Modi, ChiragChiragModiKulkarni, Sameer G.Sameer G.Kulkarni2025-08-312025-08-312024-01-0110.1109/HiPCW63042.2024.000712-s2.0-105001539004https://d8.irins.org/handle/IITG2025/28465Communication-intensive high-performance computing applications demand low latency and very high network throughput. In order to meet these needs, the hardware capabilities of modern commodity servers have been steadily advancing. Moreover, the network bandwidth has significantly increased (i.e., 400-100GbE) to outpace the processor speed. Traditionally, applications use socket API and rely on the kernel for network communication. However, it incurs high overheads due to resource contention, context switch and packet copy operations, which make the network stack a bottleneck. Notably, while the packet I/O have scaled up to tens of millions of packets per second (pps), the TCP processing is reported at 0.3 million transactions per second [1].falseConference Paper27700135187-18820240cpJournal0