Bhavana AI

AI/ML insights

Dev Log: January 18, 2026

Courses

Reviewed a research replication proposal focused on the Slicer sharding system, comparing it against a prior GPU scheduling proposal. The study takes a narrower, more focused approach: pure replication of a single algorithm using Kubernetes-native tooling rather than building custom Python schedulers.

Key differences from your GPU scheduling proposal:

  • This is a pure replication study (no extension)
  • Uses KWOK simulation instead of discrete-event simulators
  • Builds Kubernetes controllers in Go instead of Python schedulers
  • Simpler scope: one algorithm (Slicer) vs. combining multiple (Tiresias/Gavel/FGD)