| |
Click here for full text:
A Mathematical Optimization Approach for Resource Allocation in Large Scale Data Centers
Santos, Cipriano; Zhu, Xiaoyun; Crowder, Harlan
HPL-2002-64R1
Keyword(s): mathematical programming; Internet data center; resource allocation; scalability
Abstract: In this paper, we address the resource allocation problem (RAP) for large scale data centers using mathematical optimization techniques. Given a physical topology of resources in a large data center, and an application with certain architecture and requirements, we want to determine which resources in the physical topology should be assigned to the application architecture such that application requirements and bandwidth constraints in the network are satisfied, while communication delay between assigned servers is minimized. We have decomposed this complex combinatorial optimization problem into a series of tractable mathematical optimization models that can be easily solved using commercially available mathematical programming solvers. Preliminary tests of this approach demonstrated consistently that optimal or "good" solutions for problems from small to large scales can be found in seconds, which proves its better scalability compared to a Layered Partitioning and Pruning (LPP) algorithm in previous work. In addition, this Mathematical Programming approach can be extended to more general problems using the same solvers, that is, extensions of the original problem do not require the development of new algorithms. Notes: Presented at the Informs Annual Meeting, 17-20 November 2002, San Jose, CA
24 Pages
Back to Index
|
printable version
