• Publications
  • Influence
Above the Clouds: A Berkeley View of Cloud Computing
TLDR
Cloud Computing, the long-held dream of computing as a utility, has the potential to transform a large part of the IT industry, making software even more attractive as a service and shaping the way IT hardware is designed and purchased. Expand
  • 6,739
  • 367
  • PDF
Routing as a Service
TLDR
In Internet routing, there is a fundamental tussle between the end users who want control over the end-to-end paths and the Autonomous Systems (ASes) who want to control the flow of traffic through their infrastructure. Expand
  • 109
  • 5
  • PDF
Per Hop Behaviors Based on Dynamic Packet State
TLDR
This document proposes a family of Per Hop Behaviors (PHBs) based on Dynamic Packet State (DPS) in the context of the differentiated service architecture. Expand
  • 28
  • 3
Checkmate: Breaking the Memory Wall with Optimal Tensor Rematerialization
TLDR
We formalize the problem of trading-off DNN training time and memory requirements as the tensor rematerialization optimization problem, a generalization of prior checkpointing strategies. Expand
  • 19
  • 2
  • PDF
Lineage stash: fault tolerance off the critical path
TLDR
We propose the lineage stash, a decentralized causal logging technique that significantly reduces the runtime overhead of lineage-based approaches without impacting recovery efficiency. Expand
  • 6
  • 1
  • PDF
Blink: Fast and Generic Collectives for Distributed ML
TLDR
We propose Blink, a collective communication library that dynamically generates optimal communication primitives by packing spanning trees. Expand
  • 13
  • PDF
AutoPandas: neural-backed generators for program synthesis
TLDR
We present a generator-based synthesis approach to contend with the breadth of real-world APIs, and the often-complex constraints over function arguments. Expand
  • 7
  • PDF
Multi-Task Hierarchical Imitation Learning for Home Automation
TLDR
We present HIL-MT, a framework for Multi-Task Hierarchical Imitation Learning, involving a human teacher, a networked Toyota HSR robot, and a cloud-based server that stores demonstrations and trains models. Expand
  • 6
  • PDF
InferLine: ML Prediction Pipeline Provisioning and Management for Tight Latency Objectives
TLDR
In this paper we introduce InferLine, a system which provisions and executes ML prediction pipelines subject to end-to-end latency constraints by proactively optimizing and reactively controlling per-model configurations in a fine-grained fashion. Expand