This course aims to teach learners how to unlock the power of inline floating-point operations on programmable switches. The learning outcomes include understanding the limitations of existing switch hardware for application acceleration and proposing solutions to address these limitations. The course covers implementing FPISA, a floating point representation, on programmable switches and evaluating its performance in applications like distributed training for machine learning and distributed query processing. The teaching method involves presenting research findings and proposing hardware changes to improve switch efficiency. This course is intended for individuals interested in network functionality, application acceleration, and programmable switch architectures.
Overview
Syllabus
NSDI '22 - Unlocking the Power of Inline Floating-Point Operations on Programmable Switches
Taught by
USENIX