International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 11 | Nov 2024
p-ISSN: 2395-0072
www.irjet.net
Streamlining Workload Management in AI-Driven Cloud Architectures: A Comparative Algorithmic Approach Kiran Kumar Patibandla1, Rajesh Daruvuri2, Pravallika Mannem3 1 Visvesvaraya Technological University (VTU), India 2 Google Inc, USA
3ProBPM, Inc, USA
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The use of artificial intelligence (AI) in cloud architectures has significantly increased processing efficiency and scale.
However, with the development of complex algorithms and big data as well as surprisingly entered into our machine learning world; workload management becomes a significant issue in AI cloud computing. Existing workload management solutions are rule-based heuristics that may result in underutilization of resources and poor performance. For that, we present an algorithmic comparative approach to easing the burden of workload management for AI-driven cloud architectures. This is in contrast to executing a batch of tasks with different algorithms and comparing performance, cost, etc. We use ML methods to determine the best algorithm for our workload, and then deploy this in a self-contained binary that can switch between algorithms at runtime on an available resource. We validated our scheme with simulations, which demonstrates the capability of superior resource use and diminished completion time in comparison to rule-based schemes. When needed, flexibility and scalability allow you easier control over workloads that are subject to change or allocation. By simplifying AI-driven cloud workload management, the elasticity of their overall approach greatly enhances efficiency and scalability for those organizations looking to run even larger and take advantage of more complex workloads faster Tweet this Share on Facebook. Key Words: Cloud Architectures, Scalability, Large Datasets, Better Management, Cost-Effectiveness
1.INTRODUCTION Before we start, let's see what two new terms which are introduced in this work considered for deviving insights on how to design the AI-powered cloud Architecture such that workload management can be easily streamlined via [1] What is Workload Management? Monitoring and Management of Work Streams OutLook related to Cloud Architecture: Cloud architecture is the design of an application built on top of a cloud provider setup.cloud computing support [2]. Nowadays, the employment of AI cloud in computer systems not only makes workload management turned out to be even more complex due to the significant setup and functional resources needed by AI algorithms (Lopez et al., 2018) [3]…) As such, it was suggested that containing these demands may require more than the conventional devices of workload management [4]. For the rapidly changing AIdriven cloud architectures, this means requiring a straightforward workload management approach. It includes the use of AI technologies to automate the management of cloud workloads (5). AI uses historical data and usage patterns to distribute the workload among VMs in the most effective order similar to resourse allocation [6]. This reduces human intervention and offers helpful resources across the functions. The workload in AI-driven cloud architectures and managing them is critical for both effective resource utilization as well more importantly ease of meeting the increased demands from modern enterprises. Nevertheless, the efforts to streamline workload management across AI-driven clouds require [7] addressing specific key challenges. If we refer back to the AI-driven workloads in addition to just general-purpose, another challenge is bookkeeping QoS requirements. The problem with AI models is that they chew up a lot of infrastructure resources in the form of data and computational power. Traditionally, workloads like these are difficult to manage and optimize in a normal cloud environment [9] that needs special hardware and software configuration. Additionally, there is the question of proper standardization for different AI workloads. Resource Requirements: A generic workload management system that caters to all the AI workloads is pretty hard since each AI algorithm and model has its very own resource requirements [10]. This makes it very hard to predict what resource usage an app will need, so you either over-provision or oversubscribe the resources. The main contribution of the research has the following:
Novel AI-based Workload Management: The research presents novel techniques that allow for efficient workload management of the new characteristics introduced by bottlenecked cloud-native AI workloads using a hybrid approach containing both traditional static resource allocation and dynamic, AI-driven methodologies. Below let's describe one of these creative ways used by current container orchestrators to do mean resource distribution across workloads which increases the performance and allows cost optimization.
© 2024, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 113