Newsletter #2 (2/2024)

Partner News

Augmenta: Smart Agriculture Testbed Enters its Pivotal Phase!

Exciting developments unfolded during the 2nd week of February 2024 as our Smart Agriculture testbed entered its pivotal stage. In the initial phase, a significant milestone was achieved through the successful prototyping of our drone and the establishment of a robust communication link with the tractor in a simulated environment.

As we move into the second phase, our attention turns to real-world scenarios, where we deploy the two-node system in the field for extensive testing. The primary emphasis here lies on the establishment of real-time communication between two nodes and data collection, a crucial element for training the MLSysOps models. This pivotal stage not only represents a substantial leap for our testbed but also lays the foundation for the upcoming third and final phase. The culmination of these efforts positions us to fully engage MLSysOps for real-time application optimization in the project’s concluding year.

Augmenta Team in action
Chocolate Cloud uses ML to optimize Object Storage

Cloud storage typically requires users to predict future demand when choosing the amount of redundancy and data location during the setup process of their accounts or storage buckets. This often results in unnecessarily high costs due to over-provisioning redundancy or poor performance when the demand shifts significantly from one region to another.

In the MLSysOps project Chocolate Cloud is extending its S3-compatible Object Storage service with an AI-based optimizer, which reacts or predicts changes in access patterns and moves data close to the current demand, as well as adjusts the level of redundancy to what’s actually needed right now.

This novel approach will take the burden of predicting traffic patterns and manually changing configuration off the shoulder of our valued customers, who can focus on their business challenges instead of spending time on storage optimization tasks.

Contact us to join our SkyFlok Object Storage beta test if you want to try full control of data location and soon, the AI-enabled self-driving storage configuration option!

Chocolate Cloud Object Storage optimization


Coming up: NVIDIA presentation at OFC 2024
Dr. Paraskevas Bakopoulos of NVIDIA will be delivering an invited paper presentation at OFC, the premier conference on Photonic Communications, which will take place in San Diego, USA on 24th-28th of March 2024. The paper is titled “Photonic switched networking for data centers and advanced computing systems” and, among other topics, it also articulates the challenges we aim to tackle in the MLSysOps project, in order to transparently leverage AI to optimize job scheduler performance on clusters with reconfigurable topology capabilities.
NVIDIA at OFC 2023
Drs. Paraskevas Bakopoulos and Giannis Patronas presenting a poster at OFC 2023
NUBIS was at FOSDEM 2024

FOSDEM 2024 was a blast! Our team was there presenting our work around unikernels, and their interaction with the cloud-native world.

The first presentation took place in the Containers devroom, where we described the challenges of securing containers within VM or microVM sandboxes and the complexities of resource optimization on physical nodes. We presented Urunc, a CRI-compatible container runtime spawning unikernels packaged in OCI images. The talk highlighted Urunc‘s internals, covering hypervisor support, network and storage handling, and integration with high-level orchestration frameworks like Kubernetes, while also addressing network setup implications when combining unikernels and generic containers in a k8s context.

In the Microkernel devroom, we continued our exploration of unikernels with a focus on modularity. The team presented Bunny, a novel build system aimed at simplifying the Unikernel building process. Bunny adopts a layered approach, making use of the modular aspect of Unikernels. Each component represents a distinct layer, eliminating the need for users to manually build dependencies. This approach allows for the creation of minimal and specialized Unikernel images tailored for individual applications.

These presentations highlight our commitment to simplify and optimize application delivery in diverse, heterogeneous infrastructure, providing valuable insights into security and system optimization in the evolving landscape of container technology.

Charalampos Mainas presenting a new approach on libOS building at FOSDEM 2024.
MLSysOps was at HiPEAC 2024
MLSysOps at HiPEAC 2024

ENHANCE Workshop at HiPEAC 2024

Spyros Lalis (UTH) co-chaired the ENHANCE workshop (Enabling Technologies and Dependability in Cyber-Physical Systems) in the HiPEAC conference. The workshop, which includes five presentations and a discussion panel, aims to bring together researchers working on technologies that can enable the development of advanced and dependable CPS. Notably, ENHANCE is part of a 3-workshop series in HiPEAC under the wider topic of CPS.

Presentation in ENHANCE Workshop at HiPEAC 2024

The presentation delivered by Theodoros Aslanidis (UTH) focused on the application of reinforcement learning techniques in the context of the cloud-edge continuum, specifically in the domain of autonomic and intelligent system management and configuration. The talk was part of the MLSysOps project and emphasized the need for a hybrid model to overcome the limitations of both cloud computing and edge computing. The presentation explained how reinforcement learning enables adaptive decision-making at different levels of the continuum, from fine-grained resource control to strategic deployment decisions, using a hierarchical agent framework. Additionally, the talk discussed the components of a reinforcement learning system, the challenges in implementing it, and potential algorithms suitable for this domain. Overall, the presentation highlighted the role of machine learning in improving system adaptability, efficiency, and responsiveness within the complex cloud-edge environment.

FORECAST Workshop at HiPEAC 2024

Aya Moheddine (INRIA) presented possible vulnerabilities associated with the communication protocol adopted in the IoT-Edge part of the MLSysOps architecture in the context of the HiPEAC 2024 FORECAST Workshop with a speech on “A Possible Scenario to Study the Security of 5G Communication in IoT Edge Cloud Continuum” on Jan 18, 2024. These vulnerabilities are considered to study the behavior of the IoT-Edge system in normal and abnormal cases, making it possible for the agent to detect attacks or anomalies.

INRIA at HiPEAC 2024

Presentation in STEADINESS Workshop at HiPEAC 2024

Nikolaos Bellas (UTH) gave a presentation on Reconfigurable Architectures for Approximate Dense SLAM computing in robotics. In the context of autonomous robotic platforms facing size, weight, and energy constraints, SLAM algorithms are commonly implemented in resource- and power-constrained embedded systems in a Cloud-Edge ecosystem. We explore the deployment of reconfigurable computing platforms in the form of FPGA-based hardware accelerators for SLAM processing. Moreover, we use techniques such as approximate computing, which is a computing paradigm that intentionally introduces controlled inaccuracies into computation to achieve significant performance or energy efficiency gains. Our demonstration reveals that a blend of precise and approximate optimizations, coupled with the fine-tuning of algorithmic parameters, results in a speedup of up to 15.7 times compared to the precise FPGA implementation without violating the stringent constraints on the number of untracked frames.

Presenting the Smart Cities Use Case
During 2023, Ubiwhere presented MLSysOps and the Smart Cities Use Case in several industrial fairs, events and congresses in Europe, including the Smart City Expo World Congress and the IoT Solutions World Congress 2023 in Barcelona and the Portugal Smart Cities Summit.
Smart Cities Summit
Ubiwhere presenting MLSysOps and the Smart Cities use case at the Portugal Smart Cities Summit.
2nd MLSysOps Plenary Meeting
The 2nd MLSysOps Plenary meeting took place at the University of Calabria, Rende, Italy on September 28-29, 2023. During the 2-day full immersion meeting, the partners reported the current status of work and engaged in constructive brainstorming about the ongoing and upcoming activities.
2nd MLSysOps Plenary Meeting
1st MLSysOps International Workshop

The 1st International Workshop on Machine Learning for Autonomic System Operations in the Device-Edge-Cloud Continuum ( MLSysOps 2023) took place on September 25, 2023 in Rende, Italy, in conjunction with the International Conference on Embedded Wireless Systems and Networks (EWSN 2023). The goal of this workshop series is to bring together a community of researchers and practitioners who study problems at the intersection of AI/ML, autonomic and cognitive computing, Device-Edge-Cloud continuum, distributed system operation, and resilient application deployment.

This first edition featured five accepted papers with oral presentation and, notably, a very interesting invited talk by Prof. Prof. Eui-Nam Huh, Kyung Hee University (South Korea), focused on “Cloud Continuum for enabling large scale AI services”.

MLSysOps 2023
Marco Loaiza presenting the paper "Agents in the computing continuum: the MLSysOps perspective" at MLSysOps 2023.



Consortium map
Funded by the EU

The “MLSysOps” Project is funded by the European Community’s Horizon Europe Programme under Grant Agreement #101092912.

Copyright © 2024, MLSysOps EU Project, All rights reserved.

Skip to content