🧑‍🔬 Amar Phanishayee

🏛️ Affiliations

🏢MicrosoftRedmond · United States2013–2024 · 33 papers🎓Carnegie Mellon UniversityPittsburgh · United States2007–2018 · 16 papers🏢Microsoft ResearchCambridge · United Kingdom2020–2024 · 3 papers🏢MetaMenlo Park · United States2025 · 2 papers🎓Cornell UniversityIthaca · United States2006–2007 · 2 papers

From the bylines of their papers (OpenAlex, 2026-10-14), ORCID and Wikidata; institutions on papers are matched by machine. 2 more institutions on single bylines are not listed.

🧭 Fields, topics and concepts

Of its works in our library, with the number of works on each.

📅 Works by year

2020s

2010s

📄 Papers

The most cited first (OpenAlex and our library).

1. Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training Workloads (2019 · 179 citations)
   The paper examines how deep learning training behaves on a shared, production GPU cluster at Microsoft. Such clusters differ from classic big-data clusters in two ways: a GPU cannot easily be split among users, and multi-GPU jobs must have all their GPUs allocated at once (gang scheduling). The authors describe…
2. PipeDream: Fast and Efficient Pipeline Parallel DNN Training (2018 · 162 citations)
   Data-parallel DNN training gives every GPU a full model copy and synchronizes weights after each step. As models grow and GPUs get faster, network traffic increasingly dominates runtime. For VGG16, it can take up to 85% of training time. Classic model parallelism avoids exchanging all weights but leaves most workers…
3. Memory-Efficient Pipeline-Parallel DNN Training (2020 · 153 citations)
   Very large language models often cannot fit their weights and activations on one accelerator, so training must be spread across devices.
4. Heterogeneity-Aware Cluster Scheduling Policies for Deep Learning Workloads (2020 · 141 citations)
   Deep learning training clusters now mix accelerators of different generations, and models benefit unequally from newer hardware.
5. Themis: Fair and Efficient GPU Cluster Scheduling (2019 · 87 citations)
   Organizations increasingly pool GPUs into shared clusters for ML training, but fair-sharing schedulers built for big-data analytics, such as DRF, handle these workloads poorly.

All works

6 works of theirs in our library, the most cited first; 5 with a page here, the others link their original until we summarise them.

1. Analysis of Large-Scale Multi-Tenant GPU Clusters for DNN Training WorkloadsarXiv:1901.05758Myeongjae Jeon, Shivaram Venkataraman, Amar Phanishayee, et al.2019 · 179 citations
2. PipeDream: Fast and Efficient Pipeline Parallel DNN TrainingarXiv:1806.03377Aaron Harlap, Deepak Narayanan, Amar Phanishayee, et al.2018 · 162 citations
3. Memory-Efficient Pipeline-Parallel DNN TrainingInternational Conference on Machine Learning (ICML)Deepak Narayanan, Amar Phanishayee, Kaiyu Shi, et al.2020 · 153 citations
4. Heterogeneity-Aware Cluster Scheduling Policies for Deep Learning WorkloadsarXiv:2008.09213Deepak Narayanan, Keshav Santhanam, Fiodar Kazhamiaka, et al.2020 · 141 citations
5. Themis: Fair and Efficient GPU Cluster SchedulingarXiv:1907.01484Kshiteej Mahajan, Arjun Balasubramanian, Arjun Singhvi, et al.2019 · 87 citations
6. Blink: Fast and Generic Collectives for Distributed ML ↗arXiv:1910.04940 · no summary yetGuanhua Wang et al.2019 · 58 citations

🤝 Co-authors here

9 co-authors with a page, the most papers together first.

• Deepak Narayanan ×3
• Matei Zaharia ×2
• Shivaram Venkataraman ×2
• Aditya Akella
• Fan Yang
• Keshav Santhanam
• Vivek Seshadri
• Wencong Xiao
• Xie Chen

Venues

Where their works here were published, the most works first, with the editions of those that have a page.

• International Conference on Machine Learning (ICML): ICML 2020

🔗 Cited by

Papers of our library cite their works 465 times (of the 149,439 papers whose bibliographies we read); 28 citing papers with a page here →.

Updated 2 November 2026

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/authors/amar-phanishayee/