Flexible buffer management with thresholds and blocking for congestion control in multi-server computer systems

This paper is aimed at designing a congestion control system (decision maker) that scales gracefully with computer system capacity, providing high utilization, and fairness among users. The use of adequate buffer management scheme with thresholds and blocking are well-known techniques for computer systems traffic congestion control. This motivates the study of multi-server computer systems with two distinct priority classes (high and low priority traffics), partial buffer sharing scheme with thresholds and with blocking. The external traffic is modeled using the Poisson processes. The service times have been modeled using the exponential distribution. Adaptive buffer allocation algorithm (scheme) is designed to allow the input traffic to be portioned into different priority classes and based on the input traffic behavior it controls the thresholds dynamically. This algorithm allows input low priority jobs to enter into buffer if its occupancy level is less than the threshold value. Additionally, for low priority jobs a forward threshold and a reverse threshold are defined (hysteresis). Using an open Markovian queuing schema with blocking, and thresholds, a closed form cost-effective analytical solution for this model of computer system is obtained. The model of multi-server system is very accurate. It is derived directly from a state graph and a set of steady-state equations, followed by calculations of main measures of effectiveness. Based on numerical experiments and collected results we conclude that the proposed model with blocking, and thresholds can provide accurate performance estimates of multi-server systems.