’s pervasive use of data-parallel training performs well for DNNs of up to 10–20 million model parameters, but inter-machine communication dominates for models that are even 10x larger (e.g., up to 85% of time... Read More

In this seminar - Alok will present Striim, a distributed Streaming platform, and talk about the platform's motivation, distributed architecture, use cases, and open challenges. Read More

In the beginning there was a DBMS, a flexible piece of software that could be used for OLTP and OLAP workloads. When transaction throughput increased and data sizes grew the database needed to be split into two, each instance optimized for a particular workload. And so it has been ever since and the distance between the two systems has increased,... Read More

A statistical model is only as good as its training data. Systematic errors can arise when data are integrated from untrustworthy sources, collected in mixed formats, or contain inconsistent references of the same real-world entities. This talk describes the classical relational database topic of "data cleaning", i.e., the process of transforming the data to remove such issues, from a modern... Read More

Analytic workloads in data management systems are dominated by joins, aggregations, scan and filtering costs. In-Memory columnar databases have significantly optimized scans using compressed data formats and SIMD vectorization techniques, but have made little impact to the rest of the query execution plan. The Oracle Database In-Memory (DBIM) Option introduced new SQL execution operators that accelerate a wide range of... Read More

Read More

High-resolution, continuously-recording, and nearly-ubiquitous cameras provide great value for retrospective video analysis tasks in areas such as crime investigations and scientific research. This kind of video analysis task is often both interactive and exploratory in nature, where multiple queries are tried, aborted, refined, and re-executed in an iterative fashion. This exploratory video analysis usage model presents some unique challenges and... Read More