Sundaram Ananthanarayanan

Software Engineer

Union City, California, United States15 yrs 9 mos experience

Most Likely To SwitchHighly Stable

Key Highlights

Led development of Mantis stream processing engine.
Innovated SubmitQueue for continuous integration at Uber.
Transitioned Mantis to Kubernetes, optimizing costs significantly.

Stackforce AI infers this person is a SaaS-focused software engineer with expertise in stream processing and infrastructure.

Contact

Skills

Core Skills

Software EngineeringStream ProcessingData EngineeringInfrastructureContinuous Integration

Other Skills

CC++PythonJavaSoftware DesignLinuxSoftware DevelopmentKubernetesMachine LearningApache MesosApache IcebergKafkaConflict AnalysisMesosVerilog

About

At Netflix, my role as a Staff Software Engineer finds me at the forefront of developing Mantis, our stream processing engine for operational needs, as well as Flink, a stream processing engine tailored for analytical use-cases. Previously, at Uber, I led the creation of SubmitQueue and uCI, systems that were created to provide an always green master for large monorepos.

Experience

15 yrs 9 mos

Total Experience

2 yrs 11 mos

Average Tenure

6 yrs 5 mos

Current Experience

Netflix

Staff Software Engineer

Dec 2019 – Present · 6 yrs 5 mos · Los Gatos, California, United States · On-site

I am one of the leads on the data platform team at Netflix, spearheading projects in the data processing space that are both deep and wide in scope.
Some of the projects I have been involved with:
◦ Mantis: Mantis is a stream processing engine developed at Netflix, designed to address the unique challenges posed by operational data. Mantis handles 50k+ jobs/day, processing 100 PBs of data on 30k+ containers. Over the past couple of years, I have been leading its development.
Transitioned Mantis from Apache Mesos to Kubernetes with an innovative architecture. Guided a team of senior engineers from prototype to production, year-long migration and open-sourcing.
Reduced annual compute costs by millions through ML-based container optimizations.
Championed Mantis adoption to other companies including Stripe.
◦ Flink: I have also contributed to the Flink ecosystem at Netflix - a stream processing engine for analytical needs.
Created a system enabling users to backfill Flink pipelines using Apache Iceberg without code changes,
eliminating the need to maintain separate batch jobs during outages.
Designed the system to mimic Kafka properties when reading from Data Lakes, ensuring effortless integration.
The system has been adopted by hundreds of pipelines within Netflix. Helped open-source the project, now utilized by other companies using Apache Iceberg.

CC++PythonJavaSoftware EngineeringSoftware Design+3

Uber

Senior Software Engineer

May 2016 – Oct 2021 · 5 yrs 5 mos · San Francisco Bay Area

Driving next generation design & architecture of 2 systems critical to Uber's Build Infrastructure.
SubmitQueue: Conceived & developed an innovative scheduling system that keeps large monorepos shippable at all times even with 1000s of commits/hr.
uCI: Creator of uCI, a generic, scalable task-execution/orchestration system that leverages mesos for powering CI & other stateful workloads at Uber.
Submit Queue:
100s of engineers committing changes concurrently to a repository leads to frequent master breakages. Explored & conceived a new system called "SubmitQueue" that guarantees an "always-green" master in this context. At Uber, Submit Queue handles 100s of commits/hr submitted by ~1000 of engineers every day.
Individual contributions made:
Came up the idea of speculatively executing builds that are mostly likely to succeed to scale the system to 100s of commits/hr.
Designed and implemented a conflict analyzer to detect if two changes are likely to conflict or not using the build system.
Built Machine Learning models that continuously learn & estimate the likelihood of changes to pass builds.
Architected the system to handle 100s of repositories, failover-active in all of Uber's regions.
Published a research paper at Eurosys'19 introducing SubmitQueue and detailing its core components.
uCI - a reliable task execution/orchestration system for stateful workloads:
Laid the foundation for a world class CI Infrastructure at Uber by replacing Jenkins with a scalable task execution system that leverages mesos for cluster management.