PhD student and Research Fellow , School of Computer Science
Email: ssinghal74@gatech.edu, Github: @singhalshubh
LinkedIn: https://www.linkedin.com/in/shubhendra-singhal/
Google Scholar: @singhalshubh
Lab: Habanero Extreme Scale Software Research Laboratory
Hi everyone! I am currently first year PhD student at Georgia Tech specializing in Distributed Computing and High-Performance Computing for hardware-software co-design. My advisor is Dr. Vivek Sarkar. Please find my CV [here].
I primarily specialize in run-time systems and Exascale standards of programming. My focus at the current time is to improve shared memory libraries such as MPI, OpenSHMEM used for communication in a data center. One of my recent findings on Cori (commodity supercomputer) indicated that libraries in future using remote (i.e. not local to machine) shared memory can construct a scalable solution with 4x relative performance gain in communication time by simply aggregating the signals and small-packeted data transfers. Currently OpenSHMEM routines associated with put signal are under-developed and require further considerations to improve. Refer to https://github.com/openshmem-org/specification/issues/500
I am currently working on improving Conveyors, which innovates a space efficient aggregation layer. Some of the key insights deal with placement of mpp_quiet operation which can cause significant overheads (upto 15% wait) for data-intensive applications such as Triangle Counting and Jaccard. Furthermore, loopback routes of a single PE, can involve upto 4 steps to pull a packet despite the packet being on the same PE. The problem arises due to the boundary of object which do not share their symmetric memory with other object on same PE. This leads to multiple aggregations and implicit yield creation, which are not required as per the requirement – one implication of system design of Conveyors.
In addition, I am also working towards better scalability of Influence Maximization problem.
Future Goals of PhD
In the long run, I would work towards introducing energy efficiency modeling in Exascale Computing.
Personal Interests
International Law and Politics, Snooker (Big fan of Ronnie Sullivan).