Bio-Inspired Load-Balancing Framework for Loosely Coupled Heterogeneous Server Systems

Balancing load among servers is an important research challenge for a large-scale loosely coupled heterogeneous server system (LCHSS), to improve both the total throughput of the system and the quality of service experienced by clients. In practical terms, a load-balancing method for an LCHSS have to drive servers to underloaded states without unnecessary load migrations among servers. To tackle this problem, we propose a load-balancing framework inspired by biological systems that have developed adaptive, robust, and flexible behaviors through the local interactions of individual nodes with limited information. In our framework, we integrate two different biological models systematically and develop new mathematical formulas. With the developed formulas, we introduce two key rules to balance the load levels among servers in a fully distributed manner through the employment of the inter-cell signaling model and the Kuramoto synchronization model. Using a mathematical stability analysis, we provide a guide for the configuration of model parameters and prove that a system stabilizes at a steady state. Using task event logs measured at a Google cluster, we conducted a variety of case studies for the evaluation of the framework. The evaluation results verify that our framework balances the load levels among servers in spite of the variability in a system.