Jockey: guaranteed job latency in data parallel clusters
By Andrew D. Ferguson, Peter Bodík, Srikanth Kandula, Eric Boutin, Rodrigo Fonseca. European Conference on Computer Systems, 2012
347 citations recorded by the source corpus.
Subfields: Computer Networks and Communications, Information Systems
View Paper: https://www.semanticscholar.org/paper/0d868efa67bf06b1f784d60769c082fd9a58893e
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Computer Networks and Communications
- Mohammad Alizadeh
- Hongzi Mao
- Ion Stoica
- Phillip B. Gibbons
- M. Franklin
- Shivaram Venkataraman
- Chuanxiong Guo
- Srikanth Kandula
In Information Systems
- Ion Stoica
- Shivaram Venkataraman
- Chuanxiong Guo
- Mohammad Alizadeh
- B. Recht
- Srikanth Kandula
- Hongzi Mao
- M. Franklin
Citations in the corpus
- Ernest: Efficient Performance Prediction for Large-Scale Advanced Analytics (2016, background): “One approach to address this challenge is to predict the performance of a job based on monitoring the job’s previous runs [39, 44].”
- CherryPick: Adaptively Unearthing the Best Cloud Configurations for Big Data Analytics (2017, background): “Many previous works were built on top of the similarity in recurring jobs [10, 17].”
- Omega: flexible, scalable schedulers for large compute clusters (2013, background): “More elaborate approaches and objective functions, such as used in deadline-based schedulering [10], are certainly possible, but not the focus of this case study.”
- Addressing the straggler problem for iterative convergent parallel ML (2016, background): “For instance, it is not applicable to programs written in garbage-collected languages, does not handle algorithms that inherently cause stragglers during some iterations, and does not work for today’s multi-tenant computing infrastructures [17, 22, 30].”
- Optimus: an efficient dynamic resource scheduler for deep learning clusters (2018, background): “Jockey [32] and Morpheus [44] use historical traces of periodic jobs and dynamically adjust resource allocations to meet deadlines, while Optimus does not depend on the previous run of the same job since production training data often change (e.”
- Resource Management with Deep Reinforcement Learning (2016, background): “Further, the resource profile of a job may not be known in advance (e.g., for non-recurring jobs [4]), and the scheduler might have get an accurate view only as the job runs [14].”