Developing new algorithms for MVAPICH library to accelerate HPC, AI, and Deep Learning frameworks (TensorFlow, PyTorch, Horovod) using modern HPC interconnects (InfiniBand, Omni-path, 10-100GigE/iWARP, and RoCE) and accelerators (DPUs, GPUs).

The secure communication of Container-based HPC Clusters is my focus in this study.Developing a container-based cluster using Docker or Singularity is a popular trend nowadays for HPC centers. Container orchestrations like Docker swarm and Kubernetes provide mechanisms to enhance inter-node data communication security. There are several Container Network Interfaces (CNI) for Kubernetes to handle network security with different encryption methods: Calico, Antrea, Cillium, etc. However, these encryption mechanisms impose a non-negligible performance penalty, which is not acceptable in MPI applications.In this research, I have designed and implemented a novel approach to incorporate encryption in MPI libraries (mainly MPICH and MVAPICH) to reduce encryption overhead. The empirical evaluation on multiple production systems shows that my proposed encrypting approach achieves 2 to 10 times improvement (depending on message size, number of nodes and number of processes per node) over Docker swarm and Kubernetes built-in security mechanisms.

o Developed Multicore Programming by Multiprocessor and GPU using CUDAo Researched on Video Processing Methods

o Developed Genetic algorithm on GPU using CUDA in Energy-Aware Systems Labo Developed Placement algorithm for Heterogeneous Reconfigurable Systems in Data Storage Systems & Network Lab