Hyperloop

Solving performance issues for your web application

What is Hyperloop?

Hyperloop is an ongoing research project that aims at detecting and solving performance problems for web applications built with ORM (Object-Relational Mapping) frameworks, mainly Ruby on Rails. Currently it contains:

  • A comprehensive study on existing open-source applications built with Ruby-on-Rails (with a replication package of these applications).
  • Instructions and benchmarks for reproducing the common performance inefficiencies found in our study.
  • PowerStation, a RubyMine plugin to identify performance bugs in Ruby on Rails applications.
  • Panorama, a view-centric optimizer for Ruby on Rails applications.

We are working on providing better solutions to solve more inefficiencies and more handy tools for application developers.

Project Goals

From banking to online shopping, data-driven applications that comprise of a database management system (DBMS) and an application that interacts with the DBMS are pervasive in our daily lives and the scientific community. Needless to say, the performance of data-driven applications is crucial. Unfortunately, due to the quickly increasing data size and ever-changing operational environment of the data center that host such applications, many data-driven applications suffer from performance problems. Techniques and tools that can help identify existing performance problems, predict future performance problems that will arise with workload and operational environment changes, and fix these problems in data-driven applications are highly desired in the current age of big data.

Unfortunately, although previous work in programming systems and database research communities have studied performance problems and query execution-time prediction, they are insufficient to address the performance problems that arise in data-driven applications. For instance, prior work in programming systems focus on optimizing applications, and treat the DBMS as a black box with unknown performance characteristics. Meanwhile, research in the database community focuses on analyzing query performance, and is oblivious to the fact that many queries are programmatically generated by applications with predictable structures.

Research Challenges

In this project, we argue that such “black box” approach forgoes many opportunities in identifying, and fixing performance problems in data-driven applications. Our key insight is to regard the DBMS and the application as white boxes and devise new opportunities to optimize both systems simultaneously. First, we plan to use various program analysis techniques to understand the semantics of data-driven applications. We will devise new program analysis techniques to study the structure and sources of parameter values for the queries that are issued by the application, the query plans that the DBMS use to process such queries, and how the query results are subsequently used by the application.

Next, we will use such information to identify performance problems in data-driven applications. Leveraging the end-to-end performance modeling of the DBMS and the application in data-driven applications, we will design new program analysis algorithms, performance testing techniques, and performance debugging tools to identify workload and code fragments in the application that might cause performance problems.

We will evaluate our techniques using real-world data-driven applications such as web applications that store persistent data in DBMSs.

Broader Impacts

The project addresses one of the most important and fundamental challenges of the big data era, namely the performance of data-driven applications. It will broadly benefit other science disciplines and the society by improving the performance of data-driven applications. We will leverage their expertise in in database systems, program analysis, performance debugging, and software engineering to tackle the challenges. Furthermore, we have extensive connections with researchers and industry practitioners at Intel, Huawei, Google, and Microsoft, and will make use of such connections to validate their research findings.

We plan to release all software and benchmarks that are developed to public domain. The techniques developed in this project will be incorporated into graduate-level and undergraduate-level courses in data management systems and programming systems offered at the UC Berkeley.

Publications

News

Acknowledgement

This material is based upon work supported by the National Science Foundation under Grant No. IIS-1546083 (2016-2019) and No.IIS-1546543 (2016-2019), "Holistic Optimization of Data-Driven Applications," and is a collaborative effort between UC Berkeley and the University of Chicago.

Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Last updated: Mar 2020