Calendar

Dive into how modern systems escape the kernel networking stack for microsecond-scale performance: DPDK's poll-mode drivers, io_uring's submission rings, RDMA's one-sided operations, and the trade-offs each approach demands.

How moving compute to where the bits live rewrites the rules of memory-bound computation, from UPMEM's DRAM-scale PIM to Samsung's HBM-PIM and the programming model that still keeps us up at night.