Winter School 2024 on Data Systems

Welcome

Registration

Logistics

Schedule

Program

[!NOTE] Prof. Sanjam Garg, UC Berkeley, an alumnus of IITD, will give a keynote address titled “Your research career”.

Welcome

The Winter School 2024 on Data Systems, organized by the Data Systems Group of the Department of Computer Science and Engineering at IIT Delhi, will be held from December 2nd to December 6th, 2024. Supported by the Mohit Aron Endowment, this winter school provides an exceptional opportunity for final-year undergraduate students, master’s and PhD students, and industry professionals to deepen their knowledge in scalable systems for Big Data, Data Science and AI.

Participants will engage in lectures and hands-on lab sessions on a range of data systems topics, led by Prof. Kaustubh Beedkar and Prof. Abhilash Jindal.

Registration

Applications are open to final-year undergraduate students, master’s and PhD students, and industry professionals with an interest in scalable systems for Big Data, Data Science, and AI. The application process for Winter School 2024 consists of two rounds:

1. First Round - Application Submission
   Interested candidates must submit their applications via the provided online form by 5:00 pm on 14th November 2024. Only candidates who complete this submission will be considered for the next round.

2. Second Round - Online Test
   Candidates who have successfully applied in the first round will be invited to participate in an online test scheduled for 16th November 2024 17th November 2024 at 10:00 AM.

Selected candidates will receive notification of their acceptance into the Winter School by 17th November 2024 21st November 2024.

Logistics

• Venue: Room 501, Bharti building, Department of Computer Science and Engineering, IIT Delhi.
• Dates: December 2nd – December 6th, 2024
• Accommodation: Hostel accommodation including breakfast, lunch, dinner, and high tea will be provided from Dec 1st – Dec 7th, 2024. In addition, a stipend of 2000 INR will be provided to selected students. Participants will be required to check in on December 1st, 2024, and check out before noon on December 7th, 2024.
• After winter school: Participants who successfully complete the Winter School will receive a Certificate of Participation. Please note that we will NOT provide recommendation letters for graduate school applications. Students will be encouraged to apply for paid internships at IIT Delhi during the 8 weeks of summer 2025.

Daily Schedule

• 09:00 am – 10:15 am: Lecture
• 10:15 am – 10:45 am: High tea + offline discussions
• 10:45 am – 01:00 pm: Lectures with breaks
• 01:00 pm – 02:00 pm: Lunch Break
• 02:00 pm – 05:00 pm: Lab
• 03:30 pm – 04:00 pm: High tea

Program

The Winter School will cover the following topics:

Day 1: Distributed data processing

Lecture:

• Holy grail: transparently scale, and tolerate faults/stragglers;
• Difficulties in transparent locality, scalability, fault tolerance, and straggler mitigation;
• Introduction to checkpointing and replication;
• Introduction to distributed data processing with MapReduce;
• Main ideas: functional programming model; separate control and data plane; locality optimizations, re-execute lost/backup tasks, deterministic/idempotent tasks.

Lab: Write a fault tolerant computation from scratch.

Resources:

1. Distributed Shared Memory: Notes and paper.
2. MapReduce: Notes and paper

Day 2: Dataflow systems for batch processing

Keynote address by Prof. Sanjam Garg, UC Berkeley titled “Your research career”.

Abstract: In this talk, I will share my experiences and the lessons I have learned. I hope that they will help you in your research career.

Bio: Prof. Sanjam Garg is an Associate Professor at the University of California, Berkeley. His research interests are in cryptography and its applications to security and privacy. He obtained his Ph.D. from the University of California, Los Angeles in 2013 and his undergraduate degree from the IIT, Delhi in 2008. Prof. Garg is the recipient of various honors such as the ACM Doctoral Dissertation Award, the Sloan Research Fellowship and the IIT Delhi Graduates of the Last Decade Award. Prof. Garg’s research has been recognized with a test of time award at FOCS 2023, and best paper awards at EUROCRYPT 2013, CRYPTO 2017 and EUROCRYPT 2018.

Lectures:

• Spark’s Resilient distributed dataset (RDD) abstraction: write-once for consistent replication, coarse-grained transformations to reduce lineage, lineage-based re-execution of lost/backup tasks;
• Abstractions beyond MapReduce, lazy execution, stage planning, narrow and wide-dependencies, query optimizations.

Lab: Hands-on exercises with Spark.

Resources:

1. Spark: Notes and paper.

Day 3: Dataflow systems for stream processing

Lecture:

• Semantics of stream processing: unbounded streams, event time vs processing time, hopping/sliding/session windows, watermarks;
• Streaming query optimizations and continuous operator model of Flink for low-latency;
• Fault tolerance: problems due to state in continuous operators, pull state out of operators by discretizing streams, consistent checkpoints in Flink using Chandy-Lamport algorithm.

Lab: Hands-on exercises with Flink. Setup: Installing Pyflink and lab.

Resources:

1. Spark streaming: Notes and paper.
2. Flink: Notes and paper.

Day 4: Cyclic dataflow systems

Lecture:

• Iterative data processing: solving graph problems with iterative data processing, driver-based iterations vs native iterations with cyclic dataflows, changes for creating consistent checkpoints.
• Introduction to ML workflows: error handling and correction as a first class citizen, Reactive dataflow with deletes, edits, and appends.
• Backward and forward lineage for backward tracing and delete propagation, stage planning with incremental/non-incremental, monotonic/non-monotonic operators

Lab: Hands-on exercises with Popper.

Resources:

1. Popper: Paper.

Day 5: Cross-platform data processing

Lecture:

• Introduction to Wayang: a platform for cross-framework data processing, allowing optimizations across Spark, Flink, and Relational Databases.

Career session:

• Data systems research at IITD, guiding participants interested in pursuing higher studies, and research success stories.

Disclaimer: The schedule and topics are only tentative. They are subject to change depending upon attendee interests.