Search Results

Documents authored by Soudris, Dimitrios


Document
Adjacent LSTM-Based Page Scheduling for Hybrid DRAM/NVM Memory Systems

Authors: Manolis Katsaragakis, Konstantinos Stavrakakis, Dimosthenis Masouros, Lazaros Papadopoulos, and Dimitrios Soudris

Published in: OASIcs, Volume 107, 14th Workshop on Parallel Programming and Run-Time Management Techniques for Many-Core Architectures and 12th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms (PARMA-DITAM 2023)


Abstract
Recent advances in memory technologies have led to the rapid growth of hybrid systems that combine traditional DRAM and Non Volatile Memory (NVM) technologies, as the latter provide lower cost per byte, low leakage power and larger capacities than DRAM, while they can guarantee comparable access latency. Such kind of heterogeneous memory systems impose new challenges in terms of page placement and migration among the alternative technologies of the heterogeneous memory system. In this paper, we present a novel approach for efficient page placement on heterogeneous DRAM/NVM systems. We design an adjacent LSTM-based approach for page placement, which strongly relies on page accesses prediction, while sharing knowledge among pages with behavioral similarity. The proposed approach leads up to 65.5% optimized performance compared to existing approaches, while achieving near-optimal results and saving 20.2% energy consumption on average. Moreover, we propose a new page replacement policy, namely clustered-LRU, achieving up to 8.1% optimized performance, compared to the default Least Recently Used (LRU) policy.

Cite as

Manolis Katsaragakis, Konstantinos Stavrakakis, Dimosthenis Masouros, Lazaros Papadopoulos, and Dimitrios Soudris. Adjacent LSTM-Based Page Scheduling for Hybrid DRAM/NVM Memory Systems. In 14th Workshop on Parallel Programming and Run-Time Management Techniques for Many-Core Architectures and 12th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms (PARMA-DITAM 2023). Open Access Series in Informatics (OASIcs), Volume 107, pp. 7:1-7:12, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2023)


Copy BibTex To Clipboard

@InProceedings{katsaragakis_et_al:OASIcs.PARMA-DITAM.2023.7,
  author =	{Katsaragakis, Manolis and Stavrakakis, Konstantinos and Masouros, Dimosthenis and Papadopoulos, Lazaros and Soudris, Dimitrios},
  title =	{{Adjacent LSTM-Based Page Scheduling for Hybrid DRAM/NVM Memory Systems}},
  booktitle =	{14th Workshop on Parallel Programming and Run-Time Management Techniques for Many-Core Architectures and 12th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms (PARMA-DITAM 2023)},
  pages =	{7:1--7:12},
  series =	{Open Access Series in Informatics (OASIcs)},
  ISBN =	{978-3-95977-269-3},
  ISSN =	{2190-6807},
  year =	{2023},
  volume =	{107},
  editor =	{Bispo, Jo\~{a}o and Charles, Henri-Pierre and Cherubin, Stefano and Massari, Giuseppe},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.PARMA-DITAM.2023.7},
  URN =		{urn:nbn:de:0030-drops-177278},
  doi =		{10.4230/OASIcs.PARMA-DITAM.2023.7},
  annote =	{Keywords: Page Placement, Long Short-Term Memory, LSTM, Prediction, NVM, DRAM}
}
Document
Resource Aware GPU Scheduling in Kubernetes Infrastructure

Authors: Aggelos Ferikoglou, Dimosthenis Masouros, Achilleas Tzenetopoulos, Sotirios Xydis, and Dimitrios Soudris

Published in: OASIcs, Volume 88, 12th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 10th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms (PARMA-DITAM 2021)


Abstract
Nowadays, there is an ever-increasing number of artificial intelligence inference workloads pushed and executed on the cloud. To effectively serve and manage the computational demands, data center operators have provisioned their infrastructures with accelerators. Specifically for GPUs, support for efficient management lacks, as state-of-the-art schedulers and orchestrators, threat GPUs only as typical compute resources ignoring their unique characteristics and application properties. This phenomenon combined with the GPU over-provisioning problem leads to severe resource under-utilization. Even though prior work has addressed this problem by colocating applications into a single accelerator device, its resource agnostic nature does not manage to face the resource under-utilization and quality of service violations especially for latency critical applications. In this paper, we design a resource aware GPU scheduling framework, able to efficiently colocate applications on the same GPU accelerator card. We integrate our solution with Kubernetes, one of the most widely used cloud orchestration frameworks. We show that our scheduler can achieve 58.8% lower end-to-end job execution time 99%-ile, while delivering 52.5% higher GPU memory usage, 105.9% higher GPU utilization percentage on average and 44.4% lower energy consumption on average, compared to the state-of-the-art schedulers, for a variety of ML representative workloads.

Cite as

Aggelos Ferikoglou, Dimosthenis Masouros, Achilleas Tzenetopoulos, Sotirios Xydis, and Dimitrios Soudris. Resource Aware GPU Scheduling in Kubernetes Infrastructure. In 12th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 10th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms (PARMA-DITAM 2021). Open Access Series in Informatics (OASIcs), Volume 88, pp. 4:1-4:12, Schloss Dagstuhl – Leibniz-Zentrum für Informatik (2021)


Copy BibTex To Clipboard

@InProceedings{ferikoglou_et_al:OASIcs.PARMA-DITAM.2021.4,
  author =	{Ferikoglou, Aggelos and Masouros, Dimosthenis and Tzenetopoulos, Achilleas and Xydis, Sotirios and Soudris, Dimitrios},
  title =	{{Resource Aware GPU Scheduling in Kubernetes Infrastructure}},
  booktitle =	{12th Workshop on Parallel Programming and Run-Time Management Techniques for Many-core Architectures and 10th Workshop on Design Tools and Architectures for Multicore Embedded Computing Platforms (PARMA-DITAM 2021)},
  pages =	{4:1--4:12},
  series =	{Open Access Series in Informatics (OASIcs)},
  ISBN =	{978-3-95977-181-8},
  ISSN =	{2190-6807},
  year =	{2021},
  volume =	{88},
  editor =	{Bispo, Jo\~{a}o and Cherubin, Stefano and Flich, Jos\'{e}},
  publisher =	{Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{https://drops.dagstuhl.de/entities/document/10.4230/OASIcs.PARMA-DITAM.2021.4},
  URN =		{urn:nbn:de:0030-drops-136403},
  doi =		{10.4230/OASIcs.PARMA-DITAM.2021.4},
  annote =	{Keywords: cloud computing, GPU scheduling, kubernetes, heterogeneity}
}
Questions / Remarks / Feedback
X

Feedback for Dagstuhl Publishing


Thanks for your feedback!

Feedback submitted

Could not send message

Please try again later or send an E-mail