Traffic-Aware VM Placement via Balanced Minimum K-Cut
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual machine (VM) placement schemes in data centers do not effectively reduce network traffic, as they oversimplify the problem by assuming static VM placement and equal server capacities, leading to suboptimal results due to dynamic job arrivals and server heterogeneity.
Innovation Solution
A traffic-aware VM placement method that models the problem as a balanced minimum k-cut problem, incorporating incremental VM job arrivals and server heterogeneity by using pseudo and dummy VM nodes, and pseudo edges to ensure optimal distribution across heterogeneous servers, minimizing total network traffic.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing VM placement schemes are used to consolidate VMs and balance load, then resource multiplexing is improved, but network traffic is not reduced
Solution Approach 1:
The patent changes the optimization parameter from generic load balancing to traffic-aware placement by incorporating network traffic metrics into the placement decision process. The system monitors traffic patterns between VMs and uses this information to place VMs strategically, thereby reducing network traffic while maintaining resource consolidation benefits
Solution Approach 2:
The patent implements a feedback mechanism that continuously monitors network traffic patterns and uses this information to dynamically adjust VM placement decisions. The system observes actual traffic flows and refines placement strategies accordingly, creating a closed-loop control system that reduces network traffic over time
2Loss of energy
If BMKP solution is used for traffic-aware VM placement, then bandwidth utilization is minimized, but server heterogeneity and dynamic job arrivals are not considered
Solution Approach 1:
The patent transforms the static BMKP approach into a dynamic system that adapts to changing conditions. It incorporates mechanisms to handle dynamic job arrivals by continuously evaluating new placement requests against current traffic patterns and server states, and it adapts to server heterogeneity by considering specific server characteristics in placement decisions
Solution Approach 2:
The patent applies local quality by treating different servers with different characteristics rather than uniformly. It considers specific server properties (capacity, location, current load) when making placement decisions, and optimizes traffic locally by placing VMs that communicate frequently on the same or nearby servers
3Loss of energy
If advanced network topologies are deployed to expand network capacity, then traffic pressure is reduced, but capital expenditures increase
Solution Approach 1:
The patent enables the data center to self-optimize its network usage through automated VM placement decisions. The system analyzes traffic patterns and makes intelligent placement choices that reduce network traffic pressure without requiring external infrastructure changes, allowing the system to serve its own optimization needs
Solution Approach 2:
The patent changes the approach from expanding physical network capacity to optimizing virtual resource placement parameters. Instead of adding more switches and links, it modifies how VMs are placed and managed to reduce traffic pressure on the existing network infrastructure
Data Source
AI summary
A method is implemented by a computing device to provide traffic-aware virtual machine (VM) placement onto physical servers of a data center where the placement takes incremental VM job arrival and physical server heterogeneity into consideration. The method forms a graph including a new VM node, an existing VM node, and an edge between the nodes, where the edge is assigned a weight that represents a traffic demand. The method marks the existing VM node as belonging to one of the physical servers, adds dummy VM nodes to the graph, adds pseudo VM nodes to the graph, connects nodes belonging to a same physical server using an infinite weight pseudo edge, runs a balanced minimum k-cut problem algorithm on the graph to thereby divide the graph into sub-graphs, and maps the new VM to one of the physical servers based on the division of sub-graphs.