Ryan Lefever Email & Phone Number

Mountain View, Ca, Us

* YouTube Trust and Safety: Designing and developing next generation architecture to detect violations of and enforce policy across all YouTube content (https://www.youtube.com/howyoutubeworks/policies/community-guidelines/). Detection involves balancing machine learning, human reviewers, and user content flagging. Recent focus has been on adaptive detection to meet the changing constraints of COVID-19. Continued focus is on designing clean and scalable data models and infrastructure APIs.* Bandwidth Enforcer: Bandwidth Enforcer is a planet-scale control system for allocating and enforcing bandwidth constraints (https://research.google/pubs/pub43838/). Designed and deployed methods and systems for reliable distribution and validation of a global configuration. Architected an impact analysis framework for evaluating scalability changes to a planet-scale control system. Designed and built a declarative framework for testing the correctness of Bandwidth Enforcer.* Sawzall: Sawzall is a procedural programming language, developed at Google for distributed data analysis (https://research.google/pubs/pub61/). Concentrated on design and development of Protocol Buffer compilation, in support of structured Big Data, as a first class citizen in the programming language.* Click Logs Infrastructure: Click logs are vital to the operation of most Google systems, notably Ads and Search Quality. Developed reliability models for a planet-scale data collection system, where data durability is the number one goal. The models informed our resourcing and reliability strategies, particularly in more constrained areas of the world. Designed and developed a scheduling system for coordinating and ensuring the reliability of large-scale data management operations, e.g., to meet data retention obligations or improve performance, according to data access patterns. Invented techniques and developed tools to analyze schema quality and data lineage at scale.

Advanced Technology Group* Analyzed the state of fault injection at NetApp and proposed a methodology for testing distributed systems, based on global state-based fault injection. Proposed strategies for automating the fault injection process, such as automatic derivation of software state models.* Collaborated on a tool to analyze file system workloads. Designed a framework for rapid development of analysis tools.

Champaign, Il, Us

Coordinated Science Laboratory* Diverse Partial Memory Replication: Developed and evaluated a tunable approach for automated software diversity called Diverse Partial Memory Replication (DPMR), aimed at detecting memory safety errors. DPMR is a compiler transform that replicates some subset of an executable’s data memory, and applies one or more diversity transformations to the replica. Implemented using the LLVM open-source compiler (http://llvm.org). Project comprises 55k+ lines of code.* Interprocedural Program Slicing: Extended the interprocedural program slicing (IPPS) algorithm by Horwitz, Reps, and Binkley (ACM TOPLAS 1990) to accommodate memory dependencies and memory aliasing. Given a variable and a code location, IPPS determines the set of locations that can affect the variable’s value. Implemented using LLVM. Project comprises 40k+ lines of code.* Funding Proposals: Key contributor on several funding proposals, including a Next Generation Software (NGS) proposal to the National Science Foundation. The NGS project was awarded $1.2 million and ran from 2004 to 2008. See more at http://www.perform.csl.illinois.edu/projects/newNSFNGS.html.* Fault Injection Study of Coda: Experimentally evaluated the impact of correlated network failures on the directory resolution protocol for the Coda distributed file system. When a Coda client loses network contact with all of its file servers, it caches local changes and resolves them with the server when network connectivity is restored. Experiments severed network connectivity during resolution, to evaluate the performance and dependability impact of such a correlated fault.* Loki Fault Injector: Co-developed Loki, a global state-based fault injector to validate and measure distributed systems. Challenges included tracking and verifying global state, minimizing intrusion in the system under study, estimating measures, and supporting arbitrary entry and exit of nodes from the system under study.

Champaign, Il, Us

Teaching assistant in fall semesters.* Teaching assistant for a graduate-level course on computer systems analysis. Course covered stochastic modeling, simulation, queueing theory, and experimental analysis. Conducted office hours, and taught class when instructor was unavailable. Graded homework assignments and exams. Provided technical support for student use of the Mobius modeling tool (http://www.mobius.illinois.edu).

Zurich, Ch

Foreign Exchange Information Technology Group* Designed, implemented, and tested Windows NT-based client software tool and a Unix-based server to manage a publish and subscribe messaging system. Employed CORBA and C++, and learned Java to create a tool that could configure, administer, and monitor the networking system. Environment was team- and deadline-oriented, and the ability to learn quickly was essential.

Chicago, Il, Us

Cellular Infrastructure Group* Developed and tested client/server processes in C++ and IDL for Service Node call processing to provide cellular services. Utilized CORBA in a distributed-object, multi-threaded architecture on a Unix platform. Teamwork, deadlines, and quick learning were highly emphasized.

Ph.D., Computer Engineering

M.S., Computer Engineering

B.S., Computer Engineering