asethia

Ankit Sethia

Dept. of Computer Science

University of Michigan

2260 Hayward Street

Ann Arbor, MI - 48109

asethia (at) umich (dot) edu

Welcome to my webpage

I am a Research Fellow in the Department of Computer Science at the University of Michigan. I got my Bachelor's degree in E & ECE from IIT Kharagpur in 2007. After my bachelor's degree I worked with Mentor Graphics for two years. I joined University of Michigan in Fall of 2009 as a Master's student. By the end of Fall 2009, I was really interested in computer architecture. So I started working with Prof. Scott Mahlke and changed from Master's program to a PhD program. I finished my PhD under my advisor, Prof. Scott Mahlke in the Compilers Creating Custom Processors(CCCP) research group. Currently I am a Research Fellow at the same department.

Research

I am broadly interested in Computer Architecture with focus on high performance computing. The words GPU, multi-threaded and SIMD always catch my attention. I have worked on the following projects in reverse chronological order:

1) Mascar: A warp scheduler that improves overlap of computation and memory accesses in GPUs for memory intensive cases. It also improves the reuse of data in cache during periods of saturation of the memory system.

2) Equalizer: A runtime system that observes the requirements of the kernels executing on the GPU and modulates certain hardware parameters to match the requirements of the kernel.

3) APOGEE: A dynamic prefetcher that adapts to the memory access patterns and therefore reduces memory access latency for both graphics and non-graphics applications on GPUs.

4) PEPSC: An energy-efficient processor for scientific computing. The focus of this project was on looking at the reasons for energy-inefficiency on GPUs and come up with architectural solutions to overcome them.

5) MEDICS: a low power throughput engine for medical imaging. In this project i worked on creating verilog model of 16 bit and 32 bit fused ALUs.

Publications

Dynamic Hardware Resource Management for Efficient Throughput Processing
Ankit Sethia
PhD Dissertation

Mascar: Speeding up GPU Warps by Reducing Memory Pitstops
Ankit Sethia, D. Anoushe Jamshidi and Scott Mahlke
Feb. 2015, HPCA 2015

Equalizer: Dynamic Tuning of GPU Resources for Efficient Execution
(Best paper nominee)
Ankit Sethia and Scott Mahlke
Proc. 47th Annual IEEE/ACM International Symposium on Microarchitecture (MICRO) Dec. 2014.

APOGEE: Adaptive Prefetching On GPUs for Energy Efficiency
Ankit Sethia, Ganesh Dasika, Mehrzad Samadi and Scott Mahlke
Proc. 22nd Intl. Conference on Parallel Architectures and Compilation Techniques (PACT) Sep. 2013.

A Customized Processor for Energy Efficient Scientific Computing
Ankit Sethia, Ganesh Dasika, Trevor Mudge and Scott Mahlke
IEEE Transactions on Computers Vol. 61, No. 12, Dec. 2012

PEPSC: A Power Efficient Processor for Scientific Computating
Ganesh Dasika, Ankit Sethia, Trevor Mudge and Scott Mahlke
Proc. 20th Intl. Conference on Parallel Architectures and Compilation Techniques (PACT) Oct. 2011.

MEDICS: Ultra-Portable Processing for Medical Image Reconstruction
Ganesh Dasika, Ankit Sethia, Vincentius Robby, Trevor Mudge, and Scott Mahlke
Proc. 19th Intl. Conference on Parallel Architectures and Compilation Techniques (PACT) Sep. 2010.

Resume

Here is my resume in pdf format.

Personal

I am from Bhubaneswar, a city in the south-eastern part of India. I come from a community in which almost everyone has his own business for livelihood. I am still picked at my family gatherings for not having my own business and working for a salary instead.
Other than that i was active during my undergraduate years with various sports competitions held at IIT Kharagpur. It is also the place that cemented my love for Naruto, Cricket and bhaat.

Profile

To overcome procastination, I have channeled my procastination to profile myself. This plot shows an interactive graph of how i spend my time on my lab computer. The entire process is automated and a true reflection of how hard/hardly i work. There are three time-series:

1) google-chrome: It show how much time i spend browsing the internet and checking my emails.
2) terminator: This is the terminal application i use. The value on this series is the time i spend working.
3) others: Sometimes i also use other applications. All others are bundled into it.