Skip to main content

From SIGSEGV to Systems Research — My MSCS Statement of Purpose, Annotated

· 9 min read
Shreyas Atre

In late 2022, I was a third-year Electronics & Telecommunications student at VJTI Mumbai, writing statements of purpose for MSCS programs in systems and high-performance computing. This is the SoP I used to apply to Georgia Tech. I ended up at Louisiana State University with Prof. Hartmut Kaiser — the same person who mentored me during GSoC 2022 on HPX — and it turned out to be the best possible outcome.

I'm sharing this because I'm still confident this was a banger opening for a systems SoP. It leads with a real segfault, not a childhood dream. Sadly, none of the "top" programs bit — probably because the first person to read it wasn't technical and decided a second-year's debugging story wasn't what they were looking for. By the time it would have reached faculty who'd actually appreciate an ASan reference in paragraph one, it was already in the reject pile. That's how admissions works sometimes.

Below is the full text, broken into sections with short annotations on what I was trying to do with each part.

The Hook — Opening With a Real Problem​

[Why this opening]

Instead of "Ever since I was young..." I opened with a concrete debugging story. Systems faculty read hundreds of SoPs — leading with a real technical problem (segfault → GDB → ASan) signals you actually write code.

The C++ code I was working on displayed a segmentation fault. In most cases, when using GDB for debugging, I was able to zero in on the faulty line and fix it. Eventually though, I came across Address Sanitizer, which not only serves as a modification for "malloc" and "free," but also does a better job at managing poisoned memory and producing human-readable results. The mechanism behind it all seemed intriguing to me. Google's Summer of Code (2021) was my first experience with the systems domain, and I gained valuable knowledge on address sanitizers and other methods of dealing with undefined behavior. In addition to these runtime aspects, I gained experience with C++ generic programming techniques like "tag dispatching," "tag invoke," "SFINAE," and "type traits" throughout the Google Summer of Code (2022). Having completed my first C++ "type trait," I felt an indescribable sense of accomplishment. To be honest, I had no clue that pursuing this passion would lead me to a career in academia.

Open Source as Research Training — BeagleBoard, WasmEdge, LLVM​

[Framing OSS as systems work]

GSoC contributions can read as "volunteer coding" if you're not careful. The goal here was to show a progression: userspace IOCTL calls → GPU compute on SBCs → LLVM/sandboxing/virtualization → getting sponsored by a company (WasmEdge/Second State). Each step built on the last.

While participating in Google Summer of Code (2021) as a second-year student, I gained my first glimpse of the professional world. After gaining some understanding of userspace IOCTL calls and Linux's interprocess communication, I immediately set out to find ways to boost performance. BeagleBoard.org's expensive SBCs fascinated me because of the tiny GPUs they included. Those, rather than a regular CPU, would be ideal for the embarrassingly parallel algorithms. As someone who has always been curious about things like virtualization and the future of WebAssembly, this study was right up my alley. WasmEdge was where I first experienced LLVM, sandboxing, and virtualization, all of which I've since wanted to experience for their efficacy. In the end, WasmEdge sponsored me because of my contributions. Dr. Michael Yuan (CEO, Second State Inc.) acknowledged my work on the WasmEdge subsystem and approved my suggestion for an efficient Python SDK.

Standards-Level Impact — HPX, P2300, and Eric Niebler​

[Demonstrating standards-level work]

This section needed to show I wasn't just using C++ libraries — I was working at the level of the ISO C++ standardization process. Mentioning Eric Niebler's acknowledgment and the accepted fix in the P2300 reference implementation was important for credibility, not name-dropping.

The success I had with Google Summer of Code the prior year inspired me to take part again the following summer. One of my objectives for the year 2022 was to become adept in high-performance computing, something I had previously considered. To get started, I'll go over my work on the HPX runtime system. I focused on the proposals for concurrency and parallelism in the C++ standards, namely "P2300," which is an independent design for a Standard C++ framework for handling asynchronous execution in generic execution contexts. One of the fascinating integrations necessary for the HPX runtime system was one I began working on during the contribution period. Eric Niebler (Nvidia, Member of the ISO C++ Standardization Committee) even acknowledged me for pointing out an issue in the official C++ implementation in Paper P2300 and accepting my proposed solution. My contributions to Nvidia's "stdexec" repository are publicly available. These accomplishments qualified me for my ongoing internship with Professor Dr. Hartmut Kaiser at Louisiana State University, and I am prepared to implement my development and skills to be used in a related discipline. Given that further advancements in semiconductor technology are likely to be marginal at best, I plan to use my knowledge of the sender and receiver architectures to enhance the existing software stack.

Quantifiable Research — IIT-KGP and the National Supercomputing Mission​

[One number does more work than three paragraphs]

The "20% runtime memory reduction vs TensorFlow" result is the single strongest line in this SoP. Admissions committees skim — a concrete, comparative metric catches their eye immediately.

When I was given the opportunity to work with Professor Dr. Soumyajit Dey at IIT-KGP on designing a co-scheduling technique for heterogeneous GPGPU systems as part of the National Supercomputing Mission, I knew I had found my vocation in high-performance computing. As part of my programme, I was introduced to the rigorous scheduling procedures and policies used for large-scale programmes. Interestingly, I was tasked with creating a co-scheduling algorithm to optimise the runtime usage of large-scale deep learning pipelines. Despite my initial challenges, I was ultimately successful in understanding the SMT solver's static schedule. When compared to the state-of-the-art methods used by popular frameworks like TensorFlow, I was able to cut runtime memory utilisation by up to 20%.

Addressing the Major Mismatch — E&TC at VJTI​

[The elephant in the room]

My degree says Electronics & Telecommunications. I was applying for CS. Rather than ignoring it, I addressed it head-on: strong GPA, top lab performance, relevant coursework, and a capstone (FPGA-based deepfake detection) that bridges both fields.

My foundational knowledge was honed at Veermata Jijabai Technological Institute (VJTI), Mumbai, one of the oldest colleges of engineering in Asia and home to a rigorous undergraduate engineering curriculum. I graduated sixth semester with a cumulative grade point average of 8.28 out of 10 points. I have maintained this GPA while also devoting significant time to the extracurricular activities listed above. My actual major is in electronics and telecommunications, but I've really enjoyed the few computer science classes I've had to take as part of that. Because of my dedication to and enthusiasm for a hands-on approach to education, I consistently rank in the upper percentile of my class in terms of lab performance. Under the guidance of Professor Dr. Faruk Kazi (former HoD of the Electrical Department and Dean of Research and Development, VJTI), I decided to focus my final capstone research as an undergraduate on deep-fake detection through emotional inconsistencies in the audio-video temporal domain utilising field programmable gate arrays (FPGAs). I'm excited to contribute to its success because it fits precisely into my areas of expertise, which include computer architecture and organisation, specifically in the context of optimising pipelines. My knowledge of related fields, namely AI, will also prove useful in the completion of this task.

Tailoring to Georgia Tech — Courses, Labs, Professors​

[Specificity is non-negotiable]

Generic praise ("Georgia Tech is a world-class institution") wastes space. I named exact courses (CS 6210, CS 6241, CSE 6220), exact labs (Habanero, HPC Garage), and exact professors (Vivek Sarkar, Rich Vuduc) — and explained why their work connects to mine.

My motivation for pursuing a Master of Science degree in computer science is a desire to learn more about systems and make more significant contributions to the area through my studies and work. For this reason, I'd like to make an impact in the fields of operating system and compiler research, where my efforts will contribute to performance gains even on larger scales. My primary research interests lie in the field of high-speed computing. The courses at the Georgia Institute of Technology are highly applicable to my academic and career goals, and my research interests are a good fit with those of some of the faculty members there. Courses like "CS 6210 Advanced Operating Systems," "CS 6241 Compiler Design," and "CSE 6220 High Performance Computing" cater to my interests in the area of computing, which spans many sub-fields, including Computing Systems and High Performance Computing. Taking these classes would help me learn new things and hone the abilities I already have. Beyond these classes, I'm interested in Professor Dr. Vivek Sarkar's research at the Habanero Extreme Scale Software Research Laboratory. He studies software for high-performance computers, such as compilers, runtime environments, and debugging and verification tools. My past work and interests are a reflection of my dedication to the field of high-performance computing, which coincides nicely with his. Professor Dr. Rich Vuduc, at The HPC Garage, is another researcher whose work I find intriguing. His scientific interests are likewise a fantastic match for my own.

Therefore, I intend to pursue a Master's programme with a systems specialisation and subsequent PhDs and research in academia as my primary focus in order to continue my pursuit to make efficient programs and to defend against the end of performance gains through hardware by leveraging the efficient use of the current systems through performant abstractions. The education I would receive at the Georgia Institute of Technology would be unparalleled compared to any other graduate institution. I am confident that my dedication to my research, enthusiasm for my studies, and organisational skills will make me an asset to the university.


​