SearcharxivSearch

arXiv subjects

Marjan Gusev

Publications and source records attributed to Marjan Gusev.

5 recordsLinked to original sources

Scalability Evaluation of HPC Multi-GPU Training for ECG-based LLMs

Training large language models requires extensive processing, made possible by many high-performance computing resources. This study compares multi-node and multi-GPU environments for training large language models of electrocardiograms. It provides a detailed mapping of current frameworks for distributed deep learning in multinode and multi-GPU settings, including Horovod from Uber, DeepSpeed from Microsoft, and the built-in distributed capabilities of PyTorch and TensorFlow. We compare various multi-GPU setups for different dataset configurations, utilizing multiple HPC nodes independently and focusing on scalability, speedup, efficiency, and overhead. The analysis leverages HPC infrastructure with SLURM, Apptainer (Singularity) containers, CUDA, PyTorch, and shell scripts to support training workflows and automation. We achieved a sub-linear speedup when scaling the number of GPUs, with values of 1.6x for two and 1.9x for four.

cs.DC

Cloud Computing Security in Business Information Systems

Cloud computing providers' and customers' services are not only exposed to existing security risks, but, due to multi-tenancy, outsourcing the application and data, and virtualization, they are exposed to the emergent, as well. Therefore, both the cloud providers and customers must establish information security system and trustworthiness each other, as well as end users. In this paper we analyze main international and industrial standards targeting information security and their conformity with cloud computing security challenges. We evaluate that almost all main cloud service providers (CSPs) are ISO 27001:2005 certified, at minimum. As a result, we propose an extension to the ISO 27001:2005 standard with new control objective about virtualization, to retain generic, regardless of company's type, size and nature, that is, to be applicable for cloud systems, as well, where virtualization is its baseline. We also define a quantitative metric and evaluate the importance factor of ISO 27001:2005 control objectives if customer services are hosted on-premise or in cloud. The conclusion is that obtaining the ISO 27001:2005 certificate (or if already obtained) will further improve CSP and CC information security systems, and introduce mutual trust in cloud services but will not cover all relevant issues. In this paper we also continue our efforts in business continuity detriments cloud computing produces, and propose some solutions that mitigate the risks.

cs.CR

Architecture Of A Identity Based Firewall System

Classic firewall systems are built to filter traffic based on IP addresses, source and destination ports and protocol types. The modern networks have grown to a level where the possibility for users' mobility is a must. In such networks, modern firewalls may introduce such complexity where administration can become very frustrating since it needs the intervention of a firewall administrator. The solution for this problem is an identity based firewall system. In this paper we will present a new design of a firewall system that uses the user's identity to filter the traffic. In the design phase we will define key points which have to be satisfied as a crucial milestone for the functioning of the whole Identity based firewall system.

cs.CR