Research Assistant

University of Pennsylvania, United States of America

1. Working with Dr. Boon Thau Loo with Perspecta Labs on CoBRe .
2. Working towards Master's Thesis with the latest update of integration of loss, delay and restart resilience to the system. 3. Stay tuned for latest updates here!

Research Software Developer

OCaml Labs, UK , Indian Institute of Technology Madras, India

1. Developed graph500 benchmark in OCaml to address three application kernels - Large-Scale Graph Generation : Kronecker Graphs, Optimization : Single Source Shortest Path, and Concurrent Search : Breadth First Search

2. Designing Hashmap, CSR for sparse matrix, various lockfree structures for handling concurrent access and other performance choices, led to a significant speedup of 20x in concurrent search kernel, 4x in Kronecker graphs and an over all 90.93% improvement in execution time of the sequential version of the kernels across the application on Intel(R) Xeon(R) Gold 5120 CPU @ 2.20GHz(24 core).

3. Responsible for fixing the benchmark related GitHub issues on sandmark.

Research Assistant

International Institute of Information Technology, Hyderabad, India

1. Developed a uni-processor thermal-aware data freshness analysis framework considering the 2-task and N-task scenario and the EDF schedulability with no thermal violation.

2. Designed a formal proof for obtaining the maximum staleness factor under various freshness and thermal constraints. Developed a simulator in C++ for demonstrating the correctness of the proof and obtaining the variations(graphs) of schedulability with CPU utilisation, freshness and temperature bounds.

3. Lead a team of 3, to ensure productive development and preliminary research.

Research Assistant

University Of Southern California, Los Angeles, California

1. Profiled a variant of SAT, MiniSAT using GPROF to analyse which component (observed propagate() function) accounts for most of the execution time.

2. Formulated a tool AccEx to profile a program according to the user defined input of execution time. Reduced the complexity of mapping every call’s address to its function name in GPROF by assigning a machine independent attribute - call id, for every call in a program to replace the default execution time with the user-defined input.

3. Observed a pattern of exponential decrease in total execution time for linear decrease in time of propagate() function allowed up-to 76% as, the other function’s time outstrips the linear decrease, nullifying the effect of parallelism.

Research Assistant : Publications

Indian Institute of Technology Madras, India

Singhal, Shubhendra Pal, Sridevi, M., Narayanan, N.S., & Raman, M. (2019), ”Porting of eChronos RTOS on RISC-V Architecture”, Springer Singapore, Lecture Notes in Electrical Engineering - Ad- vances in Communication and Computational Technology, Proceedings of ICACCT India, vol. 668, doi:10.1007/978-981-15-5341-7. Pages 1269-1279

Shubhendra Pal Singhal and M. Sridevi, ”Comparative study of performance of parallel alpha Beta Pruning for different architectures,” 2019 IEEE 9th International Conference on Advanced Comput- ing (IACC), Tiruchirappalli, India, pp. 115-119, doi: 10.1109/IACC48062.2019.8971591. Pages 115-119