Abstract: Server consolidation based on virtualization technology simplifies system administration and improves energy efficiency by improving resource utilizations and reducing the physical machine (PM) number in contemporary service-oriented data centers. The elasticity of Internet applications changes the consolidation technologies from addressing virtual machines (VMs) to PMs mapping schemes into providing the application-to-VM-to-PM mapping. In this paper, we study the consolidation of multiple Internet applications, minimizing the number of PMs with required performance. We model the consolidation as an integer linear programming problem and present a heuristic algorithm, 3MAX, to solve it in polynomial time. Experimental results show that our heuristic algorithm consumes less than 4.3% more resources than the optimal amounts with few overheads, outperforming existing consolidation technologies.
Introduction
In modern data centers, server consolidation is the primary strategy for energy efficiency. By packing as many Virtual Machines (VMs) onto as few Physical Machines (PMs) as possible, operators can power down idle hardware. However, traditional consolidation techniques often treat the number of VMs as a fixed constant. This approach fails to account for "elastic" Internet applications—such as e-commerce or web services—where the number of instances must scale dynamically based on user demand. Furthermore, most models assume hardware homogeneity, ignoring the reality that a VM's performance can vary significantly depending on the underlying physical server hardware.
Methods
The authors address these challenges by formulating consolidation as a three-tiered mapping problem: Application $\rightarrow$ VM $\rightarrow$ PM. This model allows the system to handle horizontal scaling (adding/removing VMs) and vertical scaling (resizing VMs) simultaneously with hardware placement.
Since solving this as an Integer Linear Programming (ILP) problem is NP-hard and impractical for large-scale data centers, the researchers proposed the 3MAX heuristic algorithm. The "3MAX" logic aims to maximize the ratio of performance provided to resources consumed. It follows a greedy approach:
- PM Selection: It selects a target PM and calculates the efficiency of various VM types on that specific hardware.
- Application Allocation: It selects the application with the highest performance-to-resource ratio and allocates a corresponding VM to it on that PM.
- Iteration: This process repeats until the PM is saturated or all application performance requirements are met.
Datasets and Evaluation
The researchers evaluated 3MAX using a diverse set of industry-standard benchmarks to simulate realistic cloud workloads:
- TPC-W: A transactional web e-commerce benchmark.
- YCSB (Yahoo! Cloud Serving Benchmark): Simulates cloud-based data services.
- Apache Benchmark (ab): Simulates web server requests for varying file sizes.
- SysBench: Tests CPU-intensive prime number calculations.
- Workload Traces: Real-world traffic data from the 1998 World Cup website was used to generate fluctuating request patterns over a 24-hour period.
Key Results
The experimental results highlight the efficiency and scalability of the 3MAX approach:
- Near-Optimal Efficiency: The 3MAX heuristic consumes less than 4.3% more resources than the theoretically optimal ILP solution.
- Superior Consolidation: Compared to common packing algorithms like First Fit Decreasing (FFD) and Least Loaded (LL), 3MAX reduced the total number of required physical machines by approximately 1.06%.
- High Scalability: In simulations involving 3,000 PMs and 500 applications, the algorithm reached consolidation decisions in just 0.22 seconds, proving its viability for dynamic, real-time cloud environments.
- Performance Guarantees: The "Relative Difference" (RDi) between required and provided performance remained consistently near zero, ensuring applications received necessary resources without significant waste or performance degradation.
Conclusion
The 3MAX heuristic provides a practical and scalable solution for consolidating elastic Internet applications in heterogeneous data centers. By integrating application-level performance requirements directly into the VM-to-PM mapping logic, it achieves near-optimal energy efficiency while maintaining strict performance guarantees. Future work may extend this model to include disk I/O constraints and the specific overhead costs associated with live VM migrations.
Summary updated