ngosangns Knowledge Base

#mlops

  • DoorDash Feature Store Redis Optimization
    How DoorDash benchmarked key-value stores and tuned Redis with hashes, xxHash, protobuf, and Snappy to cut its ML feature store's cost, memory, CPU, and latency.
  • Queuing Theory In Model Serving
    A debugging story showing how modeling replayed traffic as a Poisson process, rather than fixed intervals, explained a 2x p99 latency gap in GPU model serving.
  • Monitoring Tracking
    Explains how to use the t-test to decide whether one machine learning model's accuracy is statistically better than another's.
  • Kubernetes
    Kubernetes reading list and tools, including Tilt and the Model PVC Provisioning (MPP) approach for shipping large AI models to K8s.